5.1 Introduction

This chapter aims to contribute to further understanding of the acquisition of speech from a multilingual perspective by offering a state-of-the-art overview of findings and discussing some longstanding theoretical and methodological considerations in research on third language (L3) phonological acquisition. As the discipline grows, the methodologies employed advance, yet certain aspects continue to pose a challenge including, among others, divergent research designs, varied types of L3 learners, comprehensive measures of multilingual production and perception, and phonological proficiency assessment. From a theoretical perspective, the applicability of the established L3 morphosyntactic models to phonological data is being challenged, while alternative explanatory approaches are put forward. However, the question remains which model may offer sufficient explanatory power to account for multilingual speech learning under varied acquisition circumstances.

Faced with the dynamic development of third language acquisition (TLA), which came to be recognized as an independent field of inquiry, a number of scholars have pointed to the phonological aspect of L3 acquisition as an understudied domain that should attract more attention in scientific investigations (e.g., Reference Hammarberg, James and LeatherHammarberg, 1997; Reference Cabrelli Amaro, Cabrelli Amaro, Flynn and RothmanCabrelli Amaro, 2012; Reference Wrembel and Cabrelli AmaroWrembel & Cabrelli Amaro, 2018). In line with this call, over the past decade, we have witnessed an upsurge of interest in research on L3 phonetics and phonology at international conferences, in particular at the International Conference on Third Language Acquisition and Multilingualism, New Sounds, EuroSLA, or L3 Workshop. Several workshops devoted to L3 phonology have been organized over the years, including a satellite workshop of the International Congress of Phonetic Sciences (ICPhS), which was held in Freiburg in 2007, the first of its kind on phonological transfer in TLA. This was followed by a workshop on advances in the investigation of L3 phonological acquisition at the Societas Linguistica Europaea (SLE) in Poznań in 2014, a workshop dedicated to L3 English pronunciation at New Sounds in Aarhus in 2016, a workshop on modeling the acquisition of foreign language speech at SLE in Zürich in 2017, and a special session on theoretical and methodological challenges in L3 phonological acquisition at ICPhS 2019 in Melbourne.

Having gained recognition of its theoretical relevance and practical importance, L3 phonology has become a thriving area of research generating a growing body of literature. The first special issue published in the International Journal of Multilingualism in 2010 (Reference Wrembel, Gut and MehlhornWrembel et al., 2010) was aimed to outline new methods for modeling and investigating L3 phonological acquisition, and in particular, to address transfer in L3 phonology. This issue has been widely cited and has turned out to be one of the major reference sources for L3 phonetics and phonology. The second special issue of the journal was published in 2016 (Reference Cabrelli AmaroWrembel & Cabrelli Amaro, 2016) and it focused on advances in the exploration of L3 phonological acquisition, thus contributing some novel approaches to study design, data analysis, and theoretical frameworks. It also demonstrated a number of limitations related to the scope of investigations into L3 phonetics and phonology as well as the existing methodological shortcomings (for an introductory article, see Reference Cabrelli Amaro and WrembelCabrelli Amaro & Wrembel, 2016).

In a special volume on universal or diverse paths to English phonology, Reference Gut, Fuchs and WunderGut and colleagues (2015) drew some comparisons between the acquisition of phonology from an SLA versus TLA perspective. While the acquisition of second language (L2) speech enjoys a well-grounded tradition of investigations going back to the 1960s, research into L3 phonology is a young discipline that has its roots in the late twentieth and early twenty-first centuries. Juxtaposing these two acquisition perspectives, we cannot fail to notice divergent approaches to research paradigms. SLA traditionally relied on cross-sectional experimental studies focusing mostly on L2 data and L1-to-L2 transfer, whereas L3 phonology started off with longitudinal case studies of multilinguals, which developed into corpus-based, experimental research. TLA research strives to elicit data in all the languages from the participants’ repertoire and allows for dynamic multidirectional crosslinguistic influence. What still seems necessary though is to verify further the existing theoretical and methodological assumptions relevant for L2 speech acquisition in order to expand them into more complex contexts of learning additional foreign languages.

The outline of this chapter is as follows: first, an overview of key research areas in the field will be offered. Second, some pressing methodological considerations in research on L3 phonological acquisition will be discussed and potential solutions will be suggested. Third, different approaches to theoretical modeling of L3 phonology will be critically assessed. Then, the chapter will focus on new insights into the field by overviewing state-of-the-art findings from a recent longitudinal “Multi-Phon” project on crosslinguistic influence (CLI) in multilingual phonological acquisition and, finally, some avenues for further research in the area will be suggested.

5.2 Scope of Investigation in L3 Phonological Research

This section is intended to recapitulate earlier extensive overviews presented in studies by Reference Cabrelli Amaro, Cabrelli Amaro, Flynn and RothmanCabrelli Amaro (2012), Reference Wrembel, Cabrelli Amaro, Flynn and RothmanWrembel (2012, Reference Wrembel2015b), and Reference Cabrelli Amaro and WrembelCabrelli Amaro and Wrembel (2016). The scope of investigations in L3 phonological studies is fairly broad as it ranges from segmental features including voice onset time (VOT) (e.g., Reference Llama, Cardoso and CollinsLlama et al., 2010; Reference Wrembel, Lee and ZeeWrembel, 2011, Reference Wrembel2015b; Reference Wunder, De Angelis and DewaeleWunder, 2011; Reference SypiańskaSypiańska, 2013; Reference Llama and López-MorelosLlama & Lopez-Morelos, 2016; Reference Gabriel, Krause and DittmersGabriel et al., 2018; Reference Geiss, Gumbsheimer, Lloyd-Smith, Schmid and KupischGeiss et al., 2021); vowel quality and quantity (Reference MissagliaMissaglia, 2010; Reference SypiańskaSypiańska, 2013, Reference Sypiańska2016; Reference Kopečková, Marecka, Wrembel and GutKopečková et al., 2016); to suprasegmentals featuring speech rhythm and vowel reduction (Reference GutGut, 2010; Reference Gabriel, Thulke and KupischGabriel et al., 2014; Reference Cabrelli AmaroCabrelli Amaro, 2013, Reference Cabrelli Amaro and Wrembel2016; Reference Domeno Moreno, Kabak and FuchsDomeno Moreno & Kabak, in press); as well as global foreign accentedness (Reference WrembelWrembel, 2010, Reference Wrembel, Cabrelli Amaro, Flynn and Rothman2012; Reference Lloyd-Smith, Gyllstad and KupischLloyd-Smith, Gyllstad, & Kupisch, 2017; Reference Lloyd-SmithLloyd-Smith, 2021). More recent L3 phonological studies broaden the variety of features under investigation by exploring rhotics (e.g., Reference Balas, Kopečková, Wrembel, Calhoun, Escudero, Tabain and WarrenBalas et al., 2019), the contrast between labiodental fricative versus labio-velar approximant (/v/ vs. /w/) (Reference Kopečková, Dimroth and GutKopečková et al., 2019), final obstruent devoicing (Reference Wrembel, Gut, Kopečková and BalasWrembel et al., 2020), as well as prosody (Reference Liu, Gorba, Cebrian, Calhoun, Escudero, Tabain and WarrenLiu et al., 2019; Reference Ren, Mok, Calhoun, Escudero, Tabain and WarrenRen & Mok, 2019).

These production studies seem to have dominated the field, while a relatively smaller number of studies have dealt with crosslinguistic perception in L3 (e.g., Reference Kopečková, Gut, Fuchs and WunderKopečková, 2015; Reference OnishiOnishi, 2016; Reference Balas, Kopečková, Wrembel, Calhoun, Escudero, Tabain and WarrenBalas et al., 2019; Reference Wrembel, Marecka and KopečkováWrembel et al., 2019b). These recent investigations indicate that the phonological space of multilinguals seems to be reshaped relatively early in the course of learning the new L3, and that category boundaries can be expanded to accommodate L1, L2, and L3 categories of similar phonetic types, while new L3 categories for novel phonetic types may be formed. For instance, Reference Kopečková, Gut, Fuchs and WunderKopečková’s (2015) study demonstrated that young Polish–English bilinguals are less sensitive to the differences between Polish and English vowels than their multilingual peers, thus suggesting that previous L2 learning experience enhances auditory awareness, facilitating the learner’s subsequent speech learning. Similarly, Reference OnishiOnishi (2016) showed general facilitative effects of prior linguistic experience on the perception of L3 Japanese contrasts by L1 Korean and L2 English speakers. Reference Wrembel, Marecka and KopečkováWrembel and colleagues (2019b) reported that beginner L3 Polish learners perceptually assimilate L3 sibilants to both their L1 German and L2 English categories; they could also perceive subtle differences between highly similar vowel sounds across the three languages and seemed to develop separate L3 categories for them. Reference Balas, Kopečková, Wrembel, Calhoun, Escudero, Tabain and WarrenBalas and colleagues (2019) investigated longitudinally the perception of L2 and L3 rhotic sounds in two groups of young multilinguals. The findings pointed to an effect of the learners’ L1, the role of markedness and L2/L3 proficiency in the perception of rhotic sounds; however, no developmental changes were attested between the two testing times within one school year. In turn, Reference NelsonNelson (2020) examined young and adult L3 learners’ perception of the /v-w/ contrast, reporting more accurate and faster discrimination ability in the L3 than in the L2. It led her to hypothesize a “novelty effect” for the L3 learners at initial stages of acquisition which prevents them from assimilating novel sounds to pre-existing categories of L1 or L2.

As elucidated in previous overviews of the field, the key research area in L3 phonology focuses on the nature and sources of crosslinguistic influence. With regard to the CLI nature, there is an ongoing debate whether it occurs simultaneously or sequentially, with some studies pointing to the former (e.g., Reference BarkleyBarkley, 2010; simultaneous L1 English and L2 Spanish influence on L3 Brazilian Portuguese; Reference WrembelWrembel, 2015b), and others to the latter (e.g., Reference Hammarberg, Hammarberg, Hufeisen and FouserHammarberg & Hammarberg, 2005; sequential influence of L2 German and then L1 English on L3 Swedish; Reference WrembelWrembel, 2010). Multiple sources of CLI have been attested across different phonological phenomena; for instance, Reference WrembelWrembel (2010, Reference Wrembel, Cabrelli Amaro, Flynn and Rothman2012, Reference Wrembel2015b) demonstrated hybrid VOT productions in L3, while Reference Blank and ZimmerBlank and Zimmer (2009) reported hybrid vowels in L3. The phenomenon of phonological transfer appears to be determined also by the complexity of particular subsystems (for example, Reference BenrabahBenrabah’s [1991] investigation of Arabic/French bilinguals acquiring L3 English pointed to consonants being transferred from Arabic, whereas vowels from French). Several L3 phonological studies found evidence for L1-based or dominance-driven CLI (e.g., Reference Garcia Lecumberri, Gallardo Del Puerto, Garcia Mayo and Garcia LecumberriGarcia Lecumberri & Gallardo del Puerto, 2003; Reference Llama and López-MorelosLlama & Lopez-Morelos, 2016 with English-like VOT in L3 French attested in English–Spanish bilinguals). However, the extant literature prevailingly reports cases in which L3 productions are assimilated both to the L1 and L2, thus indicating cumulative influence from the native and non-native phonological systems (L1 and L2) onto the L3. For instance, Reference SypiańskaSypiańska (2016) reported combined CLI from L1 Polish and L2 Danish on L3 English vowels in what she claimed to be a global entity formed by component languages. Reference WrembelWrembel (2015b), based on her three series of studies involving accent ratings, VOT, and phonological awareness, concluded that CLI is gradual and feature-dependent rather than holistic, mostly L1-driven with complementary influence from L2. Further, it has been shown that the type and frequency of phonological CLI depends on a combination of factors, such as the proficiency in the target and source languages, recency of use, perceived language distance between source and target languages, and the degree of metalinguistic awareness. It appears that some of these factors are only influential on specific linguistic levels and only for certain types of learners, and change depending on the varying L3 proficiency level or the experimental task applied (see, e.g., Reference Hammarberg, Hammarberg, Hufeisen and FouserHammarberg & Hammarberg, 2005; Reference WrembelWrembel, 2010).

Another key research question in the area concerns a bilingual advantage for subsequent phonetic and phonological acquisition. Scholars posit various explanations for this potential facilitation, including linguistic arguments – more extensive articulatory and perceptual knowledge and phonological awareness (Reference GutGut, 2010) – as well as cognitively-based ones – cognitive flexibility, better inhibition mechanisms regulating CLI (e.g., Reference Bartolotti and MarianBartolotti & Marian, 2012) – and the learnability of features under analyses (e.g., Reference KopečkováKopečková, 2016). While multilingual advantage in the perception of novel contrasts was evidenced in some studies (e.g., Reference Antoniou, Liang, Ettlinger and WongAntoniou et al., 2015; Reference Kopečková, Gut, Fuchs and WunderKopečková, 2015; Reference OnishiOnishi, 2016; Reference Wrembel, Marecka and KopečkováWrembel et al., 2019b), others did not attest any differences between monolingual and bilingual acquisition of new contrasts (e.g., Reference Gabriel, Thulke and KupischGabriel et al., 2014; Reference Patihis, Oh and MogilnerPatihis et al., 2015). These conflicting outcomes could be ascribed to such variables as typological distance between languages in question, varying acquisition contexts, as well as difference in language dominance and proficiency (for a more detailed discussion, see Reference Cabrelli Amaro and WrembelCabrelli Amaro & Wrembel, 2016).

Recapitulating, the extant findings from L3 phonology research suggest that any of the previously or currently acquired languages can serve as a source for CLI in the perception and production of target segments and suprasegmentals, and that this phenomenon is multidirectional. We have also a growing understanding of the combination of factors conditioning the different types of phonological CLI in L3 learning, such as proficiency in the respective languages, (psycho)typology as well as the type of phonological task performed (for an overview, see Reference Wunder, De Angelis and DewaeleWunder, 2011).

5.3 Methodological Considerations and Challenges

One of the main aims of this contribution is to shed further light on theoretical considerations and methodological challenges for L3 phonological acquisition studies. Due to the inherent complexity of research into L3 acquisition, some of the long-tested approaches valid for investigating speech from the SLA perspective seem no longer applicable. We are faced with multifarious problems and methodological shortcomings that could be addressed by widening the scope and time scale of multilingual research designs.

The research focus has shifted from the initial interest in the outcome of L3 acquisition to more attention paid nowadays to the developmental process in order to gain a more holistic picture of crosslinguistic mapping in multilingual learners, and possible changes over time. This requires the application of different methodologies and a consequent change from purely cross-sectional design to mixed or longitudinal investigations which involve several testing times to trace the developmental trajectory of the process of phonological acquisition (e.g., Reference Wrembel, Gut, Krzysik, Lewandowska, Balas, Calhoun, Escudero, Tabain and WarrenWrembel et al., 2019a, Reference Wrembel, Gut, Kopečková and Balas2020). Some scholars have attempted dense data collection with regularly repeated testing sessions (e.g., Reference Nelson, Gut, Kopečková and LowieNelson et al., 2019).

It has become a golden standard in research on L3 phonology to test all three of the learners’ languages in order to use the learners as their own control, rather than relying on monolingual norms as controls (e.g., Reference Llama and López-MorelosLlama & López-Morelos, 2016; Reference SypiańskaSypiańska, 2016; Reference Balas, Kopečková, Wrembel, Calhoun, Escudero, Tabain and WarrenBalas et al., 2019; Reference Kopečková, Dimroth and GutKopečková et al., 2019). Further, there is a departure from the initial focus on non-native languages only, to eliciting the data also in the participants’ first language in order to establish the learners’ L1 baseline and to ascertain the possibility of tracing all potential mutual interactions between phonological systems of multilingual speakers. Thus far, the investigations have been mostly limited to a single feature and/or one testing time; however, exploring multiple phonetic features and longitudinally seems paramount for better understanding of the relative effect of crosslinguistic processes in non-native speech learning.

Another question that remains vital in the discussion of research design concerns the selection of subjects. Given the variety of multilingual learning contexts as well as the complexity of multilingual participants’ profiles, we need to account for different types of L3 learners, including, among others, active bilinguals, foreign language learners or heritage speakers. For multilingual speakers, the language status can be a very complex issue. For instance, a language may be their L3 chronologically yet not in terms of language dominance, and it may be subject to change over time. Moreover, some heritage speakers may re-learn their first language (i.e., a minority language) in a formal setting as the L3 (see Reference Wrembel, Marecka and KopečkováWrembel et al., 2019b). There are considerable differences between these L3 subgroups; foreign language learners may have enhanced metalinguistic awareness and language learning strategies, while heritage speakers are usually at an advantage with respect to phonological proficiency stemming from an earlier exposure to the heritage language. While much of L3 acquisition research has primarily focused on adult sequential bilinguals who typically acquired L2/L3 in a formal classroom context, some recent investigations have expanded the participant pool to include early bilinguals or young heritage speakers (e.g., Reference KopečkováKopečková, 2016; Reference Kopečková, Marecka, Wrembel and GutKopečková et al., 2016; Reference Llama and López-MorelosLlama & López Morelos 2016; Reference Wrembel, Marecka and KopečkováWrembel et al., 2019b). These L3 studies have shifted the focus from adults to children, and recently also to adolescents (e.g., Reference Balas, Kopečková, Wrembel, Calhoun, Escudero, Tabain and WarrenBalas et al., 2019, Reference Wrembel, Gut, Kopečková and BalasWrembel et al., 2020). Another possible category of L3 learners may allow for their first exposure to the L3 as an unknown language (see perceptual investigations by Reference BalasBalas, 2018). All in all, the vast majority of studies into L3 phonology to date have involved participants at post-initial stages of third language development (however, for a multi-feature analysis of phonological crosslinguistic influence at the initial stages of L3 acquisition, see Kopečková et al., in press). Further distinctions should be pursued with respect to the level of L3 phonological proficiency, comparing initial state versus more advanced L3 learners, which has not been pursued yet on a larger scale (however, for a small-scale exploratory study in this connection, see Reference PatiencePatience, 2018).

The choice of control participants has been widely debated in the L3 literature (e.g., Reference Cabrelli Amaro and WrembelCabrelli & Wrembel, 2016; Reference Puig-Mayenco, González Alonso and RothmanPuig-Mayenco et al., 2018), with the majority arguing against the use of monolingual controls as a benchmark for multilingual performance. An alternative solution is to compare trilinguals’ performance against that of bilingual control groups (e.g., Reference Llama and López-MorelosLlama & Lopez-Morelos, 2016). This is in line with Reference Westergaard, Mitrofanova, Mykhaylyk and RodinaWestergaard and colleagues’ (2017) proposal to compare L3 learners with two groups of L2 learners, with L1s that are the same as the previously acquired languages of the L3 group, in order to trace the exact contribution of the additional language involved in L3 acquisition. For a design featuring a trilingual group alongside two bilingual groups with alternating language combinations, see Reference AmengualAmengual’s (2021) study. Other alternative approaches involve observing the same group of L3 learners over time when they act as their own controls or applying a mirror-group design within the same language repertoire with L1 and L2 covarying and keeping the L3 constant (L1 X, L2 Y, L3 Z vs. L1 Y, L2 X, L3 Z; e.g., Reference Llama, Cardoso and CollinsLlama et al., 2010; Reference Cabrelli AmaroCabrelli, 2013) or with L1 and L3 covarying and L2 being constant (L1 X, L2 Y, L3 Z vs. L1 Z, L2 Y, L3 X; e.g., Reference Kopečková, Dimroth and GutKopečková et al., 2019; Reference Wrembel, Gut, Krzysik, Lewandowska, Balas, Calhoun, Escudero, Tabain and WarrenWrembel et al., 2019a).

Faced with a dearth of comprehensive measures to create a multilingual user’s profile, there is a need to adjust the existing language proficiency measures commonly used in SLA to be applicable in studies on TLA. To this end, adaptations need to be made to the existing instruments, including background questionnaires, language learning history and use tests, and proficiency measures relevant for different linguistic domains based on, for example, the language and social background questionnaire (LSBQ; Reference Anderson, Mak and Keyvani ChahiAnderson et al., 2018), language experience and proficiency questionnaire (Leap-Q; Reference Marian, Blumenfeld and KaushanskayaMarian et al., 2007), bilingual language profile (BLP; Reference Birdsong, Gertken and AmengualBirdsong et al., 2012) or multilingual assessment instrument for narratives (MAIN; Reference Gagarina, Klop and KunnariGagarina et al., 2012). Thus far, phonological proficiency assessment has frequently relied on self-declared levels or general proficiency measures yet increasingly it has shifted to oral accentedness ratings.

Another important methodological consideration involves the selection of experimental tasks. Speech production tasks applied in L3 phonological studies, similarly to SLA research, have ranged from word lists, through sentence and text reading to (semi-)spontaneous speech in an attempt to balance out the degree of control versus ecological validity. Increasingly, speech production has been elicited by means of delayed repetition tasks aimed to avoid direct repetition, with time elapsing between stimulus offset and response (e.g., Reference Wrembel, Gut, Krzysik, Lewandowska, Balas, Calhoun, Escudero, Tabain and WarrenWrembel et al., 2019a) or picture story tasks adapted for accent mimicry and phonological awareness evaluation (Reference Kopečková, Wrembel, Gut and BalasKopečková et al., 2021). The perceptual paradigms, in turn, have mostly featured AX or ABX discrimination rather than identification tasks which seem notorious for specifying response alternatives, the problem being magnified in the case of three phonological systems in interaction. For instance, Reference Wrembel, Marecka and KopečkováWrembel and colleagues (2019b) used AX discrimination to investigate the perception of L3 Polish sibilants; and ABX was applied for vowel discrimination in L3 French and German by Reference Lewandowska, Wrembel, Calhoun, Escudero, Tabain and WarrenLewandowska and Wrembel (2019). The question remains though, whether the discrimination procedure should involve several subtests with particular pairs of languages compared separately or rather embrace a crosslinguistic design with a set of randomized auditory stimuli from all of the three languages under investigation, thus rendering it very complex if not too challenging for the respondents. Other alternative L3 perceptual tasks include a forced-choice goodness task, which allows for elicitation of an association of a given allophone across multiple languages while the complexity of stimulus identification is avoided. Reference Wrembel, Gut, Kopečková and BalasWrembel and colleagues (2020) employed the forced-good paradigm to investigate the perception of rhotics and of final obstruent devoicing in L3 German by L1 Polish subjects with L2 English. In this study the participants heard two renditions of the same phrase differing on the last stimulus items embedded in a carrier phase and had to decide which phrase sounded more natural. One rendition was a target realization and the other was an accented language realization, in which only the investigated feature was manipulated, (i.e., L1-, L2-, or L3-accented). Moreover, other perceptual measures adapted from L2 speech research (e.g., Reference Cebrian, Mora, Aliaga-Garcia, Dziubalska-Kołaczyk, Wrembel and KulCebrian et al., 2011) include a crosslinguistic similarity task. For instance, Reference Wrembel, Marecka and KopečkováWrembel and colleagues (2019b) employed crosslinguistic similarity ratings to examine whether L3 Polish vowels would be perceptually assimilated to the L1 German or L2 English phonological categories or both. However, its cognitive load was high due to double response alternatives (i.e., L3 sounds compared against those of the L1 and L2 at the same time).

Recent language acquisition research endeavors seek to combine more traditional behavioral methods with electrophysiological evidence. In this vein, L3 scholars have started to tap into multilingual processing through the application of on-line methods of brain imaging (an electroencephalogram, or EEG). Event-related potentials (ERPs) constitute an innovative method in empirical linguistics that allows to identify early stages of language (and speech) processing using such measures as mismatch negativity (MMN) for auditory stimuli in phonological studies. To the best of my knowledge such investigations are under way for L3 phonology in new undergoing projects.

Although research into multilingual language acquisition is vibrant and growing, studies typically focus exclusively on one language domain (for an overview, see Reference Puig-Mayenco, González Alonso and RothmanPuig-Mayenco et al., 2018). However, what is missing is a broader picture of multilingual acquisition, processing, and use investigated across different domains. In particular, the phenomenon of crosslinguistic influence as approached from multiple perspectives still remains largely unexplored.

Recapitulating, the methodological considerations discussed in this section seem particularly relevant if we strive to embrace a holistic approach to L3 phonology research in which diverse methodologies of data collection and analysis are applied. Only by assuming such a broad perspective can we hope to contribute to a more nuanced understanding of multilingual speech and its acquisition and processing in general.

5.4 Theoretical Modeling of L3 Speech

Currently, four different approaches to the theoretical modeling of L3 phonology are conceivable: (1) expanding traditional L2 speech models to a multilingual perspective, (2) adapting L3 morphosyntactic models to phonological acquisition, (3) narrowing down broader interdisciplinary theories, and/or (4) putting forward new proposals for a dedicated model of L3 phonological acquisition. In the subsequent sections, these four theoretical scenarios will be elaborated on in search of the most satisfactory one.

Several approaches modeling the acquisition of speech have become prominent among SLA researchers over the years, including the following models: the Speech Learning Model (SLM; Reference Flege and StrangeFlege, 1995) and its recently revised version (SLM-r; Reference Flege, Bohn and WaylandFlege & Bohn, 2021); the Perceptual Assimilation Model (PAM and PAM-L2; Reference Best and StrangeBest, 1995, Reference Best, Tyler, Munro and BohnBest & Tyler, 2007); the Second Language Linguistic Perception (L2LP) model (e.g., Reference Escudero, Boersma and HamannEscudero, 2009); the Exemplar-Based model (Reference Pisoni, Johnson and MullennixPisoni, 1996); or the Native Language Magnet model (NLM; Reference Kuhl, Iverson and StrangeKuhl & Iverson, 1995).

The SLM (Reference Flege and StrangeFlege, 1995) posits that most L2 categories are initially perceptually assimilated to native ones, yet a new L2 perceptual category can be established over time if the L2 segment is sufficiently dissimilar from the L1 category (rather than similar or the same). If a new category formation is blocked by equivalence classification in case of less dissimilar sounds, it may lead to accented production in the L2 or even to shifts in L1 production. The revised version of this model, SLM-r, deals with sound learning across lifespan, however, it adds the individual difference perspective. Its major goal is to account for the process of reorganization of phonetic systems of individual learners in response to L2 phonetic input in a naturalistic acquisition context. The authors propose more fine-grained measures of the quality and quantity of input by introducing a new notion of full-time equivalent (FTE) years of L2 input, which is calculated based on the length of residence (LOR) and the proportion of L2 use. Further, while the SLM assumed that the accuracy of L2 perception confines the production accuracy, the SLM-r posits a hypothesis that production and perception co-evolve without precedence in L2 acquisition. However elaborate and explanatory the revised model appears to be, it still focuses the cross-language investigations to L1–L2 mappings without taking into the account the existence of other additional phonological systems in foreign language learners, thus failing to provide a better understanding of how speech acquisition functions from a multilingual perspective.

According to PAM/PAM-L2 (Reference Best and StrangeBest, 1995, Reference Best, Tyler, Munro and BohnBest & Tyler, 2007), non-native phonetic segments are perceptually assimilated to native phonetic categories based on their articulatory similarity to the native gestural constellations in the process of continuous refinement of speech perception. In turn, the NLM assumes that native-language phonetic categories are organized around prototypes, and L2 learning would require the phonetic perceptual space to be reorganized around newly established prototypes. Escudero’s L2LP aims at modeling the whole developmental process of L2 speech perception, ranging from naïve, non-native to advanced, native-like performance, based on the model’s computational architecture. It relies on phonetic, phonological, and psycholinguistic constructs to account for various L2 learning scenarios, with implications for speech processing in general. In Pisoni’s model, native phonetic categories are represented as clusters of exemplars that share certain acoustic parameters. It is through matching an incoming L2 signal to previously stored exemplars, new clusters can be formed.

All in all, the majority of traditional models of L2 speech acquisition focus on phonetic rather than phonological categories (i.e., context-dependent phonetic segments, exemplars or prototypes rather than phonemes or features) and make no specific claims related to how phonology works in the L2 learner’s mind. These models share an assumption of a common phonological space where phonological categories of the native and non-native languages interact, with L1 and L2 exerting mutual (bidirectional) influence. This common space may lead to composite or merged categories (i.e., L1 and L2 hybrids), on the one hand, and to divergence and more peripheral trends in two phonological systems, on the other (see, e.g., Reference Flege and StrangeFlege, 1995). However, they fail to make any specific predictions regarding the reshaping of the phonological space in multilinguals (for a detailed discussion thereof, see Reference Dziubalska-Kołaczyk, Wrembel, Sardegna and JaroszDziubalska-Kołaczyk & Wrembel, 2022).

An attempt to extend one of the L2 speech acquisition models (i.e., PAM-L2), to the multilingual perspective was ventured by Reference Wrembel, Marecka and KopečkováWrembel, Marecka, and Kopečková (2019b). According to the authors, the general perceptual assimilation mechanism, as posited in PAM-L2, applies also to L3 acquisition; the new sounds of the L3 are assimilated both to L1 and to L2 categories. Further, they interpreted the results of their two perceptual studies as evidence for discrimination between “similar” L1, L2, and L3 sounds, indicating that their beginner L3 learners seemed able to perceive subtle differences between highly similar speech sounds that would typically follow the single-category assimilation pattern and managed to develop separate L3 categories. The authors remain cautious not to draw strong conclusions from these exploratory findings, calling for further testing of the hypotheses of the extended PAM model for L3 learners.

The second approach to modeling L3 speech consists in adapting the existing L3 morphosyntactic models. The following theory-based explanatory models, designed specifically for multiple language acquisition, have emerged as the leading proposals to date: the Cumulative-Enhancement Model (CEM) by Reference Flynn, Foley and VinnitskayaFlynn, Foley, and Vinnitskaya (2004); the L2 Status Factor Model (L2SF) by Reference Bardel, Falk, Cabrelli Amaro, Flynn and RothmanBardel and Falk (2012); the Typological Primacy Model (TPM) by Rothman (e.g., Reference Rothman2015); the Linguistic Proximity Model (LPM) by Westergaard, Mitrofanova, Mykhaylyk, and Rodina (e.g., Reference Westergaard, Mitrofanova, Mykhaylyk and Rodina2017); and the Scalpel Model by Reference SlabakovaSlabakova (2017). It is worth pointing out that the above models operate mostly in the generative paradigm and were derived primarily from investigations into CLI in L3 morphosyntax.

What follows is a very brief recapitulation of major tenets of particular L3 models, as they are further elaborated on in other contributions to the present volume. The CEM by Reference Flynn, Foley and VinnitskayaFlynn and colleagues (2004) posits that language learning is cumulative. And all previously learnt languages may influence subsequently acquired languages. Thus, the L1 is not privileged as the only source of influence. Further, transfer can only be facilitative since the accumulated linguistic knowledge necessarily enhances subsequent language acquisition, therefore, nonfacilitative transfer is neutralized or blocked due to the lack of redundancy in linguistic representation. In turn, the L2SF by Reference Bardel and FalkBardel and Falk (2007, Reference Bardel, Falk, Cabrelli Amaro, Flynn and Rothman2012) assumes a privileged position of the L2, which prevails over the L1 as the main source of CLI. This model is psycho- and neurolinguistically motivated and posits greater cognitive similarity between the L3 and L2 and co-activation of non-native languages. Based on Reference ParadisParadis’ (2009) research, Reference Bardel, Falk, Cabrelli Amaro, Flynn and RothmanBardel and Falk (2012) maintain that non-native languages (L2, L3, Ln) are sustained mostly by declarative knowledge, while the L1 by procedural; and that these two types of knowledge have various memory sources and cerebral representations. The Typological Primacy Model by Rothman (e.g., Reference Rothman2011, Reference Rothman2015) posits that multiple sources of transfer are feasible and hypothesizes wholesale initial transfer from the typologically closest language in line with the principles of cognitive economy. Rothman explains the mechanisms by means of which the parser evaluates typological proximity based on implicational hierarchical continuum of linguistic cues in the L3 input. The LPM by Reference Westergaard, Mitrofanova, Mykhaylyk and RodinaWestergaard and colleagues (2017) and Reference WestergaardWestergaard (2019) builds on some aspects of the CEM and the TPM as it posits both facilitative and nonfacilitative influence from one or both of the previously acquired languages. According to this model, all acquisition (L1, L2, L3/Ln) is step-wise and takes place property by property, not wholesale. It hypothesizes that CLI can be triggered by structural similarity with linguistic properties of any previous language. Unlike the TPM, it assumes incremental property-by-property learning rather than holistic transfer. A somewhat related proposal is Reference SlabakovaSlabakova’s (2017) Scalpel Model, which shares the assumption that the L3 is learnt property by property from the very beginning of acquisition, and that wholesale transfer from one of the previously known languages is unwarranted. This model stresses the role of several cognitive and experiential factors in the L3 acquisition process, such as the structural linguistic complexity of the properties, misleading input, construction frequency in the target L3 as well as prevalent patterns of language activation or use. Recapitulating, the existing theoretical models of TLA provide quite conflicting explanatory accounts of this phenomenon. While there seems to be agreement as to the interplay of native and non-native sources of crosslinguistic influence, the predicting factors underlying its strength and directionality remain rather inconclusive.

Several L3 phonological studies have tried to corroborate the assumptions of the proposed theoretical models of L3 acquisition, irrespective of the fact that they originated from the morphosyntactic domain. It appears that no conclusive evidence was found for any of these models when applied to the domain of L3 phonology. For instance, Reference Hammarberg, Hammarberg, Hufeisen and FouserHammarberg and Hammarberg’s (2005) case study was in line with the tenets of the L2SF. In turn, Reference Wrembel, Cabrelli Amaro, Flynn and RothmanWrembel’s (2012) and Reference Lloyd-Smith, Gyllstad and KupischLloyd-Smith and colleagues’ (2017) foreign accentedness studies offered partial support to the TPM. Reference Kopečková, Marecka, Wrembel and GutKopečková and colleagues’ (2016) study on early L3 production of vowels was interpreted as lending support to the LPM.

When exploring the role of native and non-native languages in shaping the phonological acquisition patterns in the L3, Reference WrembelWrembel (2015b) demonstrated that some predictions of the existing L3 models were inconsistent with her findings, which evidenced a combined crosslinguistic influence, not restricted to either a default L1 or L2, that was both facilitative and nonfacilitative. In a series of parallel studies conducted on four groups of multilingual learners with different language repertoires, Reference WrembelWrembel (2015b) examined three selected measures of phonetic performance. The results of particular studies appeared domain-specific rather than general in nature, with foreign accentedness ratings pointing to CLI mainly from the L1; metaphonological awareness protocols reporting CLI prevailingly from the L2; and the VOT acoustic measurements indicating that VOT values in the L3 were intermediate between L1 and L2, irrespective of the language type and L3 characteristics. Consequently, the idea of a full transfer from only one language system was not corroborated by the data, and the crosslinguistic influence attested in Reference WrembelWrembel (2015b) was found to be gradual and structure-dependent, in line with the property-by-property transfer, as proposed by Westergaard and colleagues (see also Reference WrembelWrembel, 2020 for further discussion thereof). However, it seemed difficult to provide a unified account for the sources of CLI in the L3 as the process of L3 phonological acquisition appeared to be variable across individual learners and was mediated by the type of task as well as the investigated structure.

Additional support for the LPM was evidenced in a recent study by Kopečková and colleagues (in press), which was intended to verify the predictions of the L3 acquisition models by framing them in the context of L3 phonological acquisition at initial stages. Two groups of young sequential multilinguals (with L1 Polish and L1 German) learning their foreign languages (L2 English and L3 German or Polish) at school were investigated in parallel, focusing on the production performance of selected features (i.e., rhotics, final obstruent devoicing and the /v/ vs. /w/ contrast). It was the actual phonetic/phonological feature that was found to be the strongest predictor of CLI sources in the learners’ speech. On the one hand, the L1 German learners relied on their L1 to produce the rhotic and final devoicing in their L3 Polish, whereas L2 English helped them to produce /w/ in L3 Polish. On the other hand, L2-based CLI was attested more frequently in the L1 Polish group, especially in case of the rhotics production. Kopečková and colleagues (in press) interpreted the findings as partially corroborating the tenets of Westergaard and colleagues’ LPM in that CLI was found to be structure-dependent and occurring from both previously learned languages (L1 and L2), based on perceived structural similarity of specific features. Further, although CLI seemed to be prevailingly L1-driven at the early stages of L3 speech acquisition, the emerging patterns demonstrated numerous hybrid forms and individual variation, thus pointing to the existence of other factors that may stem from processing complexity, misleading input or construction frequency, as proposed by Reference SlabakovaSlabakova (2017) in her Scalpel Model (for a broader discussion, see Reference WrembelWrembel, 2021).

The third approach to modeling L3 phonology, based on the distinction which I propose in this section, is related to narrowing down broader interdisciplinary theories. For instance, Reference Kopečková, Marecka, Wrembel and GutKopečková and colleagues (2016) propose to frame their investigation within a Complex Dynamic Systems Theory (CDST) approach to multilingualism (e.g., Reference De Bot, Cabrelli Amaro, Flynn and Rothmande Bot, 2012, Reference De Bot, Ortega and Han2017). The major tenets of the CDST framework posit that languages constitute interconnected networks of subsystems marked by variability and that these systems are sensitive to initial conditions. The complexity of systems increases as a result of internal self-organization as well as interactions with external input since language systems are seen as connected growers. Reference Kopečková, Marecka, Wrembel and GutKopečková and colleagues (2016) suggest that such a dynamically based model could account for varying degrees and directions of crosslinguistic interactions as well as individual variation in phonetic performance exhibited by their multilingual participants. Due to its holistic and interdisciplinary perspective, CDST may be expected to offer greater explanatory potential than any of the models discussed thus far. However, this framework requires the adoption of specific longitudinal study designs with dense data collection, which is a methodological constraint not easy to meet (see Reference Golin, Gut, Kopečková and LowieGolin et al., 2019 for an example of such a study).

Given the special nature of L3 phonological acquisition combining phonological representation with neuro-motor articulatory routines, scholars have wondered whether a dedicated model may need to be proposed to better explain multiple and combined CLI patterns emerging from the research in this domain (see Reference Cabrelli Amaro and WrembelCabrelli Amaro and Wrembel, 2016; Reference Dziubalska-Kołaczyk and WrembelDziubalska-Kołaczyk & Wrembel, 2017). Therefore, the fourth theoretical scenario I envisage consists in putting forward novel proposals for a dedicated model of L3 phonological acquisition. For instance, a new theory called the Natural Growth Theory of Acquisition (NGTA) (Reference Dziubalska-Kołaczyk and WrembelDziubalska-Kołaczyk & Wrembel, 2017, Reference Dziubalska-Kołaczyk, Wrembel, Sardegna and Jarosz2022) was offered as a more optimal explanatory framework that can account for the complexity of the acquisition process in a multilingual mind. The NGTA is conceived as a general theory of language acquisition, which allows us to model the acquisition of phonology as well as morphology and other language domains. It stems from the framework of Natural Phonology (e.g., Reference Dziubalska-KołaczykDziubalska-Kołaczyk, 1990; Reference Donegan and StampeDonegan & Stampe, 2009) and is enhanced by Complexity Theory (e.g., Reference KretzschmarKretzschmar, 2015). Thus, it embraces the communicative and cognitive orientation of language and the conditioning impact of extralinguistic factors within a functionalist perspective. It assumes a gradual dynamic emergence of Ln phonology, shaped by the input from L1 and other Ln, and influenced by universals, typology, and context.

The NGTA assumes that all three linguistic variables (including L1, L2, Ln universals) influence the process of acquisition, however, their impact is moderated by the configuration of extralinguistic factors such as stages of acquisition and order of acquisition, input, and frequency and recency of use, age and age of acquisition, proficiency level in particular languages, metalinguistic awareness, and individual factors, all of which are related to a given acquisition situation. Universal preferences stem from extralinguistic principles, on the one hand, and they constitute preferability generalizations based on data input, on the other hand. As motivated by a distinction between System 1 and 2 proposed by Reference KahnemanKahneman (2011), the NGTA distinguishes two levels in language acquisition and processing: (1) automatic, manifested by, for example, articulatory routines and phonetic perceptual constraints, and (2) conscious and cognitively based, an approach which is in line with a dual approach to phonological acquisition. The theory assumes that the growth of attainment in the additional language is regulated by the degree of complexity of interdependencies among the relevant variables. Thus, multilingual learners’ phonology grows along their individual natural paths of acquisition. The fundamental research question is how much the individual paths converge and to what extent they remain divergent.

Since the NGTA is conceived as a general theory of language acquisition, it may enable us to explain the acquisition of morphology, phonology, and other language domains. Guided by the general assumptions and informed by the data generated in previous extensive research on (mor)phonotactics by Dziubalska-Kołaczyk and colleagues, specific predictions pertaining to the process of multilingual acquisition of consonantal clusters were formulated and discussed in light of linguistic and extralinguistic variables of the NGTA (see Reference Dziubalska-Kołaczyk, Wrembel, Sardegna and JaroszDziubalska-Kołaczyk & Wrembel, 2022).

Our research in progress aims at complementing the data-driven support from morphonotactics by further reanalysis of the existing data from Reference WrembelWrembel’s (2015b) series of studies on L3 phonology with the view to interpreting them in the framework of the NGTA. The accentedness rating study, among others, yielded support for universal phonetic grounding in the acquisition process; the perceived foreign accentedness patterns were found to be comparable across the participants’ groups and across different acquisition settings (i.e., SLA vs. TLA). In turn, the voice onset time study showed that with a growing language proficiency, the VOT values were more target-like in the participants’ L2 and L3. This dynamic change evidenced that phonological target values were emergent with the advancement of the proficiency level, which is in line with the specific predictions of the NGTA. This study provided also data support for the general assumptions of the NGTA in that all the linguistic variables (i.e., L1, L2, L3) exerted influence on L3 VOT patterns resulting in hybrid L3 VOT, and their relative impact was moderated by the configuration of linguistic and extralinguistic factors such as the recency of use and the order of acquisition. Finally, the data generated in the metaphonological awareness part of Reference WrembelWrembel’s (2015b) study reported conscious crosslinguistic awareness of L3 pronunciation problems and, at the same time, low automatic articulation control thereof. This provided support for the two-level model in language acquisition and processing embraced by the NGTA, juxtaposing articulatory routines and phonetic perceptual constraints with conscious, cognitively-based manifestations of phonological awareness. Finally, the attested significant correlations between the awareness measure with participant variables confirmed the complex network of interdependencies between linguistic variables and extralinguistic factors as stipulated by the NGTA (for a detailed discussion, see Dziubalska-Kołaczyk & Wrembel, 2022).

As the NGTA is couched in an ecologically valid approach to data and data analysis, at first it may not appear to be as rigorous as a typical formal theory. However, in its epistemological stance it follows a logically determined set of principles, it is firmly grounded in the linguistic theory of Natural Phonology, and enhanced by a dynamic perspective of complexity theory. The dynamic nature of cross-language interactions endorsed in this theory has found support in a large body of recent psycho- and neurolinguistic research, which corroborate continuous access to different language systems in multilingual speakers (e.g., Reference KrollKroll, 2020; Reference SoraceSorace, 2020). For instance, Reference Kroll, Dussias, Bogulski, Valdes-Kroff and RossKroll and colleagues (2012) demonstrated that all the languages in the multilinguals’ repertoire constitute dynamic systems undergoing continuous change, pointing to the impact that foreign language learning exerts on the native tongue and the generated regulatory mechanism therein. Further, Reference ChangChang (2012) demonstrated L1 phonetic drift from the onset of L2 learning and ascribed it to a novelty effect that enhances the encoding. As evidenced in a comprehensive overview by Reference KrollKroll (2020), cross-language interactions are acknowledged as persistent from the very onset of multiple language learning; they have been observed in different linguistic domains – lexis, grammar, and phonology – as well as in divergent conditions (see also Reference WrembelWrembel, 2021).

Reiterating, the present contribution aims to inform the ongoing debates on the optimal explanatory frameworks in TLA, by drawing attention to the specificity and complexity of L3 phonological acquisition. As the authors of particular models call for further empirical testing of the proposed hypotheses, it seems that such theory verifications should be carried out on a broader set of data and across different domains. Further, we need to explore the intricate relationship between factors that potentially influence multilingual phonological development. It seems probable that some factors may converge, thus triggering phonological CLI with combined strength, and that this dynamic interaction may be subject to change over time. In my opinion, it is of paramount importance to expand the explanatory potential of proposed models to various acquisition situations and different linguistic domains. Therefore, further studies into L3 phonological acquisition should attempt to foster more across-domain longitudinal investigations which may feed into a wider interdisciplinary framework.

5.5 New Insights, Novel Perspectives

The present section aims at overviewing some new research insights from a recent longitudinal “Multi-Phon” project (2017–2019) on CLI in phonological acquisition. This international project was conducted in parallel in Polish and German schools and involved a pool of forty young sequential multilinguals (12 years old), divided into two groups with L1 Polish/L2 English/L3 German and L1 German/L2 English/L3 Polish. They were tested in their L1, L2, and L3 with a battery of production and perception tasks at three data collection points spanning a school year. The methods employed featured delayed repetition, picture naming, picture storytelling and accent mimicry (for speech production and phonological awareness) as well as ABX and forced-good goodness task (for perception). The phonetic features under investigation included selected vowels, rhotic sounds, and final obstruent devoicing.

Primary research questions pursued in the project involved developmental trajectories of foreign language phonologies from initial stages of the L3 over the first year of classroom instruction; complex crosslinguistic interactions over time; the production and perception interface in L2 and L3; interindividual variation as well as the effects of L1 background and language proficiency. Among the major findings, we report a nonlinear (partially feature dependent) development of foreign language phonology; transitory stages of L3 phonology reflected in intermediate realizations; cumulative (“hybrid”) CLI yet with a prevalence for L1 transfer; as well as differential learnability of phonetic features. Further, we propose and test new measures to capture L3 learners’ perceived relatedness between the languages from their multilingual repertoire and to explore phonological awareness through accent mimicry.

Although metalinguistic awareness has been long acknowledged to be a key component of multilingual competence facilitating additional language learning (e.g., Reference JessnerJessner, 2006), relatively few studies have tackled it from the perspective of L3 phonology (but see Reference WrembelWrembel, 2015a; Reference KopečkováKopečková, 2018). One of the major reasons behind the dearth of research has been an open question on how to operationalize and measure phonological awareness across multiple languages. In a recent attempt, Reference Kopečková, Wrembel, Gut and BalasKopečková and colleagues (2021) tested the suitability of the delayed mimicry paradigm for assessing phonological awareness in young multilinguals. In this study L2/L3 phonological awareness was operationalized as the ability to mimic foreign accented speech in the L1 and to consciously reflect on the phonetic and phonological manipulations when performing L2- and L3-accented picture storytelling tasks. The study design involved two delayed mimicry tasks (for each foreign language separately) and independent foreign accentedness ratings of the participants’ performance in the mimicry task. The results showed different manifestations of phonological awareness as a function of the L1 background, with the L1 Polish participants being more accurate in imitating L2 accent and offering a greater range of metalinguistic comments, while the L1 German participants were more successful at imitating their L3 accent. On the whole, the participants exhibited primarily low levels of phonological awareness at the level of noticing in their L2 and L3. However, no significant changes were attested in the development of phonological awareness over the testing period. The novel application of the delayed mimicry paradigm to the multilingual context was found to be effective as the task tapped into implicit and explicit aspects of phonological awareness and elicited data featuring both segmental and suprasegmental features from multiple languages of young instructed learners. This elaborate measure made possible the assessment of phonological awareness through quantitative and qualitative approaches, by combining a feature-based detailed analysis with a holistic evaluation based on foreign accentedness ratings.

Another novel methodological contribution stemming from the “Multi-Phon” project is the proposal of a visual psychotypological measure focused on the learners’ perception of distances between multiple phonological systems (Reference Nelson, Krzysik, Lewandowska and WrembelNelson et al., 2021). The proposed measure can be framed as holistic or alternatively as feature-based. It attempts to reliably capture perceived relatedness between all languages in a multilingual’s repertoire to account for the interconnectivity in the organization of the multilingual mind. The conducted analysis demonstrated a significant L1 effect on psychotypology, with different patterns of interactions observed; the L1 German group clearly distinguished between the individual language pairs (i.e., L1 vs. L2, L1 vs. L3, L2 vs. L3) while the Polish group perceived respective languages as more equidistant. The elicited retrospective comments were aimed to supplement the quantitative measure with qualitative insights into the rationale behind the participants’ psychotypological distance measures. On the whole, the perceived language similarity tended to overlap with purely genetic or typological similarity, with English and German ranking highest in proximity. However, the elicited reflections revealed some asymmetry in perceived language distance both across and within groups, thus providing support for the claim that typology and psychotypology are not always congruent. The participants seemed to perform their crosslinguistic proximity assessment rather holistically, by noticing salient phonological features shared by the respective languages. Finally, the validity of the proposed measure was tested by comparing it to traditional binary paired-language proximity ratings. In future research we intend to investigate to what extent the measure of perceived crosslinguistic distance correlates with the participants’ speech performance.

Faced with a scarcity of research into the relationship between speech perception and production in multilinguals, a study by Reference Wrembel, Gut, Kopečková and BalasWrembel and colleagues (2022) offers some new insights into the long-standing issue of a perceptuo-productive link in the acquisition of foreign language phonology. In this exploratory longitudinal investigation into the acquisition of rhotic sounds in English, German, and Polish, the participants’ performance on a production task (delayed repetition) and a perception task (forced-choice goodness) was tested in their L2 and L3 at two testing times. General results demonstrated disparity between the two modalities, with rhotic perception being significantly more accurate than production. The link between the two modalities appeared to be modulated by how well established a given phonological system was in the multilinguals’ language repertoire; in the more proficient L2 both modalities showed a tendency for alignment, whereas in the less proficient L3 this relationship was different, indicating dissociation between very weak production and superior perceptual abilities. Other conditioning factors included L1 background, language-specific learnability of sounds and task type. A combined analytical procedure was applied to the results, with the group-level analysis supplemented with an exploration of individual change trajectories. The latter evidenced all four potential scenarios of the perceptuo-productive link (i.e., perception leading production or vice versa and both modalities being aligned or dissociated); as well as the development over time. The study aimed to provide a deeper understanding of the intricate link between the two modalities in L3 phonology development by showcasing its varied patterns possibly stemming from different acquisition circumstances and individual learner differences.

5.6 Future Directions

Due to the dynamic nature and inherent complexity of L3 acquisition research, our current knowledge of the field is far from providing a comprehensive account of the multifarious processes involved. Therefore, future studies should continue to address the theoretical and methodological challenges for multilingual phonological investigations, which the present contribution attempted to outline. First, in order to further pursue theoretical refinement for modeling multilingual acquisition of speech, we need to expand the explanatory potential of the existing models by providing further verification against a broader set of data and by including interfaces between various language domains. In my opinion, it seems worthwhile to zoom in on those theoretical approaches that offer a greater recognition of variability in the acquisition context and diversity of the learning experience modulated by a multitude of linguistic and extralinguistic factors.

Second, as we witness the shifting of the theoretical focus from a binary conceptualization of transfer to a broader perspective of dynamic, multidirectional cross-language interactions, the L3 research agenda should be redirected accordingly. With a view to triangulating different methodologies, a range of modern behavioral and online methods should be applied, including both cross-sectional and longitudinal designs to account for process-based investigations and phonological development over time. Research into L3 speech production and perception needs to be pursued with increased sample sizes of subjects, diverse language pairings as well as phonological features that pattern differently across languages. Further, it seems of paramount importance to focus more on individual differences in L3 phonological acquisition in order to trace variables that determine varied individual developmental trajectories.

Further, systematic comparisons of multilingual speech acquired in different settings have been largely unexplored as research to date has focused either on naturalistic acquisition or formal settings, resulting in different patterns of CLI attested. Future studies should aim to bring together these two perspectives to explore differences and similarities between them while controlling for such variables as the multilingual repertoire of the participants, their age and language proficiency. Similarly, different stages of L3 phonological acquisition should be contrasted by comparing patterns of CLI in initial versus advanced multilingual learners with otherwise comparable profiles. A final recommendation would be to apply a more interdisciplinary approach to L3 phonology by expanding the research across domains in order to investigate whether the patterns of CLI occur holistically in L3 acquisition or are domain-specific. By tackling the methodological and theoretical challenges discussed in this chapter, we may hope to further advance research on the domain of speech from a multilingual perspective.

6.1 Introduction

Research on vocabulary acquisition has been conducted for some decades now, with a large number of studies addressing various aspects. These aspects include the organization of the mental lexicon, development of different aspects of vocabulary knowledge, measures of vocabulary knowledge, the relation between vocabulary knowledge and other linguistic competences, factors that guide vocabulary acquisition, and so on.Footnote ¹ At the same time, the third language (L3) acquisition field has developed with many studies devoted to lexical aspects of L3 acquisition. However, despite the many similarities between the studies, the two fields largely seem to exist in parallel, with some exceptions. For example, while Reference WebbWebb’s (2020) recent handbook on vocabulary studies presents the state-of-the-art of second language (L2) vocabulary acquisition research, there is no chapter devoted to L3 perspectives, such as background language influences on the lexicon.

There are, however, good reasons to adopt a multilingual approach when doing and discussing research on vocabulary acquisition. As will be shown in this chapter, many L3 studies have been concerned with L1 and L2 crosslinguistic influence at the lexical level. Reference Cenoz, Hufeisen and JessnerCenoz and colleagues’ (2003) book The Multilingual Lexicon is an early example, and their 2001 book Cross-Linguistic Influence in Third Language Acquisition actually also mainly concerns the L3 lexicon, even if it is not explicitly mentioned in the title. Reference Clenton and BoothClenton and Booth’s (2020) recent volume First Language Influences on Multilingual Lexicons also includes chapters on L1 and L2 lexical influences, and as the editors state in the introduction, the book aims “to explore which factors relate to the multilingual lexicon in order to understand the complex relationship between first and additional languages of language learning” (Reference Clenton and BoothClenton & Booth, 2020: 3). Thus, while there is an abundance of L2 studies on various aspects of vocabulary acquisition, vocabulary studies within the L3 field seem to mainly have focused on crosslinguistic influence.

This chapter will try to bring these two fields —vocabulary acquisition and L3 research— together in order to allow for a general discussion of the characteristics of the L3 lexicon in relation to both the L1 and L2 lexicon. The general purpose is to provide an overview of research on L3 vocabulary learning. The chapter first brings up research on vocabulary acquisition on a general level, and begins with a short discussion of what it means to know a word. Then factors that are generally considered to govern vocabulary learning are presented. As we will see, one factor that has proven to be of great importance for vocabulary learning in both L2 and L3 is comparisons with previously learned languages. Research shows, among other things, that learners can benefit from having access to a closely related language when learning words in a new language, in such a way that they can effectively understand and acquire cognates. But having access to a closely related language can also be a disadvantage with regard to so-called false friends; that is, words that have a similar form in two or more languages but where the meanings differ. Clearly, crosslinguistic influence becomes particularly intricate when it comes to L3 learning, as the learners then know words in more than two languages. In L3 research, there are many studies investigating how the involved languages affect each other on a lexical level, transfer or crosslinguistic influences. This chapter will focus on this aspect of the L3 lexicon.

6.2 Vocabulary Acquisition Research

In L2 research, vocabulary acquisition was long considered a neglected area. One of the pioneers in the field, Paul Meara, published the now frequently quoted article “Vocabulary Acquisition: A Neglected Aspect of Language Learning” in 1980, in which he called for more research on vocabulary learning, as he considered that it had been given too little space compared to research on L2 morphology and syntax. Vocabulary acquisition research did gain momentum during the 1980s and 1990s, especially on English as a second language. During these decades, a number of anthologies and monographs were published on the subject (Reference CarterCarter, 1987; Reference McCarthyMcCarthy, 1990; Reference NationNation, 1990; Reference Laufer-DvorkinLaufer-Dvorkin, 1991; Reference Coady and HuckinCoady & Huckin, 1997; Reference Schmitt and McCarthySchmitt & McCarthy, 1997).

In early research on vocabulary learning, there was often a focus on the mental lexicon (see, e.g., Reference SingletonSingleton, 1999). Some questions that interested the researchers involved were, for example, whether the vocabulary of the first language (L1) is completely separate from the vocabulary of the L2, whether there is any connection between them, or whether the L1 vocabulary and the L2 vocabulary are stored in a single common mental lexicon. These questions were often investigated using a word association test (Reference MearaMeara, 2009), which allows analysis of the type of associations learners make when they encounter a stimulus word (the associations are usually categorized as paradigmatic, syntagmatic, or phonological). It has often been shown that learners at early stages of acquisition make form-based associations and that more advanced learners rely on meaning to a larger extent (Reference RingbomRingbom, 2007; Reference LindqvistLindqvist, 2010, Reference Lindqvist, Cabrelli Amaro, Flynn and Rothman2012). As we will see, the distinction between form and meaning is central to this chapter. More and more researchers were beginning to emphasize that vocabulary is an important part of second language learning during the 1980s and 1990s, and many also claimed that vocabulary is in many ways fundamental to L2 learning (Reference SchmittSchmitt, 2010). Studies have, for example, shown that there are clear links between vocabulary knowledge and other language skills; for example, reading, listening, and writing (Laufer & Goldstein, 2004; Reference Albrechtsen, Haastrup and HenriksenAlbrechtsen et al., 2008; Reference Milton, Bardel, Lindqvist and LauferMilton, 2013). Furthermore, it has been established that it is necessary to have access to an extensive and varied vocabulary both for reception and production in an L2 (Reference SchmittSchmitt, 2010). As for vocabulary depth (how well a word is known), research has shown that it is also necessary to know many different aspects of each individual word, to be able to both use it and understand it correctly (Reference Read, Bogaards and LauferRead, 2004; Reference WebbYanagisawa & Webb, 2020).

Quite naturally, this line of research has continued, also within the L3 field (Reference EckeEcke, 2015; Reference Singleton, Lengyel and NavrascicsSingleton, 2006), but many other aspects of vocabulary began to interest researchers already in the 1990s. In recent years, the number of studies on L2 vocabulary learning has increased, both in the form of monographs (e.g., Reference NationNation, 2001; Reference MiltonMilton, 2009; Reference SchmittSchmitt, 2010), anthologies (e.g., Reference Bogaards and LauferBogaards & Laufer, 2004; Reference Daller, Milton and Treffers-DallerDaller et al., 2007; Reference HirshHirsh, 2012; Reference Bardel, Lindqvist and LauferBardel et al., 2013; Reference Jarvis and DallerJarvis & Daller, 2013), and special issues of journals (e.g., Reference Beglar and HuntBeglar & Hunt, 2005; Reference Treffers-Daller, Daller and MalvernTreffers-Daller et al., 2008; Reference LauferLaufer, 2017) and journal articles. But there is still much that remains to be investigated. Reference Treffers-Daller, Daller and MalvernTreffers-Daller and colleagues (2008) pointed out that more studies on vocabulary learning are needed, so that we can understand general principles and processes in second language learning in a better way, and much research has been conducted since. Importantly, the authors also argued that it is necessary to study target languages other than English in order to arrive at a more complete picture of the vocabulary learning process. One could also argue that more studies on L3 vocabulary are needed to gain more knowledge about vocabulary learning in general. There are reasons to believe that L3 vocabulary learning entails certain differences in comparison with vocabulary learning in both an L2 and an L1. In the next sections, we will discuss what characterizes the vocabulary of an L3. What specific characteristics can be associated with learning and using words in an L3? What is the difference compared to vocabulary learning in an L2? But first, we will quickly bring up the question of what it actually means to know a word.

6.3 Aspects of Word Knowledge

The most basic question that can be asked in relation to vocabulary learning is probably: what does it actually mean to know a word? This is an issue that has been discussed by, for example, Reference Laufer, Schmitt and McCarthyLaufer (1997), Reference SchmittSchmitt (2010), and Reference NationNation (2001, Reference Nation and Webb2020). Reference NationNation’s (2001, Reference Nation and Webb2020) proposal of what aspects are included in vocabulary knowledge, which is very often cited in these contexts, is considered to be the most comprehensive overview of vocabulary knowledge (Reference Miralpeix and WebbMiralpeix, 2020).

Reference NationNation (2001, Reference Nation and Webb2020) sees word knowledge as consisting of form, meaning, and use. Knowledge of a word’s form has to do with how it is pronounced and spelled, as well as its morphology. Knowledge of the meaning of words concerns the connection between form and meaning, concepts and referents, and associations to other words. Word knowledge also pertains to how the word can be used, what grammatical functions it has, what words it often occurs with, and in which contexts the word is usually used. Nation further divides word knowledge into receptive and productive ability. It is thus clear that vocabulary knowledge is a complex phenomenon including various quantitative and qualitative aspects, and both productive and receptive dimensions. It can also be added that from an interactional perspective, it is crucial to have access to words to be able to communicate the intended meaning to the recipient and to understand. It seems logical that gaps in vocabulary lead to greater problems in communication than gaps in morpho-syntax. If you have access to the words that have the meaning you want to convey, you usually make yourself understood, even if the words deviate somewhat at the morphological level. This is also often the case if the word order does not correspond to the one that would normally be used in the target language.

Another aspect of vocabulary knowledge is that it takes many encounters with a word to achieve full mastery in all respects. Vocabulary learning is a long and complex process. Unlike many other areas of linguistic competence, vocabulary is an aspect that can be constantly expanded and developed, as learners constantly have the opportunity to acquire new words or new meanings of a certain word.

6.4 Factors Governing Vocabulary Acquisition

Many different factors can affect the learning of new words. A division between external (frequency, availability) and internal (pronunciation, spelling, semantic) factors has been suggested (Reference Laufer, Schmitt and McCarthyLaufer, 1997; Reference Peters and WebbPeters, 2020). For the purposes of this chapter, we will focus on the internal factor referred to as comparisons with previously learned languages, as L3 research above all has tended to examine this aspect of vocabulary. As mentioned, it is interesting to note that in the vocabulary learning literature, only similarities and differences between the L1 and the L2 have been discussed (see, e.g., Reference SchmittSchmitt, 2010), although it is likely that the learners in many previous studies had knowledge of more than one language when starting to learn another one. As the present chapter will show, L3 studies have often demonstrated that learners with knowledge of more than one background language make comparisons, consciously or unconsciously, between the languages they know (Reference De AngelisDe Angelis, 2007; Reference RingbomRingbom, 2007; Reference Cabrelli Amaro, Flynn and RothmanCabrelli Amaro et al., 2012).

In an L3 context, it is of course crucial to take all the background languages of the learner into account when examining the role of the previously acquired languages. Having knowledge of several languages can result in processes and outcomes in vocabulary learning and use that are different from those in L2 acquisition. On the one hand, it can be an advantage for the learner to be able to make comparisons between the words in all the languages s/he knows. Learning can be facilitated if there are similarities between words in the involved languages in terms of both form and meaning. On the other hand, it is not certain that there will always be a successful result when comparing languages. It is well known that there are a number of words that have the same, or similar, forms in several languages, but different meanings, so-called false friends or deceptive cognates (Reference RingbomRingbom, 2007; Reference OtwinowskaOtwinowska, 2015). The word form semester is an illustrative case. The English word “semester” corresponds to the French, Italian, Portuguese, and Spanish words semestre in both form and meaning and is thus an example of a “true cognate.” However, the word only corresponds in form to the Swedish word semester, but diverges in meaning and is thus a false friend, because in Swedish, semester means “vacation.” In Polish, in turn, wakacje is similar in form and meaning, and thus a cognate, to the English word vacation and the French counterpart vacances, as well as Spanish vacaciones, and Italian vacanze. To complicate things further, the Swedish word for “semester” is termin, and has a false friend in German, where Termin means “meeting.”

Many L3 studies have shown that learners are often influenced by their previously learned languages at the lexical level, and that this can result in both positive and negative transfer. These two main types of lexical CLI will be discussed in more detail in the following sections.

6.5 L1 and L2 Crosslinguistic Influence

Lexical transfer, or crosslinguistic influence (CLI), is a commonly studied phenomenon in both L2 and L3 learning. Naturally, CLI is far more complex in the latter case, because more languages are involved, implying that there are more possibilities for CLI to occur (Reference Cenoz, Hufeisen and JessnerCenoz et al., 2001; Reference Jarvis and PavlenkoJarvis & Pavlenko, 2008). One can imagine different explanations as to why lexical transfer has come to engage so many researchers. One reason may be that especially negative lexical transfer is relatively easy to detect in learner production, for example in the form of pure code-switches, where the learner uses a word from a language other than the target language. The next section will account for results regarding negative transfer. It will discuss the extent to which the background languages (L1(s) and L2(s)) influence the L3 lexicon; and why certain background languages tend to influence the L3 lexicon to a larger extent than others, by looking at the main categories of code-switching, word construction attempts, and semantic transfer.

6.6 The Organization of the Multilingual Lexicon

In L3 research, one question that has interested researchers is whether the vocabularies in the L1, L2, and L3 are completely separate from one another, or whether all the words known to an individual are stored in one common vocabulary (Reference Cenoz, Hufeisen and JessnerCenoz et al., 2003; Reference Singleton, Lengyel and NavrascicsSingleton, 2006). Most researchers seem to think that it is likely that the vocabularies are not completely separate, as research clearly has shown that learners mix words from several languages during both the writing and speaking process (cf. above). The question is rather to what degree the vocabularies are connected. According to Reference Singleton, Lengyel and NavrascicsSingleton (2006), for example, although there are indications that there is some kind of connection and interaction between the vocabularies, it is still not sufficient to suggest that the vocabularies are fully integrated. Even if two languages appear to be activated simultaneously during language use, this does not in itself indicate that the complete vocabularies are fully integrated. One could also imagine that the vocabularies are completely separate, but that there are strong connections between them. According to Reference Singleton, Lengyel and NavrascicsSingleton (2006: 134), there is actually more convincing evidence that the vocabularies are separated. Several neurolinguistic studies examining how languages are reactivated after language loss suggest that languages can be resumed selectively (Reference UllmanUllman, 2001, Reference Ullman, Hickok and Small2016). For example, a multilingual individual may resume only one of his or her languages, which could indicate that the vocabularies are stored in different locations in the brain. However, there seems to be more evidence for integrated lexicons when it comes to bilinguals (Reference Dijkstra, Wahl and BuytenhuijsDijkstra et al., 2019). There is certainly a need for more studies on this topic from a multilingual perspective on the learning process.

The only model that focuses on L3 vocabulary learning and the role of previously acquired languages seems to be that of Reference Hall, Ecke, Cenoz, Hufeisen and JessnerHall and Ecke (2003), which is based on the model suggested by Reference HallHall (2002) on L2 vocabulary learning. The Parasitic Model suggests that vocabulary learning is governed by the principle of minimal effort, that is the process of word learning in an L2. In other words, it is argued that the L3 will cognitively rely on a search for similarities between new and old (already known L1 and L2) words (Reference EckeEcke, 2015: 149). According to the Parasitic Model, the learning of a novel word consists of three steps:

1. 1 Form representation establishing

   • During the first step, the novel L3 word is connected to a “host”, which is the most similar word in any of the background languages.

2. 2 Building connections to frame and concept representation (syntax).

3. 3 Strengthening and automatization of representation (concept), as illustrated with Ecke’s data from Spanish (L1) speaking learners of German (L3) with English (L2).

Based on these three steps, the Parasitic Model suggests that there are form-based, frame-based, and meaning-based instances of CLI. Evidence for (parts of) this model on vocabulary learning has been found in several studies on L3 learning that have investigated errors of form, frame, and concept (e.g., Reference Ecke and HallEcke & Hall, 2000; Reference Ecke, Cenoz, Hufeisen and JessnerEcke, 2001; Reference González AlonsoGonzález Alonso, 2012; Reference ChenChen, 2020; see also Reference EckeEcke, 2015: 156–158 for an overview). These studies show that these psychological connections exist and have an impact on L3 learning. However, other factors also come into play. Reference EckeEcke (2015: 150 f.) mentions learner factors, learning factors, language factors, event factors, and world factors, which can explain why there exist various types of errors and CLI.

6.7 Final Remarks

This chapter aimed to present an overview of the characteristics of the L3 lexicon. While research on L2 vocabulary acquisition has focused on many different aspects of vocabulary knowledge, the vast majority of the studies within the L3 field have looked into L1 and L2 crosslinguistic influence, or transfer, in the L3 lexicon. The overview shows that most of these studies have been concerned with negative transfer, in the form of code-switches, word construction attempts, and semantic extensions. This is probably due to the fact that this type of transfer is relatively easy to detect in learner production, as opposed to positive transfer, with the latter resulting in target-like use in the L3. Overall, it can be concluded that previous research has shown that both L1 and L2 affect the L3 at the lexical level. The extent to which a particular background language influences the L3 largely seems to be determined by the (psycho-)typology factor, as many studies have shown that a closely related background language tends to be activated and used in the L3, rather than a more distant one. However, some researchers have also suggested that L2 status plays a role, but often in interplay with the (psycho-)typology factor. The proficiency factor has also been brought up in several studies, which indicated that CLI occurred from a background language in which the learners were more proficient. However, there are also indications that L2s in which a learner is highly proficient can lose its status as an L2, thus becoming inaccessible for transfer. Proficiency in the L3 also played a role. For example, it seems that CLI decreases as the learners become more advanced in the L3. However, semantic, or meaning-based, transfer seems to prevail at later stages of acquisition too, while formal transfer is less frequent. As for positive transfer, research has shown that multilingual learners can benefit from knowing cognates, especially in receptive tasks. Importantly, both L1 and L2 have a facilitative effect.

Some previous studies have also shown that function words tend to be transferred from an L2 rather than an L1 in L3 production, while content words seem to come from both L1 and L2 to a greater extent. These results could be explained from a neurolinguistic perspective. However, more research is needed to confirm or infirm these results. It would be interesting to further examine this question in languages where some categories of function words are not present (e.g., prepositions in Finnish). What would transfer patterns look like in such cases?

In future research, there is a need for more studies with well thought-out research designs, which would allow for a closer examination of transfer patterns. Many of the studies accounted for in this chapter have looked into spontaneous oral production or written production. Other types of tasks, task combinations, and combinations of languages are needed to gain further knowledge about the characteristics of the L3 lexicon.

7.1 Introduction

How bilinguals handle their two languages within their one brain has intrigued both laypeople and researchers for a long time. What we can see in their behavior seems paradoxical: on the one hand, they can perfectly speak and understand in one given language without being constantly troubled by interference from the other; on the other hand, they can switch swiftly from one language to the other, again without much apparent trouble. So, what does this mean: are the languages separated from each other in their minds, or are they stored and processed together with some additional option to address them separately if needed? This question, and in particular, how bilinguals’ memory for words in their two languages – the bilingual mental lexicon – is structured and accessed has been studied for some decades now.

However, while it is surely to be applauded that the monolingual case is no longer uniformly assumed the standard by researchers of language processing, bilingualism does not cover the whole story either. Speaking three or more languages is certainly no exception among the world’s population; for instance, 25 percent of citizens in the EU do so to the extent that they speak these languages well enough to have a conversation in them (European Commission, 2012; see also Eurostat, 2019, for a similar number: 21 percent), often as a result of school education, possibly in combination with a multilingual environment. Note however that the prevalence of trilingualism (or more) can amount to percentages way above 50 percent in (often smaller) countries like Luxembourg, the Netherlands, Slovenia, Malta, or Denmark.

Considering these people who speak three or more languages, many questions are simple extensions of the ones asked for bilinguals: are words of the different languages represented in a shared or separate fashion, and when are they co-activated during lexical access? Such parallel questions for trilinguals compared to bilinguals have led some researchers to claim that “there is no need to develop a specific model for such multilingual processing” (Reference de Botde Bot, 2004: 17).

However, the multilingual situation is not just an extension of the bilingual case by one or more languages. Already in bilinguals, many phenomena display a large degree of asymmetry, mostly such that the more dominant language – often the native language (L1) – exerts a larger influence on the weaker second language (L2) than vice versa. In trilinguals, this configuration becomes disproportionally more complex and more dimensional. For instance, while the L2 might have been acquired through the L1, the third language (L3) might have been learned in the country where the L2 is spoken, possibly resulting in disproportionally weaker links between the L1 and L3 than between the other languages. Similarly, two out of the three languages may be typologically close or share the same script while the third may not, possibly affecting the structure of the trilingual lexicon.

This chapter presents a summary of what we currently know about the structure and function of the tri- and multilingual lexicon. This will be done from a predominantly experimental/psycholinguistic perspective, reviewing findings of studies that have gathered data on these questions in the lab under tightly controlled conditions, using experimental paradigms like lexical decision (where the participant has to decide as quickly as possible whether a letter string is an existing word; a method to study word recognition) or picture naming (to study word production), in which not only accuracy, but also reaction times are important dependent variables. While the ecological validity (i.e., how far the elicited participant behavior represents real-life language processing) is certainly lower for these methods than that for observational studies (for instance, where conversations are elicited and transcribed; see, e.g., Chapter 6 of this volume), experimental techniques have the advantage of a high degree of systematicity and standardization. They are based on the attempt to break down language processing into small and more easily manipulatable processes, like the recognition or production of single words. This serves to effectively prevent potential confounds, such as learner strategies to avoid difficult words altogether or to infer their unknown meaning from the context. Experimental evidence can be regarded as an important source of evidence that complement findings “from the field.”

Classical psycholinguistic models of (mono- or bilingual) word processing typically distinguish between word representation or storage (the stored knowledge about a word in memory) and word processing, access, or function (how this knowledge is retrieved and used when needed; e.g., Reference McClelland and RumelhartMcClelland & Rumelhart, 1981; Reference Dijkstra, Van Heuven, Grainger and JacobsDijkstra & van Heuven, 1998). While these are two independent issues, they are also tightly related: If words from different languages are stored separately from each other “in different places in memory,” chances are higher that they will also be processed independently. If they are stored together, they will likely influence each other more. The relative focus on each of these two questions has shifted through the years, and I will review findings speaking to each of them as they became predominant in the literature at a certain time.

Findings from bilingual studies will be mentioned only if relevant to the trilingual situation, which is especially the case where the historic origins of experimental research into multilingual language processing are concerned (for reviews of bilingual word processing, see, e.g., Reference Kroll, Ma, Fernández and Smith CairnsKroll & Ma, 2018; Reference Santesteban, Schwieter, Cieślicka and HerediaSantesteban & Schwieter, 2020). The chapter does not discuss the quite extensive literature on trilingual language switching, both due to space restrictions, and because the central issue in these studies – that of language control, an instance of executive control – is one that is external to the lexicon itself (Reference Crinion, Turner and GroganCrinion et al., 2006; Reference Abutalebi and GreenAbutalebi & Green, 2008; Reference Coderre, Smith, Van Heuven and HorwitzCoderre et al., 2016), typically studied with a highly specific and artificial paradigm of (forced and frequent) language switching that may lie even further away from real-life language processing than other standard laboratory techniques do.

7.2 Early Research on the Structure of the Bilingual Lexicon

One of the oldest issues in psycholinguistic research into bi- and multilingualism is the question whether a multilingual’s memory for words, the mental lexicon, is separated by language or not. The tradition of experimental enquiry into this question began with the interest in bilingualism. In 1963, Kolers asked whether, in the bilingual case, “verbally defined past experiences were tagged, or coded, and stored in common” or rather “in a form specific to the language the individual used to define the experience to himself” (Reference KolersKolers, 1963: 291). He investigated this using a free association task, varying the language (native language or L2 English) of both stimulus words and responses. The idea was that if memory is shared between languages, associations to a given stimulus (e.g., English table or Spanish mesa) should be translations of each other. In contrast, different associations depending on stimulus and response language would suggest structurally separated memory systems. The result pattern was complex and analyzed in different ways, but the critical result was that only about a third of associations to a certain stimulus (e.g., boy) in the different stimulus-response language combinations were direct translations of each other; the others were different (e.g., girl, man, trousers). Kolers interpreted these results as general evidence for the “separate memories” hypothesis.

In my impression, the separate lexicons hypothesis became the most accepted one in those days. The main finding that led to this opinion was the discovery of language switch costs (i.e., the fact that producing or comprehending language-mixed language material takes longer than unilingual material; e.g., Reference KolersKolers, 1966; Reference MacNamaraMacNamara, 1967; Reference Macnamara and KushnirMacnamara & Kushnir, 1971). The rather mechanistic interpretation of such findings was that they suggested the operation of a time- and resource-consuming language switch mechanism, which would be unnecessary if both languages were stored and processed together.

Another interesting early line of research asked whether language membership is a major organizing principle for a multilingual’s memory for words. For instance, Reference Dalrymple-Alford and AamiryDalrymple-Alford and Aamiry (1969) examined in how far bilinguals, during free recall of mixed-language word lists, cluster their responses for recalled words according to language or according to semantic categories. They found that they do both, leading them to conclude that “the main organizational principle in bilingual free recall is one in terms of conjoint category and language membership” (767).

Findings like that demonstrate that the initial line of reasoning, according to which switch costs suggest the existence of separate memory stores, was too naïve. Reaction time costs also occur when, for instance, semantic categories are switched during picture naming (as evident in, for instance, the long RTs for the first trial in a “semantic block”; Reference Belke, Meyer and DamianBelke et al., 2005; Reference Navarrete, Del Prato, Peressotti and MahonNavarrete et al., 2014); however, no one would go as far as to claim separate memory systems for words from different semantic categories. Thus, while words of the same language may indeed “cluster together” in memory, the notion of entirely separate stores with a time-consuming switch mechanism was no longer tenable.

7.3 Language-Selective versus Non-selective Lexical Access: Bilingual Models

Later on, the focus of research on multilingual lexical processing shifted from the static storage question (the “where”) to a more dynamic view of processing (the “how”): during perceiving or producing a word in one language, how far do stored memory representations from the other language(s) of a multilingual play a role as well? This question has been named the “selective versus non-selective lexical access” debate (Reference Beauvillain and GraingerBeauvillain & Grainger, 1987; Reference de Groot, Delmaar and Lupkerde Groot et al., 2000). It developed under the influence of the emerging consensus under psycholinguists that word processing is a competitive process: Both during word recognition and word production, other, non-target word representations become co-activated and compete for recognition or production (for highly prominent models of word recognition and production, see Reference McClelland and RumelhartMcClelland & Rumelhart, 1981; Reference Levelt, Roelofs and MeyerLevelt et al., 1999). For the multilingual case, the obvious question is: do word representations of the non-target languages enter the competition as well, even if undesired in a language-specific task context? Evidence for such interlingual interactions in monolingual contexts would be strong support for a fundamentally and invariably nonselective function of the multilingual lexicon.

The first, and probably still best-known model of bilingual word recognition is the Bilingual Interactive Activation Model (BIA; Reference Dijkstra, Van Heuven, Grainger and JacobsDijkstra & van Heuven, 1998; Reference Dijkstra, Van Heuven and GraingerDijkstra et al., 1998a), an extension of Reference McClelland and RumelhartMcClelland and Rumelhart’s (1981) Interactive Activation Model. While a full description of the original model is beyond the scope of this chapter, what is important is that the input (i.e., an encountered written word) activates matching word representations in a bottom-up fashion: visual features activate letter representations that contain them, which in turn activate word representations that contain them, which should in turn activate relevant word meanings (however, this semantic level is not implemented). On every level, several candidates are initially active because they partly match the input. These candidates compete with each other until the best matching representation “wins the race.”

To this existing, localist model of visual word recognition, two features were added to the BIA (Figure 7.1): first, on the word level, there are two “sub-lexicons,” each of which contains the words of one language (here, English and Dutch); and second, on top of the word level, a new language level was added, containing language nodes that represent the currently active language. They are activated by words from the corresponding language, but also possess inhibitory connections to the respectively other lexicon below them, such that, in a highly language-specific context, words from the non-target language can be suppressed. That way, the BIA model represents a compromise between language selectivity and non-selectivity: While words from the two languages function as one lexicon by default and are all activated by incoming visual input, they are also “addressable” in separation. Note that the successor model BIA+ (Reference Dijkstra and Van HeuvenDijkstra & van Heuven, 2002) does not contain this feature of top-down suppression from the language nodes anymore, as this function is taken over by an additional “task / control system.” This bears resemblance to Reference GreenGreen’s (1998) (noncomputational) inhibitory control model in which the role of nonlinguistic control processes is emphasized. This discrepancy between the two successor models already demonstrates that the regulation of the co-activation of a non-target language remains a nontrivial issue.

Figure 7.1 The Bilingual Interactive Activation (BIA) model.

(Reference Dijkstra, Van Heuven, Grainger and JacobsDijkstra & van Heuven, 1998)

As already mentioned, in these models, the conceptual level on which word meanings are represented (which should be on top of the language level) is not implemented. Generally, though, word meanings are thought to be represented in a shared fashion across languages (Reference Kroll and StewartKroll & Stewart, 1994; Reference Van De Putte, De Baene, Brass and DuyckVan de Putte et al., 2017; Reference Dijkstra, Wahl and BuytenhuijsDijkstra et al., 2019), such that the meaning representations of the Spanish word mesa and the English word table in a Spanish–English bilingual are assumed to be identical.

In word production, no computational model has, to my knowledge, been developed for the bilingual case. Reference de BotDe Bot (1992) discussed how Levelt’s production model could in principle be adapted to the bilingual case, and arrived at very similar proposals as can be found in the BIA. Roughly speaking, one can conceptualize the speaking process as the reverse from what happens during recognition: speaking starts with the to-be-expressed meaning (the semantic level), then proceeds to the selection of lexical candidates that match this meaning, and ends with the selection of the building blocks of these words (phonemes), and finally their articulation. De Bot arrived at the suggestion that while the first formulation of the message content, the “macroplanning” (comparable to the activation of concepts in recognition) is not language-specific, the subsequent “microplanning” should contain a specification of the intended language of the utterance, something that may be comparable to the language nodes in the BIA. As in the BIA, the lexicon, according to De Bot, is “one common lexicon in which items are connected in networks which enable subsets of items to be activated. One such subset can be the items from a specific language” (Reference de BotDe Bot, 1992: 14). Finally, at the lowest level, phonemes are assumed to be stored in one common system to which possibly new L2 sounds that do not exist in L1 are simply added.

7.4 Experimental Tests of Language (Non)Selectivity

The critical test for the claim of language-nonselective lexical access as a “hardwired” property of the bilingual word processing system is to create a monolingual task situation (or “language mode”; Reference Grosjean and NicolGrosjean, 2001) and to investigate whether any effects of the non-target language are still measurable. Using this logic, there are two major strands of potential evidence for non-selectivity: (1) the comparison of the processing of words that do or do not “overlap” between languages in a certain way, like cognates, false friends, and words with cross-lingual neighbors; and (2) distractor and word priming paradigms. Generally speaking, in the bilingual literature, evidence for nonselectivity of both kinds has been provided, though not unanimously for all stimulus types and tasks (for a review, see Reference Kroll, Ma, Fernández and Smith CairnsKroll & Ma, 2018). Thus, a certain extent of nonselectivity seems to be inherent to the bilingual lexical system. However, we still do not know to how many languages such nonselective activation extends: Suppose that one speaks five languages, will word candidates in all five languages then become active when only one language is the target language? This seems fairly inefficient. In what follows, I will quickly review the bilingual findings on each of these strands and then move on to whether they have been applied to trilinguals to assess the limits of nonselectivity in terms of the number of co-activated languages.

7.5 Asymmetrical Patterns of Trilingual Co-activation

While the question of (non)selectivity asked in the previous section was a simple extension of the one asked for bilinguals, tri- and multilinguals offer the opportunity to ask much more nuanced questions. For instance, if it is true that all languages known to an individual can in principle influence each other during processing, which language interacts the most with a given target language?

In bilinguals, the standard finding is that L1 influences the weaker L2 more than vice versa. However, we do not really know why this is: is it because it is learned earlier, or because it is used more often or with higher proficiency, or is it possible that the native language has a special status anyway, such that it can simply not be fully de-activated? In the multilingual situation, with either several non-native or native languages, an opportunity might arise to tear apart some of these factors. For instance, one could ask whether the L1 – as the strongest, most proficient, and most often used language learned first in life – really is always the most influential language when it comes to non-native language processing. While this is clearly what models of multilingual processing would predict – stronger, more often activated languages are the hardest to de-activate – there is also much anecdotal evidence of multilinguals especially struggling to sufficiently differentiate their several non-native languages, with language proximity as an additional, potentially crucial factor. Such anecdotes seem to suggest that the native language, albeit usually the strongest one, might also possess a “special status” that limits its influence on other languages compared to the competition between several later acquired languages. Even though we still know very little on such asymmetries of cross-language effects in trilingual processing, some evidence has accumulated that may at least give us some first ideas about the matter.

7.5.1 Patterns of Cross-Language Effects in Trilingual Word Production

Many observational studies have attested a strong role for the L2 during L3 lexical acquisition, sometimes stronger in fact than that of the L1 (e.g., Reference DewaeleDewaele, 1998; Reference Williams and HammarbergWilliams & Hammarberg, 1998; Reference Ringbom, Cenoz, Hufeisen and JessnerRingbom, 2001; see also Chapter 6 of this volume). However, there are as yet only few studies that use experimental psycholinguistic techniques to elucidate patterns of cross-language activation during trilingual processing, for instance during word production. As mentioned already in Section 7.4.4, we conducted two yet unpublished studies on the question of which of the two non-target languages, L1 and L3, influence L2 word production the most. We did this in a language set of three typologically similar Germanic languages (German, Dutch, and English). While L2 English was acquired before L3 Dutch in our population, Dutch was typically used more often and more recently due to being the immersion language. Our motivation was to test the anecdotal evidence of particularly strong interference of another non-native language during L2/L3 production, compared to the effect of L1.

Using phono-distractor effects (with distractors that overlap phonologically with the target translation), we found that both L1 and L2 distractors gave rise to similarly sized interference effects (both in RTs and error rates) compared to unrelated (Dutch or German) distractors (Reference Lemhöfer, Schriefers and SchellenbergerLemhöfer et al., submitted). However, the patterns were fairly heterogeneously distributed, with some participants having much larger effects for L3 Dutch, some for L1 German, and many about the same. Despite taking many different measures on each participant’s relative proficiency, exposure, and usage for the three languages, none of these measures reliably predicted whether an individual would be influenced more by the L1 or L3 during L2 production. Thus, more research is needed on this issue. Possibly, inhibitory control skills (that we did not measure) play a decisive role here, with some participants being very efficient and practiced especially in inhibiting the L1, but not so much the new language (L3), while others may apply less inhibition of other languages altogether. Despite this open matter, the study shows however that an L3 can be at least as strong an interferer during L2 production as an L1 is, something that has not been shown in a systematic experimental study before.

In the second series of studies, we took a closer look at direct translation distractors in the PWI paradigm (Reference Schriefers and LemhöferSchriefers & Lemhöfer, in preparation). Recall that bilingual studies have typically and paradoxically shown facilitatory effects for such (usually L1) distractors, mostly in the written modality and with the distractor appearing before the picture (i.e., with a negative SOA). Again, we used L1 German and L3 Dutch translation distractors in L2 English picture naming and additionally varied SOA (–200 or 0 ms) and distractor modality (visual vs. auditory). As can be seen in Figure 7.2, the resulting effect of translation distractors was inhibitory (indicated by positive values on the y-axis) whenever the distractor was in L3 Dutch. When it was in L1 German, it turned from inhibitory to facilitatory depending on both SOA and modality: more negative SOAs tended to make the effect more inhibitory, as did turning to the visual instead of the auditory modality.

Figure 7.2 The translation distractor effect (positive numbers equal an inhibitory effect of translation distractors) for trilinguals naming pictures in their L2 English, depending on SOA, distractor modality, and distractor language.

(Reference Schriefers and LemhöferSchriefers & Lemhöfer, in preparation)

As I have already described in Section 7.4.4, we interpret this result pattern as evidence for a mixture of two mechanisms at play during this task. The first is a predictive/conceptual/possibly strategic mechanism that causes translation distractors to facilitate responses, either through pre-activation of the target concept, or through a predictive mechanism also described in Section 7.4.4, based on the increased familiarity of words that refer to members of the item set. The second process is translation competition that causes inhibition, as in the phono-distractor effects described earlier. The clear difference in effects between the two distractor languages, in our view, is one of speed: distractors in L3, a late acquired and thus “slow” language, are simply not recognized soon enough to activate the target concept long enough before picture onset, such that the first facilitatory component does not really come into play (but note the trend toward more facilitation – or less inhibition – for visual Dutch distractors at SOA –200 compared to SOA 0). The same holds for less negative SOAs and the auditory modality (in which word recognition, when measured from word onset, takes longer).

In another trilingual translation distractor study, Reference Kujalowicz and ZajdlerKujalowicz and Zajdler (2009, exp. 1; see Section 7.4.4) observed facilitation of L3 Chinese picture naming latencies when accompanied by visual translation distractors (SOA 0) in L2 English, and (descriptively) to a lesser extent when accompanied by L1 Polish distractors. While the direction of effects differs from ours (facilitation rather than inhibition), the fact that effects were larger for another foreign language than for L1 distractors is in line with our findings. The authors interpret their results in terms of the importance of instruction language, since the participants had learned Chinese through English rather than Polish. However, whether it was indeed that factor or rather the foreign language status that makes the L2 a more potent influencer of L3 processing is impossible to work out.

In a study using the phoneme monitoring paradigm, Reference Tomoschuk, Duyck, Hartsuiker, Ferreira and GollanTomoschuk and colleagues (2021, exp. 1) showed that the L2 interfered more with L3 lexical access than the L1 did: Both in terms of RTs and error rates, the Dutch–English–French trilinguals had more difficulties rejecting phonemes not contained in the French picture name when they did occur in the English name than whey they occurred in the Dutch name. However, when the task was conducted in L2 English, it was the L1 Dutch that interfered more than L3 French, restricting the idea of a “foreign-language effect” (particularly strong influence among foreign languages) to situations when the weakest L3 is the target language. In exp. 2 of this study, using a newly taught artificial language, there was some, but not very robust and consistent evidence for a possible role of instruction language (i.e., whether new L3 words were learned via Dutch or English translations).

A final study that substantiates the idea of an especially competitive relation between L2 and L3 is an experimental investigation on L3 lexical attrition (Reference Mickan, McQueen and LemhöferMickan et al., 2020). In that study, we first taught Dutch–English bilinguals previously unknown Spanish (L3) words. On the next day, participants went through an interference phase in either Dutch (L1) or English (L2) where they had to repeatedly access the translations of half of the learned Spanish words. While we observed a general interference effect – which we interpret as an experimentally induced form of “attrition” – in the sense that Spanish words for which translations had been retrieved were more often forgotten than the non-interfered words, there was also an indication of this effect being stronger for the L2 English as an interfering language than for L1 Dutch. Thus, usage of another foreign language seems to bring about more forgetting of newly acquired L3 words than that of the mother tongue.

A final instrument to look at interlingual competition during trilingual word production is the Stroop paradigm. In a very comprehensive study on three trilingual populations differing in whether their three languages shared the same script, Reference van Heuven, Conklin, Coderre, Guo and Dijkstravan Heuven and colleagues (2011) systematically varied response and stimulus language. The general pattern was that Stroop interference and facilitation effects were mainly determined by script overlap between response and stimulus language, rather than by language status (L1/L2/L3). Languages that share the same script tended to show interlingual Stroop effects that approached intralingual effect sizes, while effects between different-script language pairs were reduced.

Altogether, while there is indeed some support of the notion that, during the production of a foreign language, another non-native language may be co-activated to a larger degree than the L1, not all experimental studies substantiate this claim. Rather, the L1 seems to remain a strong source of cross-language influence too, especially for individuals with limited proficiency in the L3 or those who learned the L3 via the L1, or when L1 and L3 are typologically close.

7.5.2 Translation and Priming Studies

Another strand of research is based on trilingual translation and priming studies, many of which are interested in the exact nature of the pairwise relationships between any two languages. An important issue in that part of the literature is how translation equivalents are connected in the mental lexicon. Two principal possibilities have been distinguished (Reference Potter, So, Von Eckardt and FeldmanPotter et al., 1984): the two-word forms may only be connected by way of the shared, supposedly language-independent semantic concept (“concept mediation”); alternatively, there may be direct links between them on the word form level (“lexical association”). Originally, Reference Potter, So, Von Eckardt and FeldmanPotter and colleagues (1984) reported evidence for the concept mediation hypothesis in bilinguals. However, additional evidence for supposedly further modulating factors led Reference Kroll and StewartKroll and Stewart (1994) to propose the famous Revised Hierarchical Model (RHM; see Figure 7.3). According to that model, L2 word representations initially possess no direct connection to their concepts, but are connected only to their L1 translations, due to the fact that novel L2 words are often acquired through association with their L1 translations. The reverse connections from L1 to L2 word forms are, however, relatively weak. With increasing proficiency, direct conceptual links may develop for the L2. This model explains the finding that L2–L1 translation is typically faster than in the reverse direction, and that semantic blocking, a manipulation thought to induce interference between related concepts, affected only L1–L2, but not L2–L1 translation RTs (Reference Kroll and StewartKroll & Stewart, 1994; for a critical evaluation of the RHM, see Reference Brysbaert and DuyckBrysbaert & Duyck, 2010; Reference Kroll, Van Hell, Tokowicz and GreenKroll et al., 2010).

Figure 7.3 The Revised Hierarchical Model of bilingual word translation.

(Reference Kroll and StewartKroll & Stewart, 1994)

For trilinguals, the question arises how an L3 can fit in this picture. Reference de Groot and HoeksDe Groot and Hoeks (1995) let Dutch–English–French trilinguals carry out a translation production and translation recognition task from L1 Dutch into the two foreign languages, English (L2) and French (L3). Word concreteness was varied as a marker of conceptual involvement. The expected (facilitatory) effect of concreteness was observed only for Dutch–English, but not for Dutch–French translation, leading the authors to conclude that only L1–L2 is conceptually mediated, while L1–L3 translation is not. This is consistent with the original proposal of the RHM, given that for the least proficient language (L3), conceptual links should still be lacking. However, highly asymmetric omission rates, which were especially large for L3 abstract words (up to 25 percent), complicated the analysis of the translation RTs. Furthermore, the L3 was also the language (French) that is the most distant from L1 Dutch, something that may have been at play as well.

In another test of the RHM, Reference Duyck and BrysbaertDuyck and Brysbaert (2008) had Dutch–English–German multilinguals name or translate number words in forward (L1–L2/L3) or backward (L2/L3–L1) directions. As an indicator of conceptual involvement, they used the magnitude effect (i.e., faster RTs for small than for large numbers). The RHM predicts that backward translation should be based on lexical association and thus show no magnitude effect, while forward translation should be largely conceptually mediated and thus show stronger signs of such a semantic effect. In contrast to these predictions, however, the results showed the opposite pattern (i.e., magnitude effects for backward, but not for forward translation, both for L1/L2 and for L1/L3 pairings). Together with a prior study on Dutch–French bilinguals that had shown even larger magnitude effects in both directions (Reference Duyck and BrysbaertDuyck & Brysbaert, 2004), the authors concluded that it is not language status (L1/L2/L3), but lexical similarity that matters: almost all Dutch number words are cognates with English, but not with French. Lexically similar items may be translatable with no or little conceptual involvement, whereas translation between lexically dissimilar ones may mostly take a conceptual route. However, this similarity account does not explain the asymmetry between forward and backward translation in the former study. Furthermore, as the authors note, number words are extremely highly frequent and early acquired words that may not be fully representative of the lexicon as a whole.

Further support for the relevance of lexical similarity, albeit during exposure to an entirely novel language (L0), comes from so-called cognate guessing studies. For instance, Reference Vanhove, Berthele, De Angelis, Jessner and KresicVanhove and Berthele (2015) asked participants to guess the German translation of nouns from an unknown Germanic language. They were native speakers of Swiss German with knowledge of standard German, English, and French. The stimuli were cognates with either German, English, or French, or noncognates. The results showed that translation guessing accuracy was best explained by “overall Germanic Levenshtein distance,” the similarity to the most similar German or English (but not French) translation. In other words, L1 or L2 status of the source language did not matter here, but both typological language similarity (disadvantaging French) and item-specific lexical similarity played a role. This confirms Reference Ringbom, Cenoz, Hufeisen and JessnerRingbom’s (2001: 66) claim that especially in L3 (or, like here, L0) comprehension, “wherever learners are able to perceive cross-linguistic similarities, they will make use of them.”

Masked translation priming is another method used to find out more about the mutual connections between languages in a trilingual lexicon. Reference Aparicio and LavaurAparicio and Lavaur (2016) had French–English–Spanish trilinguals perform lexical decisions on L2 and L3 targets that were preceded by masked (i.e., very briefly presented) primes (translations or unrelated) in either of the three languages. The results showed that translation primes affected RTs only when they were from the L1, pointing at only weak links between the L2 and L3. In another complex study on masked translation priming with animacy decision (an explicitly semantic task) on the target and all possible six prime-target directions, Reference TytusTytus (2017) observed robust (facilitatory) translation priming only between German (L1) and English (L2) in both directions, and some more fragile effects between L2 and L3 French that depended on translation direction and prime duration. No effects were observed between L1 and L3.

Altogether, the translation and priming literature seems too sparse and too variable in the used methods and trilingual populations to come to reliable conclusions about trilingual lexical processing that go beyond what we already observed in Section 7.4, namely that all three languages have the potential to affect each other during language processing. In particular, an especially close relationship between L2 and L3 is not consistently confirmed by this part of the literature. Instead, lexical similarity between languages (in general and per translation pair) seems to play a major role. Thus, while we have much evidence of complex asymmetries, the exact nature of the relationship between any two of a trilingual’s three languages seems to differ vastly depending on the specific trilingual population and experimental paradigm.

7.6 Summary and Conclusions

Possibly, the one and foremost thing that has become apparent in this chapter is how difficult it is to systematically study trilinguals experimentally, and how relatively rarely it has been done. Mostly, it is hard enough to find one trilingual population for a study that is sufficiently homogeneous with respect to proficiency profiles and acquisition history. However, these populations differ vastly from study to study in terms of these variables, as well as characteristics of the language triplet. Therefore, a systematic investigation would ideally require several trilingual groups differing only in the variable of interest (as in, e.g., Reference Ringbom, Cenoz, Hufeisen and JessnerRingbom, 2001; Reference van Heuven, Conklin, Coderre, Guo and Dijkstravan Heuven et al., 2011), which is however unfeasible for most researchers. Furthermore, while bilingual studies can look at maximally two directions of cross-language influence (L1→L2 and L2→L1), this number rises to six for trilinguals. If the laudable effort is made to look at all these six directions, it is often hard to derive a common pattern from the results. Despite these difficulties, it is certainly important to conduct more studies that succeed at reducing the many possible sources of undesired variability between language combinations and populations.

However, the one common pattern that we can state to have observed is that all three languages of a trilingual have the potential to interact during lexical processing in any of the trilingual’s languages. This conclusion from observational and experimental studies is also in line with findings showing proximity or overlap of the processing regions for three languages in the brain (Reference Klein, Milner, Zatorre, Visca and OlivierKlein et al., 2002; Reference Yazbek, Smayra and MallakYazbek et al., 2020), and the fact that differential recovery of the three languages seems to be the exception rather than the rule in trilingual aphasia (Reference Goral, Levy, Obler and CohenGoral et al., 2006).

Most notably, interactivity between all three languages can often (but not always) be observed even during L1 processing. This implies that, in “monolingual” language research, simply ignoring the fact that participants speak (one or more) foreign languages may cause flawed results; rather, trilinguals seem to constantly engage in exploiting similarities and avoiding interference from all their known languages, no matter which language they currently use.

8.1 Introduction

The central question in the study of transfer in L3/Ln acquisition has been the extent to which previous linguistic experience plays a role in multilingual acquisition (e.g., Reference Cenoz, Hufeisen and JessnerCenoz et al., 2001; Reference De AngelisDe Angelis, 2007; Reference De Angelis and DewaeleDe Angelis & Dewaele, 2011; Reference Rothman, Cabrelli Amaro, de Bot, Herschensohn and Young-ScholtenRothman et al., 2013; Reference Rothman and HalloranRothman & Halloran, 2013; Reference Angelovska and HahnAngelovska & Hahn, 2017). Within the subfield of formal linguistic approaches to L3/Ln acquisition, establishing what is transferred (the L_x or the L_y) into the L3 (initial stages) and what determines such selection has been the main line of inquiry. The importance of this is not trivial, as knowing what L3 initial grammars look like has important consequences for the modeling and understanding of L3 developmental trajectories (e.g., Reference Cabrelli, Iverson, Giancaspro, Halloran González, Molsing, Becker Lopes Perna and Tramunt IbañosCabrelli et al., 2020; Reference Puig-Mayenco, Rothman and TubauPuig-Mayenco et al., 2020b; Reference Cabrelli and IversonCabrelli & Iverson, in press).

Current discussions focus on the selection, timing, and nature of transfer. Selection and timing are important because we need to (1) understand which of the two linguistic systems the learner has at their disposal is transferred (or whether both or none are), and (2) establish the point at which such selection occurs. The focus of this chapter, however, is the third area of interest, the nature of such transfer: does transfer occur on a wholesale manner, à la Reference Schwartz and SprouseSchwartz and Sprouse (1996), or on a property-by-property basis? This question has sparked considerable interest recently, as evidenced in the recent keynote article by Reference Schwartz and SprouseSchwartz and Sprouse (2020) and its subsequent commentaries (e.g., Reference Falk and BardelFalk & Bardel, 2021; Reference Fernández-Berkes and FlynnFernández-Berkes & Flynn, 2021; Reference Ionin and Scutti SantosIonin & Scutti Santos, 2021; Reference LagoLago, 2021; Reference Pereira Soares and RothmanPereira Soares & Rothman, 2021; Reference WestergaardWestergaard, 2021; Reference WrembelWrembel, 2021). Before proceeding, however, a point of caution is needed. Although I will outline the proposals of morphosyntactic transfer in L3/Ln acquisition to contextualize the current debate, the purpose of the chapter is not to present or evaluate them (for a comprehensive overview of these approaches, see Reference Rothman, González Alonso and Puig-MayencoRothman et al., 2019; and Chapter 1 of this volume).

Throughout the chapter, I will provide an overview of two clusters of studies (Spanish–English → L3 Portuguese; Catalan–Spanish → L3 English) that have attempted to answer questions related to the nature of transfer. I will also discuss what the best evidence might be to inform this debate. One might say a priori that we want to test the entirety of an L3 grammar to see whether all possible influence comes from one source only or from multiple sources at the same time. I agree that this would theoretically be ideal. However, doing so would not be an easy task, even if it were feasible. What would then be the next best type of evidence? A good approximation to testing the entirety of a grammar would be to take a step back and look at all of the studies that test different domains of grammar within the same language triad. Such an approach is welcome on several planes: first, it allows us to have an overview of multiple properties that can discern whether the L_x or L_y is being transferred; second, we can increase the sample of speakers that are being examined, making the outcomes more generalizable; and third, gathered evidence comes from multiple tasks, tapping into different types of knowledge. Once all of this is considered, we get a better picture of transfer selection in a specific language triad, which, in turn, increases the generalizability of the outcomes we obtain from individual studies. In what remains, I will precisely take this approach and provide an overview of two clusters of studies: (1) English–Spanish learners of L3 PortugueseFootnote ¹ and (2) Catalan–Spanish bilingual learners of L3 English. The choice of these two clusters is not fortuitous. First, there are enough studies in each cluster that would allow us to take step back and look at what the overall picture suggests. Second, whereas the first cluster looks at a traditionally examined Romance–Germanic–Romance linguistic combination, the second cluster focuses on a combination that is studied to a lesser extent: two closely related languages (Romance–Romance) acquiring a Germanic one. Comparing these two clusters allows us to see whether wholesale transfer is pertinent to linguistic triads where either the L_x or L_y share strong structural overlap with the L3, or whether it also extends to triads where structural overlap is less straightforward. I will end the chapter by outlining some methodological suggestions for future studies that will give us a more complete picture of the nature of transfer, which will ultimately increase the comparability and generalizability of studies in the field.

8.2 Context

Several recent models to the field, the Linguistic Proximity Model (LPM; Reference Mykhaylyk, Mitrofanova, Rodina, Westergaard, Grillo and JepsonMykhaylyk et al., 2015; Reference Westergaard, Mitrofanova, Mykhaylyk and RodinaWestergaard et al., 2017; Reference WestergaardWestergaard, 2020) and the Scalpel Model of Third Language Acquisition (SM; Reference SlabakovaSlabakova, 2017), have shifted attention to the following question: does transfer occurs on a property-by-property basis, or does it occur on a wholesale basis, à la Reference Schwartz and SprouseSchwartz and Sprouse (1996)? Models of morphosyntactic transfer in L3/Ln acquisition can be classified according to whether they envision early multilingual transfer to come from one source holistically or from either of the known languages on a property-by-property basis.

In the first set of models, the entirety of one grammar (either the L_x or the L_y) is argued to constitute the (initial) mental representation of the L3.Footnote ² Importantly, the variables that will determine whether L_x or L_y is selected for transfer vary across hypotheses/models. First, we have the scenario in which the first language (L1) is the default for transfer selection. Such a scenario has never been formalized as a model per se, but some studies have indeed pointed in this direction (e.g., Reference HermasHermas, 2010, Reference Hermas2014). The other default alternative is that transfer would occur holistically from the L2; this is, indeed, the main tenet of the L2 Status Factor Hypothesis (L2SF; Reference Bardel and FalkBardel & Falk, 2007, Reference Bardel, Falk, Cabrelli Amaro, Flynn and Rothman2012; Reference Bardel, Sánchez, Angelovska and HahnBardel & Sánchez, 2017; Reference Falk and BardelFalk & Bardel, 2011).Footnote ³ In the most recent instantiations of this hypothesis, Reference Bardel, Sánchez, Angelovska and HahnBardel and Sánchez (2017) suggested a number of variables that could potentially lead the parser to trump L2 transfer based on other factors (e.g., working memory capacity or metalinguistic knowledge). Whereas these scenarios default transfer selection to either the L1 or the L2, there is an additional subset of models that do not default the selection to order of acquisition. The Typological Primacy Model (TPM; Reference RothmanRothman, 2010, Reference Rothman2011, Reference Rothman2015) maintains that the parser evaluates structural proximity between the target L3 input the learner receives in the initial stages against the L_x and L_y. Quite early on, upon having had enough linguistic cue validity to make an informed selection, the parser determines which whole system to transfer (Reference RothmanRothman, 2015; Reference González Alonso and RothmanGonzález Alonso & Rothman, 2017). This evaluation of the input occurs using a hierarchy of cues that allow the parser to determine which of the two systems is, overall, structurally more similar (i.e., which of the two systems has more structural overlap as a whole) to the new input that is being parsed (for a more detailed explanation of this mechanism, see Reference RothmanRothman, 2015; Reference Rothman, González Alonso and Puig-MayencoRothman et al., 2019).Footnote ⁴ More recently, a new proposal has suggested that the primary language of communication (Reference Fallah, Jabbari and FazilatfarFallah et al., 2016; Reference Fallah and JabbariFallah & Jabbari, 2018; Reference 215Jabbari, Achard-Bayle and AblaliJabbari et al., 2018) is the determining variable in transfer selection. This proposal suggests that language dominance is key in transfer selection.

According to the second set of models, transfer occurs selectively throughout L3 development and thus unfolds on a property-by-property basis. The Cumulative Enhancement Model (CEM; Reference Flynn, Foley and VinnitskayaFlynn et al., 2004; Reference Berkes, Flynn, Cabrelli Amaro, Flynn and RothmanBerkes & Flynn, 2012) argues that maximal facilitation is the determining factor; in other words, transfer only occurs when it will boost the acquisition of a given property in the L3.Footnote ⁵ The Linguistic Proximity Model (LPM; Reference Westergaard, Mitrofanova, Mykhaylyk and RodinaWestergaard et al., 2017), on the other hand, rejects this tenet. Instead, it suggests that that transfer selection is determined by the comparative linguistic structural proximity for each property individually of the L3 to the properties in the other systems. Moreover, the LPM offers the possibility that both L_x and L_y could have a simultaneous influence on the L3 in what would look like hybrid transfer. The Scalpel Model (SM; Reference SlabakovaSlabakova, 2017) aligns with the LPM in that it also proposes that transfer occurs selectively throughout development. However, it adds a series of factors that might obscure the straightforward application of the linguistic proximity principle (e.g., processing complexity and construction frequency).

8.3 Review of Two Study Clusters

8.4 Methodological Caveats and Recommendations for Future Studies

In this section, I elaborate on some of methodological considerations I have touched upon in this chapter: timing, properties examined, and type of data. Only by controlling these can we be confident that the methodological protocols will allow us to tease apart transfer from other potential effects.

8.5 Conclusions and Future Directions

The aim of this chapter was to explore some of the existing literature in relation to the nature of transfer in L3 acquisition. Although there is not yet a clear consensus within the field, we can now take a step back, aggregate the results of different clusters of studies, and see what the overall picture suggests. Taken together, the studies on the acquisition of Portuguese as an L3 by Spanish–English bilinguals suggest that, while some methodological choices might lead to mixed results, the overall picture indicates that Spanish is systematically found to be the transfer source across studies that have similar methodologies and similar protocols. However, and while this pattern seems to hold for learners of different proficiency levels, I argue that the best way to tease apart transfer effects from L3 learning processes is to look at beginner learners with low exposure. In these cases, the picture is even stronger: 100 percent of the studies in this language triad show Spanish transfer. I also presented an overview of a subset of studies on Catalan–Spanish bilinguals acquiring English as an L3. The results from this subset also point toward systematic transfer from one source, especially when ab initio learners are examined. Crucially, this second subset shows that wholesale transfer might also take place even in situations where the overall structural overlap is less straightforward than in a Romance–Germanic–Romance linguistic triad.

I acknowledge that the chapter does not provide definitive evidence for wholesale transfer because it only presents evidence from two clusters of studies. It is obvious that the field needs more studies that look at similar populations across different properties before we draw more meaningful conclusion as to what variables, if any, condition transfer selection in a wholesale or property-by-property manner. Importantly, the chapter also shows that only by adopting consistent and strict protocols across studies will we be able to adequately provide a fuller picture of the nature of transfer (and its timing and source). Ultimately, what we need to agree on is our object of inquiry. Are we interested in the nature of transfer at the initial representation? Or throughout development? Or in the interaction of language representations with other cognitive mechanisms? All of these questions are equally interesting, but not all designs allow us to equally answer each of these questions.

9.1 Introduction

This chapter discusses L3 acquisition as a step-by-step acquisition process. From the perspective of two relatively new models in the field, the Linguistic Proximity Model (Reference Westergaard, Mitrofanova, Mykhaylyk and RodinaWestergaard et al., 2017; Reference WestergaardWestergaard, 2021a, b) and the Scalpel Model (Reference SlabakovaSlabakova, 2017), we focus on theoretical arguments for crosslinguistic influence taking place property by property from either or both previously acquired language(s) and discuss methodological issues for studies that aim to identify hybrid influence. We also provide a brief overview of empirical studies finding support for this position.

9.2 Background

The formal study of L3/Ln acquisition is a field that has developed rapidly over the last couple of decades, informed by previous work on L2 acquisition (SLA). Especially important is the discussion about the initial state of L2 acquisition from the 1990s, where one of the main issues concerned L1 transfer, more specifically how much would be transferred from the L1 into the L2 – everything, nothing, or something in between. Two important models from this time are Full Transfer/Full Access (Reference Schwartz and SprouseSchwartz & Sprouse, 1996), arguing that the initial state of L2 acquisition is the complete L1 grammar, and Minimal Trees (Reference Vainikka and Young-ScholtenVainikka & Young-Scholten, 1996), arguing that only parts of the L1 grammar are available for transfer, notably lexical elements but no functional elements above the verb phrase (VP).Footnote ¹ Much of this thinking has been extended to the field of L3/Ln acquisition. This is clearly visible in the work of Reference Leung, Greenhill, Hughes, Littlefield and WalshLeung (1998, Reference Leung and Liceras2003), who argues for a direct extension of the Full Transfer model to L3 acquisition, later referred to as the Interlanguage Transfer Hypothesis (Reference Jin and LeungJin, 2009), which, however, was quickly abandoned already in Reference LeungLeung (2005). A revision of these ideas has later appeared as the Typological Primacy Model (TPM), where a number of modifications have been made in order to maintain the idea of full (wholesale) transfer (e.g., a change of focus from the initial state to initial stages), allowing some time for the parser to make a decision about which of the two previously acquired grammars to copy (Reference RothmanRothman, 2015). Other approaches to L3 acquisition have taken a different perspective (e.g., the Cumulative Enhancement Model [CEM]), focusing on the cumulative and facilitative nature of L3 acquisition (Reference Flynn, Foley, Vinnitskaya, Archer and RederFlynn et al., 2001, Reference Flynn, Foley and Vinnitskaya2004), or the L2 Status factor (L2SF), emphasizing the importance of the order of acquisition of the three languages involved and arguing that the L2 will be the main source of crosslinguistic influence, especially at early stages (Reference Bardel and FalkBardel & Falk, 2007, Reference Bardel and Falk2012).

In this chapter, we concentrate on two recently proposed L3 acquisition models, the Scalpel Model (Reference SlabakovaSlabakova, 2017) and the Linguistic Proximity Model (LPM; Reference Westergaard, Mitrofanova, Mykhaylyk and RodinaWestergaard et al., 2017; Reference WestergaardWestergaard, 2021a, b). These models stand on the shoulders of previous approaches, more specifically the TPM and the CEM, in that both argue for the importance of structural/typological similarity for crosslinguistic influence (like the TPM) and view L3 acquisition as a cumulative process (like the CEM). The LPM and the Scalpel Model explicitly define property-by-property influence as the major distinguishing property of the models. Importantly, property-by-property development in L3 acquisition is different from the L2 partial transfer models of the 1990s, in that any linguistic property is argued to be able to transfer. However, this is also different from Full Transfer (as wholesale transfer), in that crosslinguistic influence is not assumed to take place in one fell swoop. Our position is thus called Full Transfer Potential (FTP), meaning that anything can transfer, not that everything does transfer. Here we focus on the two models’ answers to several important research questions in the field, what constitutes the initial state/stages of L3 acquisition, and how building the L3 grammar proceeds.

9.3 The LPM and the Scalpel Model

9.4 Methodology for Research on Property-by-Property Influence

9.5 Empirical Support for Property-by-Property Influence

In this section, we provide a noncomprehensive overview of empirical studies that support the idea of property-by-property transfer from both previously acquired languages, mainly focusing on relatively recent work. During the first two decades of formal investigations of L3 acquisition, the research focus was mainly on which of the two previously acquired languages would provide most of the influence (and in the extreme case of the TPM, all of the influence at initial stages), in order to identify which factors play the most important role, whether it be, for example, order of acquisition (Reference Bardel and FalkBardel & Falk, 2007, Reference Bardel and Falk2012), typological/structural similarity (Reference RothmanRothman, 2011, Reference Rothman2015; Reference Westergaard, Mitrofanova, Mykhaylyk and RodinaWestergaard et al., 2017), or language dominance (Reference Fallah, Jabbari and FazilatfarFallah & Akbar Jabbari, 2016; Reference Lloyd-Smith, Gyllstad, Kupisch and QuagliaLloyd-Smith et al., 2018; Reference Puig-Mayenco, Miller, Rothman, Cho, Iverson, Judy, Leal and ShimanskayaPuig-Mayenco et al., 2018; Reference Angelovska, Roehm and WeinmüllerAngelovska et al., 2020). Thus, most studies test very few linguistic properties; in fact, often only one. In such cases it is of course difficult to find CLI from both languages. Nevertheless, in Reference Rothman, González Alonso and Puig-MayencoRothman and colleagues’ (2019) systematic review of as many as ninety-two studies, eighteen of them display hybrid influence from both previously acquired languages, either on the same or on different linguistic properties. This means that there is no lack of studies showing some kind of influence from two different languages, thus by definition also property-by-property transfer. As noted in Section 9.3.1, proponents of wholesale transfer would explain such data by making a distinction between transfer and CLI/CLE, so that influence from one language is the result of wholesale representational copying at the initial stages, and influence from the other is the result of transient processing issues – a distinction that is not made by property-by-property approaches.

In the original article introducing the LPM (Reference Westergaard, Mitrofanova, Mykhaylyk and RodinaWestergaard et al., 2017), the study compares Russian–Norwegian learners of English as an L3 (n = 22) with two groups of L2 learners, one with L1 Norwegian and the other with L1 Russian. The study investigates two word-order phenomena, one where English is similar to Russian and one where English is similar to Norwegian. While one of the properties was already acquired by all learners (S-aux inversion), the other (Adv–V/V–Adv word order) showed that the L3 learners scored between the two L2 groups, indicating that they were experiencing influence from both previously acquired languages. In a follow-up study, Reference Kolb, Mitrofanova and WestergaardKolb and colleagues (2022) added two further properties (Determiner use and V2 word order in non-subject-initial declaratives) in an investigation of Russian–German learners of L3 English (n = 66), comparing them with corresponding L2 groups. The results showed that the L3 learners were significantly different from both L2 groups in three of the four conditions (V2 word order, Adverb placement, and Determiner use), scoring between the two groups, while there was no significant difference between the three groups for S-aux inversion, as in Reference Westergaard, Mitrofanova, Mykhaylyk and RodinaWestergaard and colleagues’ (2017) study. Importantly, the two L2 groups showed a higher-than-average sensitivity on the condition(s) that was/were similar to their L1 and a correspondingly lower sensitivity on the one(s) that was/were different from their L1.

One theoretically plausible explanation of the observed in-between performance of the L3 group could be that the mean is a result of two distinct distributions (half of the L3 participants’ behavior comparable to the Language A group and the other half patterning like the Language B group). Therefore, it is important to consider a by-participant analysis and individual deviations from the group mean. Figure 9.1 plots individual results from the three participant groups on two of the conditions, Determiner use and Adverb placement. While the L1 German speakers’ individual scores are above average on Determiner use and below average on Adverb placement (the light grey frame) and the L1 Russian speakers display the opposite pattern (the dark grey frame), the L3 learners’ individual performance (the triangles) is generally found in the middle. This illustrates that the L3 group mean is not a result of a bimodal distribution, but genuinely reflects CLI from both previously acquired languages. In other words, we can argue for cumulative CLI at both group and individual levels.

Figure 9.1 Distribution of random effect sizes for two critical conditions (Determiner use and Adverb placement), showing that Russian–German L3 learners of English are different from both of the corresponding L2 groups.

(Reference Kolb, Mitrofanova and WestergaardKolb et al., 2022)

Other studies have attested similar phenomena. Reference Stadt, Hulk and SleemanStadt and colleagues (2016, Reference Stadt, Hulk and Sleeman2018a, Reference Stadt, Hulk, Sleeman, Berns, Jacobs and Nouveaub, Reference Stadt, Hulk, Sleeman, Trotzke and Kupisch2020) conducted a series of studies of word order in L3 French and L3 German with speakers of L1 Dutch–L2 English. In a longitudinal study, Reference Stadt, Hulk, Sleeman, Berns, Jacobs and NouveauStadt and colleagues (2018b) considered early and later acquisition stages of subject-initial and non-subject-initial declaratives in L1 Dutch–L2 English–L3 French (year 1–2 of French instruction). The three languages pattern in the following way with respect to word order: Dutch and French display similarity in subject-initial declaratives with sentence adverbials (i.e., V-Adv (V2) word order), while French is similar to English in non-subject-initial declaratives, where both exhibit XP-S-V (non-V2) word order. The results from first-year students with 50 percent English immersion demonstrate a strong influence from Dutch, as V2 is used in both structures, target-consistently in subject-initial and nontarget-consistently in non-subject-initial declaratives. However, the ungrammatical non-V2 word order (i.e., the English Adv-V) is accepted at a rate of 14.6 and 34.7 percent in a Gap Filling and Grammaticality Judgment Test respectively. Furthermore, non-V2 is used appropriately at approximately the same rates (20.1 and 44.1 percent) in non-subject-initial declaratives. These results indicate that both previously acquired languages are activated and available for CLI at early stages of L3 acquisition.

Additional factors are brought into the picture in a later comparison of the acquisition of V-Adv word order by intermediate learners of L3 German and L3 French in order to explain a higher occurrence of (nontarget-consistent) Adv-V word order in the latter group (Reference Stadt, Hulk, Sleeman, Trotzke and KupischStadt et al., 2020). It is suggested that the substantial activation of L2 English happens only when learners are sufficiently exposed to the L2 on a daily basis. In the case of L1 Dutch–L2 English–L3 German, the high frequency of V2 structures in the L1 and L3 increases learner awareness of the structural resemblances between these two languages at later stages, when sufficient exposure to the L3 is reached. This is when L1 Dutch becomes the more suitable language for transfer in L3 German. Thus, a multiple group methodology (e.g., with the same L1 and L2 and different L3s) and cross-sectional designs can provide important insights into CLI at the initial stages and beyond.

In a similar study, Reference Dahl, Listhaug, Busterud, Leal, Shimanskaya and IsabelliDahl and colleagues (2022) examine the acquisition of subject-initial declaratives with sentence adverbials in L1 Norwegian–L2 English–L3 German in an Acceptability Judgement Test, comparing this property to the acquisition of word order in non-subject-initial declaratives across four age groups (years 1–4 of German instruction). For both phenomena, there is structural similarity between L1 Norwegian and L3 German in the form of V2 word order (V-Adv and XP-V-S). In the earliest learners (years 1–2 of German instruction), the authors find no clear evidence of transfer from either L1 Norwegian or L2 English, since the mean scores on both structures show no preference for either word order. The same is observed in the individual data. According to the authors, this indicates insecurity rather than transfer from one of the previously acquired languages, which could be “due to the potential availability for transfer of structures from two previously learned languages, e.g., along the lines of full transfer potential” (191). Interestingly, the lack of a word order preference in L3 German in early years occurs despite the participants’ near target-like performance on both sentence types in L2 English (across all age groups). Thus, both word orders seem to compete in an emergent L3 grammar, suggesting that both previously acquired languages are activated. In years 4 and 5, a development toward target-like word order is observed, which falls into place earlier in subject-initial than non-subject-initial declaratives.

Not many L3 studies investigate more than one or two properties. Reference Ben AbbesBen Abbes (2020) is one of the few studies designed to investigate several morphological and syntactic properties, including gender, number, and articles (definiteness/specificity) in two groups of beginner learners of L3 French with L1 Spanish–L2 English and L1 Turkish–L2 English. Despite this multiple-property design, only one structure presents a good test case, the definiteness/specificity distinction in the L1 Turkish–L2 English group, where L3 French is similar to English and different from Turkish. In this case, the study is able to isolate facilitative evidence from L2 English, since L1 Turkish learners reach target-like knowledge of the indefinite article in L3 French only when they have attained advanced proficiency in L2 English. These results suggest that both previously acquired languages are available for CLI in L3 acquisition, and that structural similarity with the L3 determines the property to be transferred, with proficiency as a confounding factor. The results from the L1 Spanish–L2 English language pair reveal no effects for any of the properties, since the participants perform at ceiling in L3 French. From the perspective of the LPM and the Scalpel Model, the results from L1 Turkish–L2 English–L3 French demonstrate that even a single group methodology can be revealing of incremental grammar construction in L3 acquisition.

More substantial evidence in support of the LPM and the Scalpel Model is reported by Reference Jensen, Mitrofanova and AnderssenJensen and colleagues (2021) from the acquisition of L3 English by Norwegian–Russian bilinguals across three linguistic modules: syntax (word order), morphology (subject-verb agreement and copula), and the syntax-semantics interface (genericity and definiteness). In each domain, at least one condition targeted a property that was similar in Norwegian and English (while Russian was different) or similar in Russian and English (while Norwegian was different). The comparison of the bilinguals with two groups of monolingual controls (L1 Russian and L1 Norwegian) follows two scenarios: (1) the L3 learners pattern with the higher-accuracy L2 group (i.e., the L3-ers have learned to inhibit the non-facilitative language); or (2) they score in between the L2 groups (as in Figure 9.1). This is illustrated in Figure 9.2, which demonstrates simultaneous facilitative and nonfacilitative influence from both previously acquired languages on subject-verb agreement. Additionally, the study highlights the importance of other factors such as complexity and saliency, which account for different developmental slopes across investigated properties.

Figure 9.2 Distribution of individual accuracy scores on subject-verb agreement by group: Bilingual (2L1 Norwegian/Russian–L3 English), Norwegian (L1 Norwegian–L2 English), Russian (L1 Russian–L2 English).

(Reference Jensen, Mitrofanova and AnderssenJensen et al., 2021: 13)

To conclude, the selected evidence reviewed in this section is in line with the LPM and Scalpel Model’s proposal that an L3 grammar is built incrementally in small steps and that it is influenced by both previously acquired languages. While subtractive group and multiple property methodology is particularly useful in isolating the influence from previously acquired languages, cross-sectional and especially longitudinal designs present great potential, especially in the absence of control groups. Furthermore, in order to address relevant theoretical underpinnings, the study design must involve a high degree of sophistication, especially with regard to the selection of test properties, their linguistic characteristics, and acquisition trajectories. The relevance of these considerations is based on evidence from, for example, Reference Slabakova and García MayoSlabakova and García Mayo (2015), Reference Guo and YuanGuo and Yuan (2020), as well as Reference Feijoo and García MayoFeijoo and García Mayo (2021), where a set of factors is hypothesized to modulate CLI in important ways.

9.6 Summary and Outlook

In this chapter we have discussed crosslinguistic influence in L3 acquisition from the perspective of the LPM and the Scalpel Model, arguing for property-by-property acquisition. According to these models, the initial state of L3 acquisition is the entirety of the two previously acquired languages, which are always active in the L3 acquisition process. Like L1 and L2 acquisition, L3 acquisition is learning by parsing/processing, and the L3 grammar is built up incrementally in small steps. In this process, the full grammars of both previously acquired languages remain available for CLI, which we refer to as Full Transfer Potential. During processing of the L3, the previously acquired languages will be activated to different degrees, based on structural similarity with the L3 – and at early stages also superficial lexical/typological similarity. We have also devoted a section to the optimal methodology for L3 research, focusing on the type of populations to compare, the combination of properties to investigate, and the timing of testing. For the LPM and the Scalpel Model, it is important (1) to use a subtractive language group design in order to isolate the effect of each previously acquired language; (2) to select properties that share similarities and differences across the three languages; and (3) to choose an acquisition stage that avoids ceiling and floor effects. Additionally, in line with the Scalpel Model, it is necessary to investigate (and/or control for) additional factors such as frequency, proficiency, recency, and so on. The field of multilingual language acquisition is only in its infancy, and we believe we will see considerable development in this area in the coming years.

10.1 Introduction

Recently, there has been an upsurge in research on how the experience of acquiring and using two (bilingualism) or more languages (multilingualism) affects language and nonverbal cognition (e.g., Reference BialystokBialystok, 2017; Reference Cenoz, Barron, Gu and SteenCenoz, 2017; Reference Wright, Boun and GarcíaWright et al., 2017; Reference de Houwer and Ortegade Houwer & Ortega, 2018). This growing research interest has developed in response to an increasing awareness that the majority of today’s world population is bi- or multilingual (e.g., Reference Cenoz, Soler and JordàCenoz, 2008; Reference Grosjean and LiGrosjean & Li, 2013). In this chapter, we review the research on multilingualism that focused on a particular aspect of cognitive functioning; that is, pragmatic performance, which broadly refers to the effective production and interpretation of meaning in context (Reference TaguchiTaguchi, 2009). We take a wide view of bi- and multilingualism as subsuming not only individuals who started using two or more languages in childhood, but also second language (L2) or third language (L3) learners, who started using their additional language(s) later in life.

Bilingualism and multilingualism have a lot in common and the two terms are often used interchangeably, but the multilingual experience is different from that of bilinguals in various respects (Reference Cenoz, Soler and JordàCenoz, 2008, Reference Cenoz2013; Reference Safont-Jordà and ChappelleSafont-Jordà, 2013a; Reference Rothman, González Alonso and Puig-MayencoRothman et al., 2019). One first main difference lies in crosslinguistic influence, the effect of one language on another at the competence and/or performance level (Reference Rothman, González Alonso and Puig-MayencoRothman et al., 2019); specifically, in multilingualism, crosslinguistic influence can occur between multiple rather than between only two languages (e.g., from the first language [L1] to the L3, from the L3 to both L1 and L2, and so on). Second, multilingualism presents more variability than bilingualism. In terms of age of acquisition, for example, bilingualism involves only two possibilities – simultaneous development of two languages from birth or sequential acquisition of an L2 sometime after birth – but in multilingualism there are four; that is, all three languages can be acquired simultaneously from birth or sequentially (L1, then L2, then L3) or two languages might be learned simultaneously before the L3 (L1 and L2 from birth, then L3) or after the L1 (L1 from birth, then L1 and L2). Levels of exposure to and use of each language are also different in multilinguals compared to bilinguals and monolinguals, because language experience is divided across more languages. This suggests that multilinguals might exhibit performance in each language that differs in some respects from bilinguals or monolinguals. For instance, bilinguals are often reported to exhibit a slower developmental rate in some aspects of language or slower lexical access than monolinguals, when each of their languages is considered separately (e.g., Reference Bialystok, Craik, Green and GollanBialystok et al., 2009). This finding has been attributed to the different amount and quality of experience that bilinguals have in each of their languages compared to monolinguals (e.g., Reference Bialystok, Craik, Green and GollanBialystok et al., 2009). Third, learning an L3 as a bilingual is distinct from acquiring an L2 as a monolingual, in that bilinguals have already been through the process of learning an additional language. In this regard, bilinguals are more experienced language learners and they have likely developed language learning skills and strategies, which they can use in the task of acquiring an L3 (e.g., Reference CenozCenoz, 2013; Reference Rothman, González Alonso and Puig-MayencoRothman et al., 2019). Finally, bilinguals learning an L3 can bring to the L3 acquisition process an altered, compared to monolinguals, set of general cognitive skills that can affect aspects of language learning. For instance, even though the evidence is still mixed, bilingualism has been associated with (1) enhanced metalinguistic awareness, the ability to explicitly reflect on language and manipulate it (e.g., Reference Adesope, Lavin, Thompson and UngerleiderAdesope et al., 2010); (2) better executive functions, a set of cognitive processes including working memory (the ability to hold and use at the same time information in memory), inhibition (the skill to ignore nonrelevant information), and switching (the ability to seamlessly shift attention from one task to another; e.g., Reference BialystokBialystok, 2017); and (3) advanced Theory of Mind, a cognitive system that is responsible for attributing mental states, such as intentions and feelings, to oneself and to others, and for understanding behavior by considering these mental states (e.g., Reference SchroederSchroeder, 2018). Moreover, some evidence suggests that bilingual children exhibit more sensitivity to the speakers’ communicative needs (Reference Genesee, Tucker and LambertGenesee et al., 1975), a better-developed ability to detect communicatively infelicitous utterances (Reference Siegal, Iozzi and SurianSiegal et al., 2009) and to use pragmatic cues such as the interlocutor’s perspective or gaze direction for inferring their referential intention (e.g., Reference Yow and MarkmanYow & Markman, 2011b; Reference Fan, Liberman, Keysar and KinzlerFan et al., 2015); and an increased reliance on pragmatic, relative to purely linguistic, cues when processing or acquiring language (e.g., Reference Yow and MarkmanYow & Markman, 2011a). All these cognitive skills and linguistic strategies may affect general facets of language acquisition and processing, as well as pragmatic development and performance in particular (see, e.g., Reference Antoniou and TaguchiAntoniou, 2019; Reference Antoniou and MilakiAntoniou & Milaki, 2021).

In the next sections, we provide some background on pragmatics and then move on to research that has focused on specific aspects of pragmatic performance in multilinguals. The bulk of research on multilingualism and pragmatics has examined speech acts, but we also consider studies that have investigated other pragmatic phenomena, such as using the appropriate language with different interlocutors in the early years of child multilingual acquisition; and the comprehension of pragmatically implied meanings (implicatures) in children and adults. Moreover, this research has examined pragmatic performance using traditional behavioral measures, including child speech samples during conversations (e.g., Reference MontanariMontanari, 2008; Reference Safont-JordàSafont-Jordà, 2011), response rates to questionnaires (e.g., Reference Tiv, Rouillard, Vingron, Wiebe and TitoneTiv et al., 2019), rates of target responses in production tasks (e.g., Reference Safont-JordàSafont-Jordà, 2003, Reference Safont-Jordà2005a; Reference Zand-Moghadam and AdehZand-Moghadam & Adeh, 2020), appropriateness ratings of target expressions (e.g., Reference Safont-JordàSafont-Jordà, 2003; Reference Portolés-FalomirPortolés-Falomir, 2015), and rates of correct responses or manual response times to target utterances in comprehension tests (Reference Antoniou and KatsosAntoniou & Katsos, 2007; Reference Antoniou, Culbertson, Perfors, Rabagliati and RamenzoniAntoniou, 2022). Newer psycho- and neurolinguistic methods, such as eye-tracking and Event-Related Potentials, have not, to date, appeared in this field. We return to this issue briefly in the final section of this chapter, where we discuss directions for moving research in this area forward. This is not intended to be an exhaustive review of all research on multilingual pragmatics. Rather, we selectively focus on key studies and topics in this body of work to present central lines of research and main findings, as well as to identify avenues for future research.

10.2 Pragmatic Performance, Speech Acts, and Implicature

Pragmatic performance includes a wide array of skills related to the use of context – the characteristics of the speaker and listener, aspects of the physical setting, or other shared information between interlocutors – alongside explicit language, to convey and comprehend intentions or intended meanings (e.g., Reference GriceGrice, 1989).

In this regard, pragmatic performance encompasses the ability to perform speech acts; that is, to produce utterances that communicate a specific intended meaning and have a certain effect on the interlocutor, such as requests, refusals, apologies, or compliments (Reference AustinAustin, 1962; Reference SearleSearle, 1969). Speech acts are potentially face-threatening communicative acts, in that they carry the risk of damaging the face, that is, the public self-image, of the speaker or the addressee (e.g., Reference Brown and LevinsonBrown & Levinson, 1987). They can be formulated with linguistic strategies varying from direct to indirect and with additional linguistic items to reduce their negative, face-threatening force. The choice of strategy and the use of linguistic mitigation items depend on contextual factors and, specifically, on politeness considerations, such as the power relation or degree of familiarity between the interlocutors (e.g., Reference Alcón-Soler, Safont-Jordà and Martínez-FlorAlcón-Soler et al., 2005). For example, requests such as in (1) are a specific type of speech act used by the speaker to get the listener to perform an action for the speaker’s benefit (see Reference Portolés-FalomirPortolés-Falomir, 2015: 48–53). Since requests may threaten the addressee’s freedom (Reference Brown and LevinsonBrown & Levinson, 1987), they can be mitigated with linguistic items such as please. More specifically, requests may be comprised of two parts. First, the head act, which is the main part of the utterance performing the function of requesting and, second, the peripheral linguistic devices, which are optional and can be used to soften, mitigate, or aggravate the face-threatening nature of the request. The head act might be realized through a direct (e.g., Close the window, please), conventionally indirect (e.g., Would you please close the window? I do appreciate it), or an indirect (It’s really cold in here, isn’t it?) strategy. Also, the peripheral linguistic items might be internal, in that they appear within the head act, or external, in that they appear either before or after the core act. Indirect strategies and mitigation items might be used in high face-threatening contexts – such as when the interlocutor is of higher social status relative to the speaker – because they make the utterance more polite (e.g., Reference Portolés-FalomirPortolés-Falomir, 2015). Moreover, because of their structural complexity, conventionally indirect strategies and the use of modification items appear more consistently later in development during L1 and L2 acquisition and depend on proficiency level in the target language (e.g., Reference Portolés-FalomirPortolés-Falomir, 2015). Example 1 illustrates a conventionally indirect request and its components.

Pragmatic performance further entails the ability to communicate and interpret implied meanings. Conversational implicature is a term introduced by Reference GriceGrice (1989) to describe cases in communication in which the speaker intentionally conveys information that is not part of what they explicitly said, and the listener understands this implied speaker’s meaning through an inference-making process about the speaker’s intentions behind the utterance. According to Grice, this reasoning process is triggered by the assumption that the speaker, in producing an utterance, has been cooperative and has adhered to certain conversational rules, which he calls maxims. These maxims exhort speakers to be relevant; to give no less and no more information than is needed for the purpose of the talk exchange; to be brief, orderly, and avoid ambiguity and obscurity; and to provide information that is true and for which there is adequate evidence. These correspond to the maxim of relevance, the first and second maxims of quantity, and the maxim of manner and of quality, respectively. For instance, consider the following dialogue in the context of a scenario where B returns home looking very disappointed after a very bad job interview.

1. (2) A: How did the interview go?

   B: It went fantastic!

2. (3) The interview went terribly.

B’s utterance in (2), possibly accompanied by further cues such as a distinctive intonation, will be understood as implying (3), even though this has not been stated explicitly. Within the Gricean account, the implied proposition is an implicature – specifically, an ironic interpretation – that is derived by exploiting the maxim of quality: It is obvious to A that B said something they do not believe to be true; however, since B is cooperative, this means that they are trying to get across a different, truthful proposition. The most obviously related truthful proposition is the contradictory to what B explicitly said.

Reference GriceGrice (1989) approached communication and conversational implicatures from a philosophical perspective. However, recent pragmatic accounts have developed Grice’s ideas into cognitive-psychological models of language interpretation, production, and communication (e.g., Reference Sperber and WilsonSperber & Wilson, 2002); specifically, Relevance Theory has made two important claims about pragmatic comprehension: (1) that it requires cognitive effort, which has been interpreted by some researchers as longer processing time (e.g., Reference Bott and NoveckBott & Noveck, 2004) and/or the use of additional cognitive resources (e.g., Reference de Neys and Schaekende Neys & Schaeken, 2007) when interpreting implicatures relative to literal meanings; and (2) that, since pragmatic comprehension involves understanding the speaker’s intentions in producing an utterance, it further draws on Theory of Mind skills (Reference Sperber and WilsonSperber & Wilson, 2002).

In this context, the time course of and the cognitive factors that underpin pragmatic interpretation have been the subject of ongoing debate. Much work on the time course of pragmatic understanding has focused on a specific implicature type, scalar implicatures, such as when the use of the scalar term some implies not all. This research has provided some evidence that scalar implicatures (SI) are associated with a cost in terms of processing time, using a range of measures, such as manual reaction times from sentence judgment tasks (e.g., Reference Bott and NoveckBott & Noveck, 2004), reading times from the self-paced reading paradigm (e.g., Reference Breheny, Katsos and WilliamsBreheny et al., 2006), mouse-tracking (Reference Tomlinson Jr, Bailey and BottTomlinson et al., 2013), and eye movements in the visual world paradigm (e.g., Reference Huang and SnedekerHuang & Snedeker, 2009). However, other studies with similar methods, specifically, eye tracking and self-paced reading, indicate that, in some contexts, SI can be accessed as fast as literal meanings (e.g., see Reference Breheny, Katsos and CumminsBreheny, 2019). A similar picture emerges for other pragmatic phenomena, including metaphors and irony (see e.g., Reference Gibbs and ColstonGibbs & Colston, 2012: 58–127). Finally, research on the cognitive underpinnings of pragmatic comprehension in children and adults has revealed effects of working memory (e.g., Reference Chiappe and ChiappeChiappe, & Chiappe, 2007; Reference Marty and ChemlaMarty & Chemla, 2013; Reference Antoniou, Veenstra, Kissine and KatsosAntoniou et al., 2020; Reference Antoniou and MilakiAntoniou & Milaki, 2021) and Theory of Mind (e.g., Reference Filippova and MatthewsFilippova, 2014; Reference Spotorno and NoveckSpotorno & Noveck, 2014; Reference Fairchild and PapafragouFairchild & Papafragou, 2021), even though, again, these results are not always reported (see, e.g., Reference Matthews, Biney and Abbot-SmithMatthews et al., 2018). Overall, the mixed findings suggest that pragmatic interpretation is in some cases costly, but that various linguistic and extralinguistic cues (e.g., certain linguistic expressions, contextual support) may combine to increase or decrease processing effort (e.g., Reference Gibbs and ColstonGibbs & Colston, 2012: 58–127).

10.3 Multilingualism and Pragmatic Differentiation

An important aspect of bilingual and multilingual children’s pragmatic competence is the skill to pragmatically differentiate their languages; that is, to use the appropriate language with speakers of different languages. Bilingual children can adjust their language use based on their interlocutor’s language by around the age of two (e.g., Reference Genesee, Nicoladis and ParadisGenesee et al., 1995; Reference Comeau, Genesee and LapaquetteComeau et al., 2003; Reference Genesee, Nicoladis, Hoff and ShatzGenesee & Nicoladis, 2007). Reference MontanariMontanari (2008) and Reference QuayQuay (2008) have shown that similar findings hold for simultaneous trilingual children.

The two studies examined children’s spontaneous speech in simultaneous interactions with speakers of different languages from the age of 1;10 until 2;4. In both studies, there was evidence that children pragmatically differentiated their languages. In Reference MontanariMontanari’s (2008) study, for instance, the child, Kathryn, modified her language use with each of her interlocutors; specifically, she used the most Tagalog with her Tagalog-speaking mother, the most English with her English-speaking interlocutor, and the most Spanish with her Spanish-speaking relatives. Similarly, Xiaoxiao, the child in Reference QuayQuay’s (2008) study, used more Mandarin than her other languages when addressing her mother, who was a Mandarin native speaker, and more English with her father, a native speaker of English. Furthermore, when addressing both parents, Xiaoxiao predominantly relied on Japanese. Reference QuayQuay (2008) attributed this preference to the fact that Japanese was Xiaoxiao’s dominant language, as well as to the child’s awareness that this was a language shared by her parents. Finally, for Xiaxiao, further evidence of context sensitivity was found when comparing her language production at home and at her monolingual Japanese-speaking daycare; specifically, the child almost exclusively used Japanese at daycare despite using all her languages at home.

For both children, there were also instances of language mixing; that is, use of elements from more than one language in a context where only one language was appropriate. However, even these cases were justifiable and indicated pragmatic sensitivity to the speaker. Reference MontanariMontanari (2008) argued that Kathryn’s inappropriate language choices were driven by three factors: (1) her proficiency in each language, as she often relied on her dominant language or resorted to an inappropriate language to fill lexical gaps in the relevant language; (2) her interlocutors’ attitude toward language mixing; and (3) her awareness that her interlocutors were multilinguals. Kathryn’s inappropriate language use was particularly evident with her father and paternal grandmother, who were more accommodating of her language mixing. Moreover, her mother, father, and paternal grandmother often used English – their non-native language – to communicate with each other, hence providing the child with cues that they were multilinguals. These factors rendered language mixing an appropriate and effective communication strategy with these interlocutors.

Xiaoxiao’s mixing behavior also provided evidence for pragmatic differentiation. Xiaoxiao mainly mixed English and Japanese when addressing her English–Japanese bilingual father, but used each of these languages in dual combinations with Mandarin when addressing her Mandarin-speaking trilingual mother. In addition, her mixed utterances to her trilingual mother included more triadic combinations of languages than to her bilingual father. These findings indicate that the child was sensitive to her parents’ language proficiencies, even in her mixing.

Thus, evidence suggests that multilingual children can adjust their language use to their interlocutors and context. Language mixing does occur in their speech, particularly in multilingual situations where speakers of different languages are present. However, this does not indicate confusion or a fused, undifferentiated language system. Rather, multilingual children often mix their languages because they borrow words from a nonrelevant language to fill lexical gaps in the target language. Multilingual children’s language mixing also shows a great deal of pragmatic sensitivity, being guided by the interlocutor’s attitude toward inappropriate language choices, by attention to the specific language context, and by an awareness of the addressee as multilingual and of the interlocutor’s language proficiency and preference.

10.4 Multilingualism and Implicature

Research on implicature has mainly focused on bilinguals and L2 learners (e.g., Reference Antoniou and TaguchiAntoniou, 2019; Reference Taguchi, Yamaguchi and TaguchiTaguchi & Yamaguchi, 2019). Two broad accounts regarding bilingual pragmatics emerge from this literature. According to the first view, bilinguals enjoy a pragmatic advantage. This account is mainly supported by research with preschoolers showing that bilinguals outperform monolinguals in the comprehension of SI (e.g., Reference Siegal, Matsuo, Pond, Otsu and OtsuSiegal et al., 2007) and other pragmatic skills (see Section 10.1). Explanations for this benefit include bilinguals’ superior executive control (e.g., Reference Siegal, Iozzi and SurianSiegal et al., 2009), their increased experience in using the appropriate language with speakers of different languages (e.g., Yow & Markman, 2001b), and a compensation process that makes up for bilingual’s slower language development with accelerated pragmatic acquisition (e.g., Reference Siegal, Tallandini, Pelizzoni, Michelin, Siegal and SurianSiegal et al., 2012).

The second account proposes a unified model of pragmatic development, representation, and processing in bilinguals and L2 learners. Drawing on previous theoretical work on bilingualism, L2 learning, and pragmatics (e.g., Reference GriceGrice, 1989; Reference Sperber and WilsonSperber & Wilson, 2002; Reference TomaselloTomassello, 2008; Reference SlabakovaSlabakova, 2010; Reference SoraceSorace, 2011; Reference Sorace2012; Reference KecskesKesckes, 2015), this account suggests that bilinguals and language learners develop a single pragmatic system – at least for pragmatic knowledge that might be considered universal, such as Grice’s maxims – that is independent from their languages,Footnote ¹ and is acquired and functions in a similar way to monolinguals or native speakers (Reference Antoniou and TaguchiAntoniou, 2019; Reference Antoniou, Veenstra, Kissine and KatsosAntoniou et al., 2020). On this view, nonverbal pragmatics is indistinguishable in bilinguals, language learners, and monolinguals in terms of real-time use, rate, and path of acquisition. However, linguistic pragmatics depends to some extent on target language proficiency, because, arguably, some degree of language processing is necessary for pragmatic interpretation to proceed. This language proficiency effect is particularly evident, for monolinguals, bilinguals, and language learners alike, during initial stages of language acquisition and/or for difficult pragmatic phenomena. At this phase, language is less developed in terms of competence and automaticity and, thus, gaps in language knowledge and increased demands of language processing may negatively affect pragmatic performance. Moreover, some aspects of language develop slower in bilinguals (e.g., Reference Bialystok, Craik, Green and GollanBialystok et al., 2009). Thus, this account suggests that, during initial stages of language acquisition, when bilinguals and language learners have markedly lower target-language proficiency than monolinguals or native speakers, they will exhibit lower pragmatic understanding skills. This difference, however, is not in pragmatic processing or knowledge per se, but is due to insufficient proficiency in the target language, and will be overcome when bilinguals and language learners attain adequate command of the target language for language processing, to the degree necessary, to proceed unobstructed and allow further pragmatic processing. This view is based on two empirical findings: first, target-language proficiency effects on pragmatic performance in bilinguals and L2 learners; and, second, that bilinguals and L2 learners are often reported to exhibit performance comparable to monolinguals or native speakers in the comprehension of various types of implicatures (Reference Syrett, Lingwall and Perez-CortesSyrett et al., 2017; Reference Antoniou and TaguchiAntoniou, 2019).

A third view stems from a general account of bilingual language processing, the Interface Hypothesis (Reference Sorace and SerratriceSorace & Serratrice, 2009; Reference SoraceSorace, 2011, Reference Sorace2012). According to the Interface Hypothesis, computationally demanding linguistic phenomena, typically at the interface of language-internal (e.g., syntax) and external domains (e.g., pragmatics), are the locus of particular difficulties in bilinguals and language learners, which cannot be overcome even at the highest level of language proficiency. Such linguistic processes are particularly problematic in bilinguals and language learners because, for them, language processing is less automatic and because, perhaps relatedly, they need to devote effort in managing competition from the non-target language. In this respect, bilinguals and language learners have fewer resources to allocate to the language interpretation process. A prediction that may derive from the Interface Hypothesis, then, is that bilinguals and language learners are likely to exhibit difficulties in pragmatic comprehension, even at near-native levels of language proficiency. This is because (1) there are theoretical reasons and some empirical evidence to suggest that the interpretation of pragmatic meanings is resource demanding; and (2) pragmatic phenomena involve the interaction of language with extralinguistic factors (see Section 10.1).

Studies that examined implicature comprehension in multilinguals can be divided into two groups: (1) research that compared multilinguals and monolinguals (e.g., Reference Antoniou and KatsosAntoniou & Katsos, 2017); and (2) work that examined various multilingual experiences on a continuous scale without a monolingual or native comparison group (e.g., Reference Tiv, Rouillard, Vingron, Wiebe and TitoneTiv et al., 2019). The results from two studies that took the comparative approach are more in line with the second account of bilingual pragmatics (Reference Antoniou and KatsosAntoniou & Katsos, 2017; Reference Antoniou, Culbertson, Perfors, Rabagliati and RamenzoniAntoniou, 2022); specifically, they suggest that multilinguals can exhibit similar to monolinguals pragmatic performance, even for linguistic processes such as irony, which depends on nonlinguistic factors and is evidently demanding from a theoretical, developmental, and processing perspective (Reference Filippova and MatthewsFilippova, 2014; Reference Spotorno and NoveckSpotorno & Noveck, 2014; Reference Antoniou and MilakiAntoniou & Milaki, 2021).

The two studies compared multilingual, bi-dialectal, and monolingual participants. Multilinguals were bi-dialectal speakers of two Greek dialects – Cypriot and Standard Greek – and also used one or more languages. The first study tested school-aged children and reported no group differences in the comprehension of manner, relevance, scalar implicatures, and novel metaphors (Reference Antoniou and KatsosAntoniou & Katsos, 2017). Target-language proficiency was also a positive predictor of overall pragmatic performance in this study. Similar results were obtained in the second study with young adults, which focused specifically on the accuracy and speed of processing irony (Reference Antoniou, Culbertson, Perfors, Rabagliati and RamenzoniAntoniou, 2022). Irony comprehension in this study was difficult for all participants, shown by lower accuracy rate and slower reaction times for ironic compared to literal items. However, once more, there were no group differences in irony processing.

Research that examined multilingualism as a continuous, multicomponent rather than a dichotomous experience has not exclusively focused on multilinguals, but we review this work here because it used mixed samples of bilinguals and multilinguals (Reference Tiv, Rouillard, Vingron, Wiebe and TitoneTiv et al., 2019; Reference Tiv, Deodato, Rouillard, Wiebe and TitoneTiv et al., 2021). Reference Tiv, Rouillard, Vingron, Wiebe and TitoneTiv and colleagues (2019) found that L2 proficiency positively predicted general use of sarcastic irony – a particular form of verbal irony – as measured with a self-report questionnaire, across multilinguals’ languages. Further analyses revealed that multilinguals used sarcastic irony largely with the same communicative functions as monolinguals, such as, for example, for frustration and embarrassment diffusion. Another study by Reference Tiv, Deodato, Rouillard, Wiebe and TitoneTiv and colleagues (2021) examined the interpretation of two types of ironic utterances in the L1: ironic compliments – cases of irony where the speaker says something negative to communicate the opposite – and ironic criticisms – where the speaker says something positive to mean something negative. Their results showed that participants were slower to decide whether ironic compliments made sense; and were also likelier to judge them as less sensible than ironic criticisms. This finding replicated past findings with monolinguals (Reference Tiv, Deodato, Rouillard, Wiebe and TitoneTiv et al., 2021). Finally, Reference Tiv, Deodato, Rouillard, Wiebe and TitoneTiv and colleagues (2021) reported that higher L2 proficiency predicted greater accuracy in evaluating whether statements made sense for both irony types, and shorter reaction times for ironic compliments.

Overall, results from research that has examined implicature in multilinguals compared to monolinguals show that the predictions of the second account of bilingual pragmatics (Reference Antoniou and TaguchiAntoniou, 2019) further hold for multilinguals. First, there is some evidence for a positive effect of target-language proficiency on pragmatic interpretation, and second, findings reveal that multilinguals have implicature understanding skills comparable to monolinguals, at least in one of their languages (Reference Antoniou and KatsosAntoniou & Katsos, 2017). This is also true for particularly demanding pragmatic phenomena such as irony (Reference Antoniou, Culbertson, Perfors, Rabagliati and RamenzoniAntoniou, 2022). Moreover, the two studies that treated multilingualism as a continuous experience reported that increased multilingualism – as reflected in higher L2 proficiency – is positively associated with greater use of sarcastic irony across languages and better L1 irony comprehension (Reference Tiv, Rouillard, Vingron, Wiebe and TitoneTiv et al., 2019, Reference Tiv, Deodato, Rouillard, Wiebe and Titone2021). At face value, the latter results are more consistent with the bilingual pragmatic advantage account. These findings, however, indicate a relative advantage for more proficient compared to less proficient multilinguals, but do not necessarily show an absolute multilingual advantage relative to monolinguals or native speakers. All three accounts relevant to bilingual pragmatics express their predictions with reference to monolingual or native speakers, and, hence, for their predictions to be tested, a comparison with a monolingual or native group is necessary. Thus, the findings of studies that adopted a continuum approach cannot adjudicate between the different views. However, they do show that certain aspects of multilingualism influence pragmatic performance and need to be incorporated into these accounts.

10.5 Multilingualism and Speech Acts

The bulk of studies on multilingualism and pragmatics have focused on speech acts. We review this work by compartmentalizing it into three sections: studies focusing on the bilingual advantage in L3 pragmatic acquisition, research examining crosslinguistic influence and the effect of target-language factors on pragmatic processing in that language; and studies that investigated pragmatic performance through a multilingual approach that considers all languages spoken by multilinguals. We divide the literature this way solely for ease of exposition and do not necessarily imply that the studies in each section represent discrete lines of research. On the contrary, the studies in the three sections have often offered complementary perspectives and insights on various aspects of multilingual speech act performance and acquisition.

10.6 Conclusions and Directions for Future Research

This chapter provided an overview of research on pragmatic competence, processing, and acquisition in multilinguals. In this concluding section, we briefly summarize and synthesize the main findings from this literature, pointing, where appropriate, to the differences and similarities of multilingual relative to bilingual and monolingual pragmatic performance. We close this section by suggesting directions for future work.

Studies that examined pragmatic differentiation in simultaneous multilingual children have generally shown that these children, just as previously reported for simultaneous bilingual children, can differentially and appropriately use each of their languages in a context-sensitive manner, from around the age of two. Code-mixing is evident in multilingual children’s language production, but this does not indicate a lack of pragmatic awareness. Rather, inappropriate language use might be caused by proficiency gaps in the contextually relevant language and is often pragmatically guided (Reference MontanariMontanari, 2008; Reference QuayQuay, 2008). For instance, multilingual children code-mix more when the communicative environment provides them with cues that such linguistic behavior is an acceptable and effective communicative strategy, such as when their interlocutors are accepting of their code-mixing and/or are multilinguals. Moreover, some evidence from speech acts suggests that sequential multilingual children exhibit similar pragmatic differentiation, in that they can differentiate and appropriately use their L3 pragmatic system from the early moments of L3 learning (Reference Safont-JordàSafont-Jordà, 2011).

The evidence from research on implicature reveals that multilinguals exhibit pragmatic interpretation skills comparable to monolinguals, at least in one of their languages (Reference Antoniou and KatsosAntoniou & Katsos, 2017; Reference Antoniou, Culbertson, Perfors, Rabagliati and RamenzoniAntoniou, 2022). However, this work further indicates that degree of multilingualism, measured continuously through L2 proficiency, positively affects general use of sarcastic irony across languages and L1 irony understanding (Reference Tiv, Rouillard, Vingron, Wiebe and TitoneTiv et al., 2019, Reference Tiv, Deodato, Rouillard, Wiebe and Titone2021). This evidence does not necessarily demonstrate a multilingual pragmatic advantage compared to monolinguals, but does show that additional factors are at play during multilingual pragmatic processing compared to monolinguals. Moreover, the work on implicature, but also studies on speech acts, have further shown that, like what has been found with bilinguals, pragmatic skills in the target language (L1, L2, or L3) are positively influenced by increasing target-language proficiency (e.g., Reference Safont-JordàSafont-Jordà, 2005a; Reference Stavans and Webman ShafranStavans & Webman Shafran, 2018). However, this linear association for the L1 and L2 might be disrupted with the introduction of an L3 in young children (Reference Safont-JordàSafont-Jordà, 2011).

Studies that have examined speech acts from the perspective of crosslinguistic influence have provided some evidence that the L1 can affect L2 and L3 speech act performance through the transfer of socio-pragmatic elements from the L1 to the L2 and L3, though the evidence is mixed (Reference Koike and PalmiereKoike & Palmiere, 2011; Reference Webman ShafranWebman Shafran, 2019). Furthermore, this crosslinguistic influence is not unidirectional, but can also occur from the L3 to the L1 and L2, as Pau’s case study illustrates (Reference Safont-JordàSafont-Jordà, 2011). Moreover, the literature on speech act performance has shown that bilinguals, and especially productive or balanced bilinguals, enjoy an advantage relative to monolinguals in learning the pragmatics of an additional language (e.g., Reference Safont-JordàSafont-Jordà, 2005a; Reference Alcón-SolerAlcón-Soler, 2012; Reference TrebitsTrebits, 2019). Thus, this research identifies bilingualism as an experience that is uniquely relevant to multilingual pragmatic performance and acquisition. Finally, the research on speech acts indicates that explicit pragmatic instruction positively influences pragmatic acquisition in an additional language, even though bilinguals learning an L3 might benefit more than monolinguals learning an L2 (e.g., Reference Safont-JordàSafont-Jordà, 2005a; Reference Alcón-SolerAlcón-Soler, 2012).

Clearly, the work conducted so far has contributed significantly toward a better understanding of multilingual pragmatic performance and development. However, various questions remain open and await further investigation. A first issue is methodological and relates to how the concepts of interest – multilingualism, specific multilingual experiences, and pragmatic performance – are measured. For instance, much of the reviewed research has focused on language proficiency as an attribute relevant to bilingual and multilingual pragmatics. This construct arguably involves multiple knowledge (e.g., morphosyntax, lexicon) and skill (e.g., listening, speaking) components (Reference Hulstijn, Blom and UnsworthHulstijn, 2010). However, language proficiency in the reviewed studies has been often evaluated through single assessments, such as vocabulary tests, that only partially reflect its multidimensional nature (e.g., Reference Antoniou, Culbertson, Perfors, Rabagliati and RamenzoniAntoniou, 2022). Relatedly, the reported work has examined pragmatic performance by exclusively relying on behavioral measures, which only crudely reflect the end-product of pragmatic processing. Other psycho- and neurolinguistic tools, such as, for instance, eye tracking and Event-Related Potentials, have, to our knowledge, not been employed in this field. This is despite their past use in and the knowledge accumulated from monolingual studies (see Section 10.2), which could inform the work in this research area. Moreover, such methods have the potential to measure performance on a more fine-grained level of analysis and to examine multilingual effects on the stream of processing (e.g., Reference Grey, Tagarelli, Phakiti, De Costa, Plonsk and StarfieldGrey & Tagarelli, 2018). This may reveal subtle multilingual effects that could not be detected in past research. Thus, future studies should examine the effect of multilingualism and multilingual experiences by using measures that account for the multicomponent nature of these constructs; and by taking advantage of other techniques from psycho- and neurolinguistics. This work may increase confidence in or revise past findings and conclusions but either way it will certainly enhance our understanding of multilingual pragmatics.

Second, research on the bilingual advantage in L3 pragmatic learning has exclusively focused on speech acts. Hence, it is important for future work to investigate whether this benefit generalizes to other pragmatic phenomena. Moreover, it is still unclear from this literature which factors are responsible for bilinguals’ superior performance in L3 pragmatics. Possible explanations include the skills and strategies acquired from bilinguals’ past language learning experience, and their enhanced metalinguistic awareness, executive control, and communicative sensitivity (e.g., Reference Safont-JordàSafont-Jordà, 2003: 59, Reference Safont-Jordà2005a; Reference Alcón-SolerAlcón-Soler, 2012; Reference TrebitsTrebits, 2019). Opinions on this matter have been so far speculative, with no direct empirical studies. Moreover, another issue is whether bilingualism merely accelerates the rate of pragmatic acquisition in an additional language, leading to an initial advantage over monolingual learners, which disappears with time, or whether bilingualism affects the pragmatic system in a more permanent way.

Third, current research on crosslinguistic influence in multilingual pragmatic performance has offered inconsistent evidence about transfer effects and the factors involved in such influence (e.g., proficiency level, language similarity; Reference Koike and PalmiereKoike & Palmiere, 2011; Reference Webman ShafranStavans & Webman Shafran, 2019). Future work using well-characterized multilingual groups should more closely examine the conditions under and the direction in which multilinguals’ languages influence each other in terms of pragmatics.

Finally, it is necessary to investigate multilingual pragmatics by taking a holistic approach that examines not only one, but all languages spoken by multilinguals (Reference CenozCenoz, 2013; Reference Safont-Jordà and ChappelleSafont-Jordà, 2013a, Reference Safont-Jordà, Cenoz, Gorter and May2017; Reference Cenoz and GorterCenoz & Gorter, 2019). So far, this approach has been employed only in studies on speech acts (e.g., Reference Safont-JordàSafont-Jordà, 2011; Reference Portolés-FalomirPortolés-Falomir, 2015). Evidently, such a perspective, adopted in investigations of different pragmatic phenomena and with different multilingual groups, would reveal a clearer and more complete picture of multilingual pragmatic knowledge, processing, and acquisition.