1 Recursion: Easy to Describe, Not Always as Easy to Find
Recursion is defined in a number of ways, from what Pinker and Jackendoff (Reference Pinker and Jackendoff2005:203) call “a procedure that calls itself, or ... a constituent that contains a constituent of the same kind,” or in the words of Rodgers and Black (Reference Rodgers, Black, Pieterse and Black2004), “a data structure that is partially composed of smaller or simpler instances of the same data structure.”Footnote 1 Classic examples of recursion in syntactic theory arise with rewrite rules of the form developed in Post (Reference Post1943) and subsequently familiar throughout all syntactic theory that include, among many other rules of course, the following key interacting rules:
| (1) | a. S → NP VP |
| b. NP → that S | |
| c. VP → V NP | |
| d. NP → the N |
The combination of these rules yields indirect recursion, as expansion of (1a) into the VP rewrite rule in (1b) yields an infinite set of sentences of the form The dog thought that the cat saw the rat or The dog thought that the bird said that the cat saw the rat. With only the four rules in (1), an infinitude of sentences can be derived, and as Bar-Hillel (Reference Bar-Hillel1953:164) pointed out, in the recursive analysis of such sentences with expansions like (1b), “we had to move from ‘sentence’ to ‘nominal’, then back to ‘sentence’, and finally once more to ‘nominal’.”
Similar properties arise once conditionals are added, as in (2), which draws on examples formulated in Chomsky (Reference Chomsky1957):
| (2) | S → if S then S |
Thus we derive If the cat saw the rat then the dog told the bird and even self-embedded conditionals like If if the cat saw the rat... Indeed, one can apply such a formalism to NPs themselves:
| (3) | a. NP → the N’ |
| b. N’ → N PP | |
| c. N’ → N | |
| d. PP → P NP |
This set of rewrite rules will yield The book on the table near the mat and so forth.
Taking what these have in common, we may say that an algorithmic generative rule system is recursive if the output of a given rule R2 (like 1b or 3d) contains a symbol or sequence of symbols Z that is also part of the input of a rule R1 (like 1a or 3a), such that R1 immediately or remotely generates Z. Note that R1 and R2 may be identical, as in (2).
This is an incredibly powerful property for languages to have. Husserl’s (Reference Husserl1913) Logical Investigations wrote of a way of obtaining “a boundless multitude of further forms” (as translated in Goldsmith and Laks Reference Goldsmith and Laks2016) by the recursive combination of propositions in this way, a characteristic of human language whose nature has been studied in virtually all subsequent work on the topic. Notably, however, the discussion of recursion in phrase structure in the first few decades of research on the topic focused almost exclusively on recursion involving sentential embedding (see Graffi Reference Graffi, Busà and Gesuato2015 for discussion). However, recursion, whether of the center-recursion kind in (2), or the edge-recursion kind in (3), can be found across domains beyond sentential embedding alone, and once PP and NP recursion are included, it is potentially found in some form or another in all languages. This depends, of course, on the formalism used to express it grammatically, as a language that allows, say, up to four adjectives (the shiny bright expensive green bottle) has a more parsimonious grammatical description than one that needs to add a new phrase structure rule for every adjective that is added (as in (5)):
| (4) | a. NP → the N’ |
| b. N’ → Adj N’ | |
| c. N’ → N |
| (5) | a. NP → the N |
| b. NP → the Adj N | |
| c. NP → the Adj Adj N | |
| d. NP → the Adj Adj Adj N | |
| e. NP → the Adj Adj Adj Adj N |
Nonrecursive formalisms (such as (5)) could be written if, say, a corpus of a language never found more than four adjectives in a row, and a linguist who decided to model a grammar based on this insisted such a corpus was truly representative of the language. Conversely, one might question whether, indeed, the limit to four adjectives was found specifically because this was a finite corpus, based on limited genres like monologue narratives with no real-time conversational interactions, instead of being from a wide range of dyads, facing goal-oriented demands, or contextually varied situations of wooing, scolding, or planning. If elicitation tasks reveal that narrative-based corpora present only a limited sample of the language (see Davis, Gillon, and Matthewson Reference Davis, Gillon and Matthewson2014), then formalisms such as (5) will be inferior to the coverage and compactness of those in (4).
How to tie the widespreadness (or lack thereof) of a given recursive pattern in actual usage to its formalism? It is well established (see Chomsky and Miller Reference Chomsky, Miller, Luce, Bush and Galanter1963; Gibson and Thomas Reference Gibson and Thomas1999) that center-embedding becomes very difficult to process, to the point of rendering grammaticality judgments difficult after three embeddings. Clearly, therefore, recursion in this domain involves more processing costs. Focusing on this kind of recursion would make one doubt its existence. By contrast, focusing on possessor recursion might make one quite confident that recursion is alive, kicking, and easy to process; Lima and Kayabi (this volume) find, for example, that Kawaiwete-speaking children correctly answer questions like “What is Pedro’s friend’s brother’s basket’s color?” Comparison of recursion across domains is therefore crucial, and particularly across a range of languages and with a range of methods, especially when grappling with the question of whether recursion is at the center of every language (as many interpret Hauser, Chomsky, and Fitch Reference Hauser, Chomsky and Fitch2002).
It becomes immediately clear that no matter how easy it may be to write down a recursive set of rules such as (1)–(3), the full-blown use of recursion does not occur as widely as we might expect, as reflected in typological skewedness, language-specific limitations, real-time processing costs as measured behaviorally, with eye-tracking or brain-imaging, and finally in its acquisition profile. Children’s acquisition of clausal embedding of the type generated in (1), studied by Bloom et al. (Reference Bloom, Rispoli, Gartner and Hafitz1989), for example, has been argued to be somewhat limited in its usage at an early age. Diessel (Reference Diessel2004), in a thorough study of the uses of clausal embedding introduced by “I think that...” in child language, argues that in such cases, the “apparent matrix clauses are nonassertive: rather than denoting a mental state or an act of perception, they function as epistemic markers, attention getters, or markers of illocutionary force” (p.3). In fact, children sometimes omit the complementizer ‘that’ in even German and French, languages in which the complementizer is obligatory for adults, and frequently use “I think” as a sentence-final parenthetical. In such a characterization, the relationship between expressions such as “I think” and the proposition they introduce does not follow the hallmarks of true embedding, where embedded propositions are syntactically and semantically integrated in the matrix clause and marked as dependent structures that are formally incomplete without the matrix clause. Instead, one might treat these utterances with a rewrite rule:
| (6) | S → NP think that NP VP |
A rule such as (6) is nonrecursive in the sense that it explicitly rules out double-embedding. Diessel’s argument is that younger children start out with rule systems such as (6), only gradually later giving way to a revision of the type that yields (1).
Along somewhat similar lines, not all languages employ direct clausal embedding of the type that (1) would derive. Levinson (Reference Levinson2013), for example, claims that recursion in Kayardild is limited to one level deep, precisely as (6) would restrict it, although for reasons of case morphology related to being polysynthetic. Indeed, it is sometimes claimed that the type of grammatical reanalysis that is responsible for children’s transitions from paratactic embedding structures as in (6) to fully fledged embedding structures that would be generated by (1) finds parallels at longer timescales, in the kind of grammaticalization that diachronically yields clausal embedding from formerly paratactic structures. Givón (Reference Givón2009) shows how this kind of reanalysis takes place with examples from Bambara, Hittite, Germanic, and Uto-Aztecan, arguing that one of the parallel components in this reanalysis is when the previously adjoined material now “falls under a single merged intonation contour with the main clause” (p.202). Similarly, Hale (Reference Hale and Dixon1976:78) argued that:
[in] a large number of Australian languages, the principal responsibility for productive recursion in syntax is shouldered by [an] adjoined relative clause. It is typically marked as subordinate in some way, but its surface position with respect to the main clause is marginal rather than embedded.
Similarly, Nordlinger (Reference Nordlinger2006) argues that structures such as they drink grog, they’ll fight in the Australian language Wambaya are fully ambiguous between coordinated and subordinated (if-then) relations (although she contends that the subordinate construal may be forced by prosody). Nordlinger’s discussion, alongside a detailed overview of Australian language data by Legate, Pesetsky, and Yang (Reference Legate, Pesetsky and Yang2014), makes clear that it is far from correct to say that subordination is lacking in these languages, but nonetheless that this kind of recursion is not as freely used as it is in English, or in comparison to adjectival recursion in the same languages. If anything, then, while it is definitely too radical and simplistic to say that any language ‘lacks’ recursion, its distribution parallels that of, say, interdental fricatives like /θ, ð/ in English: positionally limited (found only before r in clusters), not often found with more than one instance (thither), and notoriously difficult in L1 and L2 acquisition.
What leads to these markedness-like restrictions on ‘a constituent that contains a constituent of the same kind,’ and how do they line up with different kinds of recursion? Why is it that some languages are more restrictive with CP recursion of the type exhibited in (1) than they are with PP recursion exhibited in (3)? Why do some languages allow recursive noun-noun compounding (e.g., comic book club) while others do not (Roeper and Snyder Reference Roeper and Snyder2004)? Recursion looks like a candidate for the application of Jakobson’s (Reference Jakobson1941) criteria of markedness as compellingly linked in his book Kindersprache, Aphasie, und Allgemeine Lautgesetze, in which he argued that marked phonological structures are rarer cross-linguistically as well as later to develop in child language. Again, these questions must be asked with respect to a much richer range of structures than center-embedding relative-clauses or sentential embedding of speech reports. What is the typological, acquisitional, and processing profile of DP recursion? Of PP recursion? Of causatives, evidentials, imperatives, and other underexplored XPs potentially allowing being pushed inside ‘a constituent of the same kind’? Coming back to (3), with the case of PPs for example, interpretational issues arise. How do we know whether the interpretation of the book on the table next to the mat is really one of next to the mat modifying the book, rather than the table? How do we know that, for example, real-time instances of these are implemented in comprehension by recursion, rather than iteration? How does cross-linguistic transfer of recursive structures take place in L2 acquisition (see Nelson Reference Nelson2016 for PP recursion across Spanish-English L2)?
Moreover, certain kinds of XPs also naturally lend themselves to embedding more easily than others. For example, imperatives like Go fetch the water are crosslinguistically less likely to be embeddable. On the other hand, the kind of recursion found with possessors, as in My sister’s husband’s boss got us the tickets, are extremely natural in adult English. (And as mentioned, in Lima and Kayabi’s chapter in this volume, up to four levels of possessor recursion are easily processed by Kawaiwete-speaking children.) In other languages, however, restrictions are imposed on such structures. As argued in Nevins, Pesetsky, and Rodrigues (Reference Nevins, Pesetsky and Rodrigues2009b), such restrictions may be in some instances morphosyntactic: languages with certain kinds of morphological systems for case or finiteness may disallow one instance of finite morphology to be c-commanded by another. Rizzi (Reference Rizzi2013) contends that while the formal operations yielding recursion are fully available as part of general computation, its applications are modulated by the specific properties of the lexical items it acts upon. Just such an investigation of which properties, and how they modulate the possibilities of recursion, is precisely at stake in this current volume.
How does the possibility of recursive operations within particular domains come to change over various timescales – the timescale of a child that integrates language use with other cognitive systems, for example? Consider sentences such as (7):
| (7) | What did the girl say she bought? |
De Villiers (Reference Villiers, Baron-Cohen, Tager-Flusberg and Cohen1999) showed that when children under four years are asked to answer questions like this one, they tend to answer with what the girl actually bought rather than what she said she bought, which of course might be different. According to de Villiers, these children have not yet analyzed (4) in terms of an embedding syntax and semantics; indeed, she argues that cognitive development and language acquisition are mutually dependent, and that the development of faculties such as Theory of Mind and false belief go hand in hand with the analysis of (7) as fully fledged embedding. The dramatic connections between lack of subordination and false belief – the question of when children realize the “opacity” of complement sentences – parallels the evolution of subordination types along the acquisition path. These, in turn, mirror the variation across languages discussed above. A subpath of stages, with stepwise semantic changes as well (see Roeper and de Villiers Reference Roeper, de Villiers, de Villiers and Roeper2011; de Villiers et al. Reference Villiers, Roeper, Harrington and Gadilauskas2012), precedes the full form of recursive complementation that syntactically and semantically represents false belief. These types of subordination are stepping stones to the full subcategorization of clauses that enables recursive structure for multiple points of view. In their contribution to this volume, Roeper and Oseki argue that there is an acquisition path from coordination to indirect recursion.
While recursion across the domains of sentential embedding, prepositional phrases, causative structures, possessors, and relative clauses can all be described formally with the same means (e.g., the rewrite rules in (1)–(3)), their distribution across and within the world’s languages is clearly not equal. With a focus on cross-linguistic diversity, this volume includes experiments on PP recursion, possessive recursion, relative clause recursion, adjective recursion, sentential recursion of both tensed and non-finite clauses, and discourse recursion. This will allow us to begin to consider new kinds of questions: are there clusters of recursive structures that are reflected in typology, dialects, or language change? Are they acquired in a systematic way with one kind triggering another (as discussed in Roeper and Oseki’s chapter)? Finally, what interfaces do they have with morphology, parametric variation, and lexical representation?
Our goal in this volume is to bring new data and emerging research methodologies from a gamut of less familiar languages to this study. This book aims to address a host of topics about recursion woven together across different dimensions of linguistic research: formal analysis, theoretical exploration, experimental fieldwork, and several methodologies (intuition, comprehension experiments, event-related potentials, and reaction time studies). Recursion is held up against its interaction with reference, evidentiality, second-order beliefs, and prosody across domains and latitudes, and it is compared to non-recursive structures (e.g., the ‘coordination default’) across fifteen languages, thereby informing the question of its distribution, processing, and restrictions across grammatical domains and across grammars.
2 Recursion across Latitudes
This book presents analyses of recursive constructions in a broad array of languages representing a great areal, typological, and genetic diversity, spanning wide latitudes. Fifteen languages are examined in the eighteen chapters of this book. Dutch, English, Portuguese, and Japanese are studied, alongside eleven Amerindian languages of South America, coming from six distinct language families or genetic affiliations.
The region of the world that is currently most prominent in contemporary discussions of recursion is lowlands South America (east of the Andes), in particular the regions surrounding the Brazilian Amazon, in large part due to the claims that arose in Everett (Reference Everett2005) that Pirahã, an indigenous language of Brazil, has ‘no embedding,’ and the subsequent debates that arose about (a) what, in fact, are the right kind of empirical tests to be conducted to make such claims (see Sauerland and Sandalo et al. in this volume for new kinds of experimental inquiry for this language) and (b) whether, if indeed, Pirahã were to be missing recursion in one particular domain (e.g., CP), what consequences this might have for developing a theory of markedness of recursive structures. Given particularly the linguistic diversity of Brazil, the unfamiliarity of many of these languages to mainstream debates on recursion, and the vibrant presence of a range of interdisciplinary methods applied by researchers working on them, their contribution to the questions laid out above is extremely valuable.
We turn, therefore, from a mapping out of linguistic domains to a necessary introduction to the distribution of recursion across latitudes as reflected in the research reported on languages within Brazil.
Brazil has, at present count, around 150–160 indigenous languages (Moore Reference Moore, Mello, Altenhofen and Raso2011). The present volume considers phenomena from ten living Brazilian indigenous languages (depicted in Figure 0.1), as well as Tupinambá, an extinct language spoken on the coast of Brazil when the first colonizers arrived. These comprise thirteen chapters in the book. Alongside these are English, Portuguese, Japanese, and Dutch, which are examined in the other five chapters, constituting the total eighteen chapters of the book. In addition to their wide areal distribution, ranging from Kotiria and Wapichana in the northwest and north regions of the country, Pirahã in the south Amazon area and Karitiana in the western state of Rondonia, to Guarani in the south and southeast, with a concentration of languages in Central Brazil (Karajá, Kawaiwete, Kĩsêdjê, Kuikuro), the languages studied in the book are also representative of some of the main linguistic stocks and families in Brazil: Tupian, Macro-Jê, Carib, Arawak, and Tukano (alongside an isolate, Pirahã).
Figure 0.1 Living languages with approximate locations
The Tupian stock, the largest South American genetic group, is represented by languages from both its western branch (Karitiana, the last surviving language of the Arikem family) and eastern branch (Kawaiwete, Tenetehára, Mbyá Guarani), all belonging to the Tupi-Guarani family, the largest family in the stock. Damaso Vieira analyzes recursive constructions in Tupinambá, an extinct language of the Tupi-Guarani family, drawing on data registered in the sixteenth and seventeenth centuries.
Languages of the second largest genetic group in Brazil, the Macro-Jê stock, are also well represented in the book: Kĩsêdjê, a member of the Jê family, the largest family in the stock, and Karajá (Karajá family). The Carib family, which contains the second largest number of languages in a single family, is represented by Kuikuro, a language classified in the Xinguan southern branch of Carib. The Arawak family is represented in the book by Wapichana, the North Arawakan language with the largest number of speakers. Kotiria represents the Eastern Tukano family, spoken in the northwestern Amazon basin. Finally, Pirahã, which is examined in three of the chapters in the book, is the last surviving language of the small Mura family, spoken in the south Amazon area.
Even though the Brazilian indigenous languages investigated in the book present different levels of vitality according to UNESCO criteria, there is a clear consensus among linguists working on indigenous languages in Brazil that all of these languages are classified at least as “vulnerable” by UNESCO’s endangerment criteria. Spoken by a mostly monolingual population, Pirahã has a high level of intergenerational transmission, but the sheer low number of speakers (less than 400) makes it vulnerable. With a population of around 7,000 speakers in Brazil, Wapichana, on the other hand, is considered “definitely endangered” by UNESCO, based on the fact that less than half of the population can speak the language (see Moore Reference Moore, Mello, Altenhofen and Raso2011) and that at least 80 percent of the population is bilingual in Portuguese. While most of the languages discussed are largely understudied, the research presented in this book, with the exception of the study on extinct Tupinambá, are based on first-hand data, collected by the authors of the chapters themselves.
All of the studies presented in the book were developed based on a cross-comparative methodology, in which theoretical questions guide data collection. Some of the studies engage in the precision offered by experimental methods, a new endeavor which is being called “experimental fieldwork,” facing the challenge of bringing together crucial dimensions of linguistics such as theory of grammar, psycholinguistic methods, and fieldwork procedures in order to attempt to uncover grammatical processes that could never be discovered solely on the basis of corpus building. Thus, while our focus in this section has been on the typological diversity of the languages covered here, of equal importance in our organization of this volume is to convey the range of similar methodologies that can and should be applied across them. There are thus direct connections between the issues raised and Theory of Mind, PP recursion, and coordination and subordination in parallel investigations with English, Portuguese, Japanese, and Dutch found throughout this book (see especially the contributions by Terunuma and Nakato, Hollebrandse, and Pérez-Leroux et al.). At the same time, the Brazilian languages represented herein involve empirically foci that are largely absent from well-studied European languages, e.g., evidential marking, switch-reference marking, and embedded imperatives.
3 Recursion and Embedding across Domains
This volume is organized into four grammatical domains in which recursion is examined. Clearly, at some points there may be intersections or transversal connections possible across individual chapters within distinct sections.
3.1 Speech Reports, Theory of Mind, and Evidentials
The formal property of sentential embedding is invariably linked to the semantic and illocutionary types of elements they link up. Most canonically, these are speech reports, which involve reporting the beliefs or speech acts of others, and thus intersect with questions of Theory of Mind, second-order beliefs, the cognitive development of Theory of Mind in children, and evidential reporting more generally.
Sauerland’s chapter examines speech reports such as Toi said that he has been to the moon, which can be used to distinguish between subordination and coordination structures for speech reports. Specifically, if the relationship between the embedded proposition and the higher attitude verb is one of subordination, then the sentence as a whole can be judged as true. However, if Pirahã really lacked embedding, then sentences of this sort would actually be coordination structures, akin to Toi spoke, and he has been to the moon. The chapter discusses the result of an experiment conducted with sixteen Pirahã speakers, who as a whole end up distinguishing subordination (Toi said that he has been to the moon) from coordination (Toi spoke, and he has been to the moon), where the former as a whole can be judged as true at the same time that the latter is judged as false. The results therefore provide a new empirical base for the conclusion that Pirahã grammar contains at least one level of embedding, and moreover outline a technique for the study of speech reports that can be straightforwardly employed and replicated in experimental fieldwork situations with relatively understudied languages.
A further challenge, only recently beginning to be studied experimentally, is the extent to which one can trace the distribution and development of second-order belief ascription, as broached in the chapter by Hollebrandse, focusing on sentences like The judge knows that the jury thinks that Malcolm is guilty. These structures are particularly interesting for the subordination versus coordination dichotomy because, as Hollebrandse argues, there is virtually no way to anaphorically or paratactically express the second-order belief: only the recursive embedding serves to do so. They are also highly relevant, within the acquisition context, for understanding the relationship between linguistic encoding and the psychological faculty of Theory of Mind. The chapter presents two methods that can be used to elicit second-order belief ascription in language acquisition tasks, which, as we have discussed, are highly valuable as they are broadly adaptable to fieldwork situations in which one cannot always be certain what purely introspective acceptability judgments would be probing.
Focusing on Brazilian Portuguese, Corrêa, Augusto, Marcilese, and Villarinho present two sets of psycholinguistic experiments designed to investigate indirect recursion in children’s comprehension of specific reference in complex DPs with an internal modifier and in complement clauses subcategorized by non-factive mental state verbs. Relying on the assessment that recursion increases computational cost, the authors explore the extent to which children rely on indirect recursion in the analysis of complex structures. They call into question the proposal of a developmental path in which direct recursion would be the default analysis, previous to the acquisition of indirect recursion. The authors claim that 6 year olds are able to engage in second-order false belief reasoning, regardless of the type of structure in which false beliefs are expressed, and that the true difficulty lies in coping with a sentence composed of propositions having different truth values.
The relation between embedding and evidentiality comes to the fore again in Stenzel’s contribution. Stenzel reviews evidential categories in Kotiria in order to assess whether evidentials can be viewed as recursive constituents in the language, focusing on how to express the syntactic structure underlying the development of systems with multiple evidentials. After analyzing the structural means employed in the expression of four firsthand categories of evidentials, the author proposes that embedding can indeed be used to express detailed semantic concepts. However, there are boundaries to the number of derivable notions, since only one hierarchical level of embedding is possible, and there are few such morphological markers.
Thomas’ chapter turns to a different type of embedded speech act: imperatives, which can be embedded under certain verbs of saying in languages such as Mbyá Guarani. Thomas formalizes the phenomenon of embedded imperatives as recursive instances of the functional head ForceP, embedding a semantic type known as Speech Act Potentials. Thomas demonstrates that embedded imperatives in Mbyá Guarani are not some kind of paratactic quotation, but rather true instances of embedding occurring under verbs meaning ‘to say’ and ‘to ask to,’ and even under the reportative evidential particle je, so that an embedded speech act can involve a structure like Give me the maté! (‘I heard that order’). By developing a formal theory of speech act potentials, the chapter directly enables understanding these structures as an instance of recursion.
3.2 Recursion along the Clausal Spine
The chapters in this part of the book examine the phenomenon of embedding in categories along the clausal spine, from VP complementation in control configurations to IP coordination in switch-reference, TP-level embedding, and recursive causative constructions, across a number of indigenous Brazilian languages. They trace new empirical routes for investigating specific morphosyntactic phenomena, many of which can serve as future applicable diagnostics for recursive embedding, alongside developing specific analyses of how these constructions are grammatically assembled.
The chapter by Rodrigues, Salles, and Sandalo focuses on obligatory control in Pirahã. Using elicitation experiments, the authors analyze the syntax of desiderative constructions such as I want to eat fish. Using diagnostics based on the scope of temporal and locative adverbs as well as negation, the authors argue that in Pirahã, the complement VP to eat fish is distinct from the matrix verb, is subordinate to it (and not coordinated with it), and that the word-order alternation in which such nonfinite complements can appear either after or interestingly before the matrix verb indicate that this type of complementation indeed constitutes an instance of embedding a VP inside another VP. The data contribute towards the ongoing question of the extent to which various types of self-embedding can be unambiguously identified, and provide a set of diagnostics that can be replicated and extended to parallel questions in a number of other languages.
Nonato’s contribution examines the phenomenon of ‘switch reference,’ in which a morphosyntactic marker indicates whether the subjects of two coordinated TP clauses are coreferent or not. As the author reports, this was initially a useful diagnostic for what is known as symmetric versus asymmetric coordination (the latter being two clauses that have a causal relation or a temporally sequential relation and thus are not commutatively reordered), and potentially would enable modeling the two structures in terms of direct recursion and indirect recursion, in a transparent way that relates to the ‘coordination default.’ However, despite the initial appeal from earlier fieldwork that this morphosyntactic marker was a flag of different kinds of recursion, the author provides an update on more recent and rigorous elicitation conducted on Kĩsêdjê showing that even in symmetric coordination, switch-reference marking is obligatory. He presents the suggestion that further research both in Kĩsêdjê and across languages might find evidence for asymmetric coordination in the phenomenon of subset and superset switch-reference.
Duarte examines the derivation of embedded clauses in Tenetehára, in which clausal complements of perception verbs, such as Sergio saw Pedro waiting for the tapir, involve an unexpected complementizer-final word order in the embedded clause. He provides a detailed syntactic analysis in terms of phrasal roll-up movement of the embedded VP to the specifier of the embedded TP and then CP, with this step of predicate raising therefore resulting in the complementizer-final order with the VP preceding everything else, including a set of particles that he identifies as the tense position. The study is germane to the question of how strategies of embedding may be syntactically achieved within a formal model of the clausal cartography of tense projections.
Damaso Vieira examines recursive constructions at both the syntactic and the morphological levels, demonstrating that one of the most well-studied languages in Brazil, Tupinambá, the object of grammatical analysis since the sixteenth century, exhibits indirect recursion across different domains, such as complement clauses and verb incorporation, and causative, reflexive, and possessive constructions. She provides a comparison with the modern-day Tupi-Guarani language Mbyá, where at the syntactic level, recursion applies to complement clauses and possessives, and at the morphological level, recursion is observed in the contexts of verb incorporation and causatives, concluding that the parallel existence of recursion in these two subcomponents of the grammar has consequences for understanding word-formation in terms of a partially syntactic component. This contribution represents the application of questions of recursion to affixes in morphologically complex languages, a pursuit which has as of yet been all too little undertaken (though cf. Lander and Letuchiy Reference Lander, Letuchiy and van der Hulst2010).
3.3 Recursive Possession and Relative Clauses
The chapters in this part of the book deal with two very important types of linguistic constructions that have been at the center of research on recursion in recent years. Recursive possessives have been a recently lively area of study by acquisitionists looking into how children interpret and acquire multiple embedded constructions.
The first such chapter is by Terunuma and Nakato. In their chapter, the authors describe multiple embedded genitive constructions in Japanese, comparing them to their English counterparts. They present two experiments inspired by an original test proposed by Tom Roeper and previously used by other scholars (e.g., Limbach and Adone Reference Limbach and Adone2010; Leandro and Amaral Reference Leandro and Amaral2014; Lima and Kayabi, this volume). Terunuma and Nakato’s findings lead them to propose a specific sequence of acquisition of recursive possessives in Japanese with three stages. Their analysis supports current hypotheses about the acquisition of recursion based on developmental paths.
Lima and Kayabi offer another study into how children interpret recursive possessives. In the same chapter, they also show the results of an elicitation task with multiple embedded locatives introduced by PPs. Lima and Kayabi studied both constructions in a Tupi-Guarani language spoken in central Brazil called Kawaiwete. In this language, possessive relations with full NPs are expressed by juxtaposition with the possibility of multiple embedding. They piloted a similar experiment to the one designed by Terunuma and Nakato. In contrast to Japanese, Kawaiwete has no additional possessive particles or morphemes. For their second experiment with embedded locatives headed by postpositions, Lima and Kayabi conducted an elicitation task that showed that adult speakers use a strategy that avoids center-embedding, creating a word order where PPs are right-adjoined, thereby triggering successive extraposition of PPs.
Relative clauses constitute another important example of how different natural languages license recursive constructions. In this volume, there are two chapters that look into relative clauses in two distinct native Brazilian languages. Amaral and Leandro describe different types of multiple embedded relative clauses in Wapichana, their potential ambiguity, and their patterns of interpretation by speakers of different ages. In Wapichana, relative clauses can be verbal or nominal in nature, and there are no restrictions for multiple embedding, as long as the head of the previous sentence subcategorizes for a noun that can be further modified by another relative clause. Amaral and Leandro find that there is a clear difference in the interpretation of lower clauses in terms of high or low attachment, depending on the part of speech of the head of the main clause. Their research is a pioneering example of how to conduct psycholinguistic experiments with multiple embedded relative constructions in an understudied indigenous language.
In many of the less studied languages, we are in need of careful descriptions to prudently advance our knowledge of how recursive structures should be represented, and indeed judgments of ungrammaticality can be informative in this respect. That is exactly what Storto, Vivanco, and Rocha propose in their chapter, where they provide a detailed analysis of relative clauses in Karitiana, along with elicitation tasks to show preferred and/or accepted grammatical patterns for relative clauses in that language. They discuss the results of an elicitation task in which subjects overwhelmingly prefer the SOV order instead of OSV for Karitiana relative clauses. They then present a thorough analysis of relative clauses that function as oblique objects of main verbs, alongside examples of multiple embedded relative constructions. The chapter provides an excellent example of how fieldwork elicitation can be improved by using techniques from experimental linguistics.
3.4 Recursion in the PP Domain
The part of the book focusing on PP constructions presents five chapters that use different methodological approaches to examine these structures in Portuguese, English, Japanese, and in two Brazilian indigenous languages, Karajá and Kuikuro. Though very diverse in their theoretical scopes and in their methods of data collection and analysis, the five chapters present a noteworthy convergence in their specific conclusions. Taken together, they enable a strong takeaway message that embedded PPs are more complex to represent, acquire, and process than corresponding structures with conjoined PPs.
Roeper and Oseki set the stage with the proposal that recursion is represented by three primary types of recursion: Direct Unstructured Recursion, Direct Structured Recursion, and Indirect Recursion. The authors elaborate careful arguments about the relevance of Bare Phrase Structure for both linguistic theory and language acquisition, and add a new dimension to the typology of recursion, with the theoretical space carved out by Feature Sharing. The novel typology proposed by the authors allows them to state clear predictions about the child language acquisition path: since complexity of recursion increases from Direct Unstructured Recursion through Direct Structured Recursion to Indirect Recursion, the acquisition path should proceed in the same direction.
In their chapter, Sandalo, Rodrigues, Roeper, Amaral, Maia, and da Silva present two pilot tests investigating PPs in Pirahã. In a picture description test, three sets of pictures were presented to two Pirahã subjects, who were asked to indicate which of the pictures would better fit a description expressed in previously elicited Pirahã sentences. In the elicitation phase of the test, the Pirahã speakers systematically introduced the coordinative particle piai to describe the pictures, indicating coordination. However, in the interpretation phase, the participants correctly pointed at pictures, unequivocally demonstrating the capacity to discriminate between conjoining and embedding. The second test was an act-out game, in which participants had no problem in either formulating or following imperative commands involving two or three levels of PP conjoining or embedding, even though their more spontaneous preferences seemed to favor PP coordination.
Pérez-Leroux, Castilla-Earls, Béjar, Massam, and Peterson employed a referential elicitation task in order to test whether children differentiate between PP recursive and nonrecursive double modification in English. Results indicated that children (and adult controls) produced target descriptions twice as often to non-recursive than recursively PP modified NPs. According to the authors, the recursive and non-recursive conditions differ with respect to the distribution of phase boundaries, and unlike nonrecursive constructions, the recursive ones would require speakers to attend to a referent that would be no longer accessible in the active derivational workspace. They argue that this resistance to recursion is not caused by the syntax of the construction in itself, but has its roots in referential demands at the interface.
Franchetto’s chapter describes and analyzes the operations available in Kuikuro for the construction of DP and PP recursive structures. The author reviews general characteristics of Kuikuro phrasal prosody in order to propose that prosodic integration is the key for the identification of these constructions. After showing that DP recursive structure is distinct from DP coordination through evidence from intonation patterning in possessives, the study focuses on PPs and documents a mismatch between prosody and syntax, while at the same time suggesting a match between prosody and cognitive integration. This chapter demonstrates that pitch tracks, extremely straightforward to generate with contemporary software, provides evidence for distinct treatment of DP and PP recursion.
Finally, Maia, França, Gesualdi-Manhães, Lage, Oliveira, Soto, and Gomes present results of psycholinguistic picture matching tests and electrophysiological experiments run with Karajá and Brazilian Portuguese (BP) subjects. Results indicated that recursive PP constructions were more difficult to process than coordinated PP constructions across all languages tested. However, the on-line electrophysiological test suggested that once participants were engaged in the recursive algorithm, subsequent embeddings seemed to be facilitated. In PP embedding, a more salient ERP component was found 400ms after the first PP, probably connected in the recursion condition with integration efforts. In contrast, the N400 in coordination was less salient. Crucially, in comparison with the first embedded PP, the next embedded PP shows shorter latencies in all the relevant regions of interest, demonstrating that the processing of the PPs seems to be facilitated after one enters the recursive mode. Based on these results, the authors conclude that recursion can be viewed as the result of an algorithm that is costly to launch, but once established, does not pose increasing effort for the system. This study, together with the others in this part of the book, therefore establishes the existence of a ‘coordination default’ encountered across production, processing, and acquisition in the case of multiple PPs, but one that can nonetheless be overridden.
4 Avenues Opened Up By This Volume
The highly variable forms of embedding and self-embedding that recursion enables have a long history in linguistics and computer science. Phenomena like recursive clausal embedding, PP-embedding, possessive embedding, coordination, and compounding each have special properties, and they can be remarkably different across languages. Many forms of recursion (like possessor recursion and nominal compounding) are completely absent in whole language families. Each language can be expected to have its own acquisition path across a variety of types, both left- and right-branching. Every language studied in the world, as far as we know – including Pirahã – has some such examples, as this volume incontrovertibly reveals. The late Ken Hale, the father of theoretical fieldwork, once remarked that he “always looked for the most complex parts of a language first, because they displayed the central regularities of a language most clearly” (personal communication). Such a perspective justifies the notion that we need to identify where self-embedding recursive structures are for every language and for every child acquiring a new language. This book pursues that goal, alongside its own social goal: creating a community of scholars that can share ideas, methods, and particular experimental devices, as the collection of chapters on PP-recursion convincingly illustrates.
Recursion, as a formal mechanism, has been argued to permeate many cognitive spheres, from planning to discourse (Corballis Reference Corballis2007; Lobina and García-Albea Reference Lobina, García-Albea, Taatgen and van Rijn2009; Levinson Reference Levinson2013). In a way, it is a microscope that can be used across domains as a point of comparison because of its centrality to phrase-structure descriptions of languages, alongside at the same time its elusiveness in showing full-blown unrestrictedness. In a way, the fact that it is so easy to formally represent yet so difficult to encounter vindicates the hunch that its limitations come from many syntax-external systems. Tracking reference, second-order beliefs, speech act potentials, and morphological and prosodic integration across multiple dependent issues of the same kind of constituent presents challenges for a number of cognitive systems, the nature of which are much more complex than the format of rules like (1) themselves.
As such, it is an extremely useful guiding heuristic that leads us to the core of some of the contemporary questions in the language sciences: what is the relative weight of syntactically specific mechanisms as opposed to language-external systems in terms of an overall explanation of a given phenomenon? Although many studies have addressed this question with respect to the highly familiar territory of center-embedding in English, the experimental fieldwork and transversal comparisons conducted here represent relatively recent modes of exploration of recursion and its processing in unexpected corners of the grammar, and in less well-trodden corners of the world. The burgeoning literature on the computational and genetic underpinnings of recursion (e.g., Friederici et al. Reference Friederici, Bahlmann, Friedrich and Makuuchi2011) must be in dialogue with an understanding of the ubiquity and limits of its distribution. The investigation of recursion in unfamiliar and often isolated societies, whose grammars have emerged under little contact with the much-studied languages of developed societies, bear clear consequences for discussions of whether the regularities and nature of its patterning have biological, and not culture-dependent, origins.
There are many aspects in the study of recursion that we have not even touched upon here. Recent work in the semantics of attitude verbs by Kratzer (Reference Kratzer2013) has begun to question whether the semantic relation between an embedding verb and its propositional complement in sentences like John yelled that it was hailing are indeed direct function application, or whether they semantically look more like adjunction structures that are combined by predicate modification. At the same time, syntactic research by Aboh (Reference Aboh2004), Arsenijević (Reference Arsenijević2009), and Krapova (Reference Krapova2009) indicates that sentential complementation is achieved by the same syntactic means as relativization, mediated by a nominal complement. Both of these strands of research suggest that CP-recursion of the type in (1) may not always be the appropriate representation for sentential embedding. It is only through further crosstalk with ongoing work in semantics, and with comparative work with the morphosyntactic evidence for complementation-as-relativization, that work on the markedness of recursion may move forward in tandem. Similar work on ‘alternative’ strategies to compensate for PP and DP recursion reveal themselves in the evidence brought forth by processing and prosody studies presented herein.
In closing, beyond recursion and embedding themselves, it is our hope that the overall organization of a team of interacting researchers that put theory and data collection in dialogue within the reach of a single core computational topic with a complex set of interfaces can yield an overarching research strategy to be applied to the exploration of other recurrent, defining characteristics of grammar in experimental fieldwork across languages and their interfaces with processing and cognition (such as, say, the representation of nominal number, or logical connectives such as disjunction). We contend that the study of syntactic markedness as understood in formal, computational, and experimental approaches to grammar becomes more robust through the kind of emphasis adumbrated here, one that productively sums the contributions of each constituent of diverse comparative fieldwork across grammatical domains. And now, on to the chapters!
