2.1. PG morphophonology

PG is an agglutinative language; many different morphemes are expressed as prefixes or suffixes within the verbal complex (Ayala Reference Ayala1996; Hamidzadeh & Russell Reference Hamidzadeh, Russell, Weber and Chen2014; Zubizarreta Reference Zubizarreta2022). The full verbal complex template is shown in (2):

This order of elements in the verbal complex interacts with morphophonological processes in the language. The root and its prefixes form a prosodic word, which is the unit of nasalisation and stress assignment (Russell Reference Russell, Elkins, Hayes, Jo and Siah2022b), while suffixes constitute separate prosodic words of their own (Dbkowski Reference Dabkowski, Jurgec, Duncan, Elfner, Kang, Kochetov, O’Neill, Ozburn, Rice, Sanders, Schertz, Shaftoe and Sullivan2022).

2.1.1. Agreement

PG makes use of two sets of agreement morphology, which I term ‘Set A’ and ‘Set B’ here. In Table 1, I present the full independent form of each subject pronoun, followed by the corresponding Set A and B forms. Set A agreement has been previously analysed as true phi-agreement, as opposed to the Set B clitics (Woolford Reference Woolford, Legendre, Putnam, de Swart and Zaroukian2016; Zubizarreta & Pancheva Reference Zubizarreta and Pancheva2017).

Table 1 PG pronoun paradigm.

Set A agreement is used when agreement is controlled by the transitive subject (as in (3a)) or active intransitive subject (as in (3b)).

• 

Set B forms are used when agreement is controlled by the transitive object (as in (4a)) or stative intransitive subject (as in (4b)). The system of argument cross-reference in PG is a case of hierarchical agreement: subject and object cross-referencing morphemes compete for a single slot, and the winner is the form higher on the person hierarchy (Velázquez-Castillo Reference Velázquez-Castillo1991; Nichols Reference Nichols1992; Siewierska Reference Siewierska1996). Additionally, Set B forms are used in possessive contexts, as in (4c).

• 

3.1. Nasality

PG distinguishes six vowel qualities, as shown in Figure 1. Vowel length is not contrastive. All oral vowels have nasal counterparts.

Figure 1 Paraguayan Guaraní phonemic vowel inventory.

Vowel nasality is contrastive only if the vowel is stressed (Barratt Reference Barratt1981; Walker Reference Walker1999). Typically, the vowel which receives the primary stress within a lexical item is the root-final vowel (Adelaar Reference Adelaar1994; Cabral & Rodrigues Reference Cabral and Rodrigues2011; Mistieri Reference Mistieri2013). Preceding a stressed root-final nasal vowel, all previous vowels within the word also predictably surface as nasal. I interpret this generalisation as evidence that only a vowel which receives stress may be specified phonologically as either oral or nasal: all other vowels within the phonological word are underlyingly unspecified for nasality, and predictably surface as either nasal or oral due to the influence of the stressed vowel. Therefore, the minimal pair [tupa] ‘bed’ vs. [tũp] ‘God’ exists in the language, but there are no possible counterparts like *[tũpa] or *[tupã], in which nasality is contrastive on an unstressed, non–root-final vowel.

The inventory of consonants in PG, including those of ‘mixed’ articulation, is represented in Table 2. Notably, there are no voiced stops in the inventory. However, segments of ‘mixed’ articulation, in which a consonant begins as a nasal and ends as a voiced stop, are present in PG. The status of these ‘mixed’ segments has long been debated in the Tupí–Guaraní literature. Analyses vary between arguing that they are the surface result of the pre-nasalisation of oral stops (Gregores & Suárez Reference Gregores and Suárez1967; Rose Reference Rose2008; Daviet Reference Daviet2016) versus the post-oralisation of nasal consonants (Piggott Reference Piggott1992; Cardoso Reference Cardoso2009; Lapierre & Michael Reference Lapierre and Michael2018; Estigarribia Reference Estigarribia2021). I take the latter position, that these ‘mixed’ segments are post-oralised allophones of nasal consonants. The surface form of an underlying nasal consonant is determined by its immediate phonological environment: before an oral vowel, a nasal stop surfaces as its partially oralised allophone, as the result of shielding (Stanton Reference Stanton2017). Before a nasal vowel, a nasal consonant surfaces faithfully as nasal, while an underlying oral approximant surfaces as nasalised. Consonants alternate as in Table 3.

Table 2 Paraguayan Guaraní consonant inventory, including phonemes and mixed-articulation allophones.

Table 3 Allophonic consonant alternations before oral vs. nasal vowels.

3.2. Nasalisation

Both regressive (leftward) and progressive (rightward) spreads of nasalisation are attested in PG. I argue that these two types of nasalisation differ in important ways. In this study, I investigate what conditions variation in nasalisation, and how that variation is connected to the direction of nasalisation.

3.2.2. Progressive nasalisation

Nasalisation in PG also has a limited rightward spread. Following a stressed nasal vowel, the locative enclitic /=pe/ may undergo nasalisation to surface as [=mẽ], as illustrated in (12).

• 

While regressive nasalisation spreads throughout roots and all prefixes, progressive nasalisation occurs only with a very select group of targets (Lunt Reference Lunt and Anderson1973; Humbert & Piggott Reference Humbert, Piggott, Booij and van de Weijer1997). In his grammar of PG, Estigarribia (Reference Estigarribia2020) lists 11 suffixes and enclitics affected by progressive nasalisation: each morpheme has one allomorph following an oral vowel, and a different (nasal) allomorph following an oral vowel; these are shown in Table 4. I find that only five of these 11 suffixes productively undergo nasalisation for the speakers with whom I worked, listed as ‘productive’ in Table 4.Footnote ⁵ The productive morphemes include the totalitative (total), which indicates application and/or exhaustion of a given event to the whole object or subject (Estigarribia Reference Estigarribia2020); the locative (loc); an enclitic indicating ‘until’; the incipient (incip), representing the immediate future; and the nominal past (n.pst), which restricts the temporal interpretation of a noun to the past, similar to English ‘former’ (Tonhauser Reference Tonhauser2007). The unproductive morphemes include the collective (coll), indicating plurality of countable items; the collective plural (coll.pl), indicating a place of abundance; various passive nominalisers; and the enclitic meaning ‘towards’.

Table 4 Allomorphs of target suffixes in Paraguayan Guaraní.

While voiceless stops are transparent to regressive nasal harmony, progressive nasalisation exclusively targets morphemes with initial voiceless stops.Footnote ⁶ Some suffixes and enclitics undergo total nasalisation – for example, loc [pe] $\sim $ [mẽ] – while the initial voiceless stop of others pre-nasalises – for example, total [pa] $\sim $ [mba]. This distinction is attributable to historical origin (Russell Reference Russell, Marianne, Reisinger and Underhill2023): suffixes which pre-nasalise have earlier origins as roots,Footnote ⁷ and the pre-nasalisation of voiceless-stop-initial roots was a productive process in Proto-Tupí–Guaraní (Estigarribia Reference Estigarribia2021).

While regressive nasalisation may be triggered by either a phonemic nasal vowel or a consonant, the same is not true for progressive nasalisation. If a nasal consonant is followed by an oral vowel, and therefore surfaces as post-oralised, as in the root /h-enu/ ‘listen’ in (13), nasality cannot spread rightwards, as shown in (13b). Therefore, only stressed nasal vowels should be able to trigger progressive nasalisation in PG.

• 

Progressive nasalisation involves the rightward spread of nasality. The trigger is a phonemic nasal vowel, and the target is a suffix-initial syllable beginning with a voiceless stop belonging to one of the five productive suffixes listed in Table 4. The surface effect of progressive nasalisation differs based on the suffix: some suffixes show the effects of full nasalisation, while others pre-nasalise.

3.3. Interim summary

I have described two processes of nasalisation in PG. In Table 5, I compare several properties of these processes. Regressive harmony targets all preceding segments except for voiceless obstruents, which are transparent. Progressive nasalisation, on the other hand, exclusively targets syllables with initial voiceless stops. Additionally, regressive harmony appears to be fully productive, while progressive nasalisation is limited to a very small number of particular suffixes and enclitics, with morpheme-specific effects.

Table 5 Comparing directions of nasalisation in Paraguayan Guaraní.

I now turn to the interactions of these two nasalisation processes with roots of Spanish etymological origin. Russell (Reference Russell2022a) finds that the presence of a nasal consonant in a Spanish-origin root can trigger the nasalisation of prefixes. The observation that regressive nasal harmony also applies to some extent to Spanish-origin roots provides support for the claim that it is in fact a productive process in the language. However, the interactions of progressive nasalisation and Spanish-origin roots have gone unreported in prior literature; I seek to fill that gap here. Gaining insight into these interactions is crucial to an understanding of the differences between regressive and progressive nasalisation in PG.

4.1. Phonotactic adaptation

Individual Spanish-origin lexical items display various repairs of violations of PG phonotactics. Pinta & Smith (Reference Pinta, Smith, Estigarribia and Pinta2017) propose five lexical strata in PG, based on phonological repairs of loanwords from Spanish, as reproduced in Table 6. They identify four different properties: the presence of a nasal coda (N Codas), any coda consonant (Codas), non-final stress (Non-final stress and complex onsets (#CC). The more properties are repaired, the more native-like a stratum is, and conversely, the more properties are tolerated, the more foreign-like a stratum is. Pinta and Smith do not consider behaviour in contexts of nasalisation in their analysis of lexical strata.

Table 6 Lexical strata in Paraguayan Guaraní; reproduced from Pinta & Smith (Reference Pinta, Smith, Estigarribia and Pinta2017: 306).

I provide an example of a lexical item from each stratum in Table 7; examples are taken from Pinta & Smith’s (Reference Pinta, Smith, Estigarribia and Pinta2017) discussion of their proposed strata. In the nativ(ised) stratum, which includes native Guaraní items as well as fully nativised loans, all syllables are open, stress is final and no complex onsets occur, as visible from the adaptation of culantro ‘coriander’ to [kuɾãtũ]. In the mostly nativised stratum, all syllables are open and stress is final, but onset consonant clusters are tolerated, as in the initial syllable of ‘Pluto’. In the partially nativised stratum, all syllables are open, but non-final stress is tolerated, and so are onset consonant clusters, as evidenced by the adaptation of ‘London’ with the onset cluster in the second syllable intact, but the final coda /s/ deleted. In the barely nativised stratum, non-nasal codas are tolerated, as are non-final stress patterns and onset consonant clusters; however, nasal codas are repaired, as visible from the adaptation of the nucleus and coda of ‘department store’ as a nasal vowel. Finally, in the unadapted stratum, non-final stress and onset consonant clusters are tolerated, and codas of all kinds are tolerated, including nasal ones, as in the first syllable of ‘salad’.

Table 7 Examples of adaptations sorted by stratum in Paraguayan Guaraní.

The stratum into which a loan falls is partially a reflection of the time depth of the loan, à la Itô & Mester (Reference Itô, Mester and Tsujimura1999), but is mainly dependent on the loan’s phonotactic similarity to PG in its unadapted form. For instance, adaptation of the Spanish word papá ‘father’ into PG necessitates no overt repair of PG phonotactics, and would thereby be classified as Level 1 (‘nativ(ised)’). I acknowledge the limits of this system of lexical strata, and use it here simply as a proxy measure for quantifying the well-formedness of a Spanish-origin lexical item vis-à-vis PG phonotactics.

4.2. Nasality

Paraguayan Spanish has three phonemic nasal consonants (bilabial /m/, alveolar /n/ and palatal //), but no phonemic nasal vowels (Cassano Reference Cassano1971). It is not the case that vowels are never pronounced with phonetic nasalisation in Spanish; however, nasalisation is not a target of speech production in Spanish vowels, and nasalisation is not a phonologically active feature of vowels in the language (Solé Reference Solé1992). Spanish borrowings contain surface phonotactic environments that never occur natively in PG, since phonemic nasal consonants in PG always surface as partially oralised allophones before oral vowels. An underlying sequence /ma/ must be realised either as [mba] or as [mã] on the surface in PG due to phonotactic constraints in the language; however, lexical items of Spanish origin are not subject to the same phonotactic constraints. In some older borrowings, this mismatch in phonotactics is in fact repaired, as shown in Table 8. Some sequences of a vowel and nasal coda consonant in Spanish borrowings are reinterpreted in PG as phonemic nasal vowels. In words like ‘heart’, ‘pants’ and ‘soap’, a word-final VN sequence in Spanish is reinterpreted as a stressed nasal vowel in PG, and these words obey nasal harmony within the root. The word for ‘pillow’ demonstrates the outcome when a nasal consonant in a Spanish-origin root does not appear within the same syllable as the stressed vowel: this /m/ spreads its nasality leftwards, and surfaces as post-oralised [mb], exactly as expected given the phonotactic constraints of PG.

Table 8 Adaptations of selected Spanish-origin items with nasal consonants into Paraguayan Guaraní.

However, most Spanish-origin items are pronounced roughly as they are in Spanish: nasal–oral sequences are not repaired, and these items do not exhibit root-internal nasal harmony. The difference in strategy between adaptations to morpheme-internal nasal harmony and apparent non-adaptations is likely related to the diachronic expansion of PG–Spanish bilingualism in Paraguay. Spanish influence on PG took place slowly and gradually over a long period of time until quite recently, when urbanisation has accelerated the rate of Guaraní speakers acquiring and using Spanish (Zajiícová Reference Zajiícová2009; Fernández Barrera Reference Fernández Barrera2015). Since the majority of users of PG today are bilingual with Spanish, they seem to be more tolerant of violations of PG phonotactics by Spanish-origin items (Pinta Reference Pinta2013; Pinta & Smith Reference Pinta, Smith, Estigarribia and Pinta2017). Earlier borrowings, therefore, underwent nativisation more than recent borrowings, which maintain the phonology of Spanish more faithfully.

4.3. Nasalisation

The presence of a nasal consonant in a Spanish-origin root triggers nasalisation of prefixes (Thun Reference Thun2005; Russell Reference Russell2022a), even when the vowels and consonants in between are oral, as illustrated in (14). Examples below are presented using a four-line method, in which the top line is the assumed underlying representation, with Spanish-origin roots (in Spanish orthography) italicised, and underlying nasal consonants bolded. In the second line – the surface IPA representation – the targets of nasalisation are underlined.

The presence of a nasal consonant may also trigger nasalisation in suffixes, even when the vowels and consonants in between are oral, as in (15).

However, importantly, both consultants in an elicitation setting often expressed uncertainty regarding the acceptability of the combination of a Spanish-origin root and suffix nasalisation, as illustrated in (16).

Spanish-origin roots may participate in nasalisation of prefixes and suffixes, but not in root-internal harmony, with the exception of the few nativised borrowings. The interactions of Spanish-origin items with regressive nasal harmony have previously been analysed as the innovation of a novel system of consonant harmony, in which a nasal consonant in a Spanish-origin root is in correspondence with consonants in prefixes, resulting in nasalisation of those prefixal consonants (Russell Reference Russell2022a). Interactions of Spanish-origin roots and progressive nasalisation are still unclear at this point. Although both consultants produce and accept some forms in which a nasal consonant in a Spanish-origin root triggers progressive nasalisation in an elicitation setting, they also verbally express uncertainty about the acceptability of similar forms.

5.1. Methods

The data presented here come from a corpus of sociolinguistic interviews with native PG speakers in Paraguay (Bittar Prieto Reference Bittar Prieto2021). Twenty-six interviews were conducted; each interview lasted roughly an hour. Half of the interview data was collected in the urban Bañado Sur neighbourhood of Asunción in 2015. These conversational interviews were conducted by Israel Pedrozo, a resident of Bañado Sur and native speaker of PG and Paraguayan Spanish. The age of the participants in the urban community ranged from 18 to 68 (mean = 43.9, sd = 18.7). The other half of the data was collected in the rural community of San Juan Nepomuceno, roughly 200 kilometres from Asunción. The interviews in this community were conducted between October 2019 and January 2020 by Antonio Zena, a native of San Juan Nepomuceno. The ages of the participants in the rural area ranged from 18 to 73 (mean = 45.6, sd = 15.7). For each interview, all utterances were transcribed in PG and translated to Spanish by the interviewers themselves as well as by Josefina Bittar, a linguist and heritage speaker of PG.

All potential sites of nasalisation – both instances in which nasalisation of an affix actually occurred and those in which it could have but did not – were added to a data frame. The relevant variable context is the combination of an affix affected by nasalisation and a nasal root: either a PG-origin root containing a phonemic nasal segment, or a Spanish-origin root containing a nasal consonant. Data collection was limited to cases of certain affixes, specifically those affixes in which consonants clearly alternate between oral and nasal, listed in Table 9. Nasalisation of affixes that comprise only a single vowel, or a single vowel and an obstruent, was not included in the data set, due to the limitations of measuring nasality in these contexts without data about nasal airflow. Additionally, tokens were excluded when it was not possible to ascertain the etymological origin of the root (e.g., mamá ‘mother’).

Table 9 Targets of nasalisation included in the data set.

Though the nominal past tense suffix k^we had been found to productively nasalise in the elicitation context, there were no tokens of its nasal form in the corpus, so it was excluded from the data set. The corpus was also checked for the six target suffixes which had not been found to productively nasalise in the elicitation context (cf. Table 4): the collective ; the collective plural tɨ; the passive nominalisers pɨ, pɨ and ; and the enclitic ‘towards’ oto. The corpus included two tokens of the nasal allomorph of the collective both following the lexical item mĩtã ‘child’. These tokens were not included in the final data set. The corpus included no tokens of the collective plural or any of the passive nominalisers, and no tokens of the nasal allomorph of ‘towards’ .

The total number of tokens included in the data set was 3,641. Each token, defined as an affix that is a potential undergoer of nasalisation, was coded for the dependent variable – viz., whether nasalisation of an affix occurred – as well as a number of linguistic and social factors. Social factors included gender (self-identified as male or female), age and community affiliation (rural or urban). Linguistic factors included direction of nasalisation (regressive or progressive), morpheme identity of the target affix, etymological origin of the root (PG or Spanish), morphological category (adjective, noun or verb) and log-transformed root token frequency within the corpus.Footnote ⁸ Gender, age and community affiliation (rural or urban) were included in order to investigate whether social factors have any significant impact on application of nasalisation. Older people and rural communities are often associated with a form of PG which is considered more ‘pure’, while young people and urban communities may be associated with bilingualism and language mixing (Rubin Reference Rubin1968; Gómez-Rendón Reference Gómez-Rendón, Matras and Sakel2007). The direction of nasalisation was identified as either regressive (leftward) or progressive (rightward). I have argued that the two types of nasalisation differ in important ways, and therefore hypothesise that they will show significantly different patterns. Several characteristics of the root were identified and included in the analysis: etymological origin (PG or Spanish), morphological category and log-transformed token frequency.

Tokens were also coded for three additional factors which are relevant only for Spanish-origin items: phonotactic well-formedness of the root with respect to PG phonotactics (measured from 1 to 5 as per Pinta & Smith’s classification), distance (in segments) between the trigger and target, and whether or not the nasal consonant trigger appears in the stressed syllable. Phonotactic well-formedness (lexical stratum) is relevant only for Spanish-origin items, as all PG-origin items in the corpus belong to the native stratum. If the phonological similarity of a Spanish-origin root to native PG items is a factor in predicting its rate of nasalising affixes, lexical stratum is expected to have a significant effect. The distance between the trigger and target was measured in terms of number of intervening segments, in order to investigate if distance has a significant impact on application of nasalisation. Since oral consonants and vowels may intervene between a trigger in a Spanish-origin root and a target, this distance ranges in the corpus from zero to ten segments. In every case of a PG-origin item triggering nasalisation, though, the distance is zero. Finally, stress was included as a factor in order to assess if the relationship between stress and nasality in PG holds for Spanish-origin items as well: namely, that nasality is contrastive only on a stressed vowel, as described in §3.1. Stress was coded as binary: Yes, if the nasal consonant trigger in a root appears within the stressed syllable, and No if the nasal consonant trigger is in an unstressed syllable. Because of PG-internal phonotactics, regressive nasal harmony can be triggered by either a nasal consonant or a stressed nasal vowel. If a nasal consonant in a Spanish-origin root behaves like a phonemic nasal consonant in a PG-origin root, it is predicted to be able to trigger regressive nasal harmony. However, progressive nasalisation may only be triggered by a stressed nasal vowel, never by a phonemic nasal consonant (cf. (13)). Therefore, every trigger of suffix nasalisation is predicted to be a nasal vowel in a stressed syllable. This factor was included in order to evaluate whether a root in which a nasal consonant trigger appears within a stressed syllable nasalises suffixes at a significantly higher rate than those in which the trigger of nasalisation appears in an unstressed syllable.

5.2. Findings

As indicated in Table 10, the corpus includes more tokens of regressive nasal harmony than progressive nasalisation. Tokens of PG-origin triggers of nasal harmony slightly outnumber Spanish-origin triggers. However, Spanish-origin roots appear in contexts of progressive nasalisation more frequently than PG-origin roots. This is likely attributable to the high frequency of suffixes attaching to proper nouns like place names, which are often of Spanish origin. The rate of nasalisation triggered by PG-origin roots is over 90% for both directions. With Spanish-origin roots, however, we see a drastic difference in nasalisation rate between regressive harmony, where nasalisation occurs 77.6% of the time, and progressive nasalisation, where it occurs with a mere 5.7% of the tokens. It is clear that a more in-depth investigation of the difference in nasalisation rates along the axis of direction, as well as by etymological origin of the root, is necessary.

Table 10 Distribution of tokens by direction and etymological origin.

The nasalisation rates for specific affixes, shown in Table 11, prove to be particularly enlightening.Footnote ⁹ A PG-origin root triggers nasalisation of most prefixes over 90% of the time, with three notable exceptions, namely the three Set B morphemes (i.e., the forms used when the transitive object or stative intransitive subject controls agreement). The lower nasalisation rate of these morphemes may reflect their distinct morphological status, as they have been analysed as proclitics (Woolford Reference Woolford, Legendre, Putnam, de Swart and Zaroukian2016; Zubizarreta & Pancheva Reference Zubizarreta and Pancheva2017). The rate of prefix nasalisation triggered by Spanish-origin roots is generally slightly lower, but consistently above 85%. With the three Set B morphemes, however, nasalisation rates are again much lower. When it comes to suffixes and enclitics, we see differences between individual morphemes. The large gap between the relatively high rates of nasalisation following a PG-origin root and the quite low rates of nasalisation following a Spanish-origin root is apparent.

Table 11 Distribution of tokens by affix and root etymological origin.

5.2.1. Regressive nasal harmony

The total number of tokens included in the data set of regressive harmony was 2,747 (1,575 tokens preceding roots of PG origin, plus 1,172 tokens preceding roots of Spanish origin). The variation in regressive nasal harmony was statistically modelled using mixed-effects logistic regression.Footnote ¹⁰ The model included seven fixed effects – gender, age, community affiliation, target affix identity, root etymological origin, morphological category and frequency – and a by-speaker random intercept. The three factors specified in §5.1 that are relevant only for Spanish-origin items (lexical stratum of the root, distance between trigger and target and whether or not the nasal consonant trigger appears in the stressed syllable) were not included in this model due to the asymmetry in the data. The model also included an interaction term between the target of harmony and the etymological origin of the root. None of the social factors significantly improved model fit at the threshold of $p<0.05$ . Additionally, morphological category had no significant impact on model fit. Though root origin was not significant as a main effect, it is significant in its interactions with target identity. The non-significant predictors were removed, with the exception of those involved in significant interactions, resulting in the final model in Table 12. The marginal $r^2$ of this model is 0.348, representing the variance explained solely by the fixed effects, and the conditional $r^2$ is 0.363, representing the variance explained by the entire model. In the model summary, a positive estimate indicates that the given factor level favours the nasal variant compared to the baseline, while a negative estimate indicates that the level favours the oral variant. Plots are provided in Figures B1 and B2 in Appendix B.

Table 12 Summary of model of regressive harmony.

Set B morphemes nasalise at a significantly lower rate than other morphemes. Root token frequency is significant in the model of regressive nasal harmony: the more frequent a root within the corpus, the higher the predicted rate of nasalisation. Etymological origin of the root is significant only in its interaction with two of the three Set B morphemes.

Variation in regressive nasal harmony triggered by only PG-origin items was modelled using mixed-effects logistic regression to isolate the factors relevant for those items. The total number of tokens included in this data set was 1,575. This model included six fixed effects – gender, age, community affiliation, target identity, morphological category and root token frequency – as well as a by-speaker random intercept. Again, none of the social factors significantly improved model fit at the threshold of $p<0.05$ . Morphological category and root frequency also had no significant impact on model fit. The non-significant predictors were removed, resulting in the final model in Table 13. The marginal $r^2$ of this model is 0.233, and the conditional $r^2$ is 0.240. The only significant predictor was the target identity of the morpheme: specifically, Set B morphemes nasalise at a significantly lower rate than other morphemes.

Table 13 Summary of model of regressive harmony triggered by PG-origin items.

Variation in regressive nasal harmony triggered by only Spanish-origin items was modelled using mixed-effects logistic regression to isolate the factors relevant for those items. The total number of tokens included in this data set was 1,172. This model included nine fixed effects – gender, age, community affiliation, target identity, morphological category, frequency and two additional factors: root lexical stratum and distance – as well as a by-speaker random intercept, and an interaction between the target of harmony and the lexical stratum of the root. Again, none of the social factors significantly improved model fit at the threshold of $p<0.05$ . Morphological category also had no significant impact on model fit. The lexical stratum of the root was not significant, either as a main effect or in an interaction with the target morpheme identity. The non-significant predictors were removed, resulting in the final model in Table 14. The marginal $r^2$ of this model is 0.398, and the conditional $r^2$ is 0.422. Plots are provided in Figure B3 in Appendix B.

Table 14 Summary of model of regressive harmony triggered by Spanish-origin items.

In summary, Set B agreement morphemes nasalise at a significantly lower rate than all other morphemes. The token frequency of the trigger of nasal harmony significantly positively correlates with nasalisation rate. Spanish-origin roots do not trigger nasalisation at a significantly lower rate than PG-origin roots; however, root etymological origin is significant in its interaction with certain Set B morphemes. In such cases, Spanish-origin roots trigger nasalisation of Set B morphemes at a lower rate. When looking specifically at Spanish-origin roots, the lexical stratum of the root does not significantly affect the rate of nasalisation. An increase in distance between trigger and target significantly negatively correlates with nasalisation rate.

5.2.2. Progressive nasalisation

In statistical analysis of the progressive nasalisation data, the incip suffix was excluded from the model due to the low number of available tokens ( $n=7$ ). The total number of tokens included in the data set of progressive nasalisation was 880. The variation in progressive nasalisation was statistically modelled using mixed-effects logistic regression, including seven fixed effects – gender, age, community affiliation, target affix identity, root etymological origin, morphological category and frequency – and a by-speaker random intercept. The model also included an interaction term between the target morpheme identity and the etymological origin of the root. Again, none of the social factors significantly improved model fit. Additionally, neither the morphological category of the root nor root token frequency had a significant impact on model fit. The non-significant predictors were removed, resulting in the final model in Table 15. The marginal $r^2$ of this model is 0.779 and the conditional $r^2$ is 0.793. A plot is provided in Figure B4 in Appendix B.

Table 15 Summary of model of progressive nasalisation.

All three morphemes included in this model display distinctly different patterns of nasalisation. Spanish-origin roots are predicted to trigger nasalisation at a lower rate than PG-origin roots across the board; however, the size of this effect is specific to each morpheme.

Variation in progressive nasal harmony triggered by only PG-origin items was modelled using mixed-effects logistic regression to isolate the factors relevant for those items. The total number of tokens included in this data set was 389. This model included six fixed effects – gender, age, community affiliation, target identity, morphological category and root token frequency – as well as a by-speaker random intercept and an interaction term between target identity and root frequency. Again, none of the social factors significantly improved model fit at the threshold of $p<0.05$ . Morphological category also had no significant impact on model fit. The non-significant predictors were removed, resulting in the final model in Table 16. The marginal $r^2$ of this model is 0.228, and the conditional $r^2$ is 0.300.

Table 16 Summary of model of progressive nasalisation with PG-origin roots.

When including only tokens of progressive nasalisation following a Spanish-origin root, the total number of tokens in this data set was 497. This model included nine fixed effects – gender, age, community affiliation, target affix identity, morphological category, frequency, root lexical stratum, stress and distance – and a by-speaker random intercept, as well as an interaction term between lexical stratum and affix morpheme identity. Again, none of the social factors significantly improved model fit, and neither the morphological category of the root nor frequency had a significant impact on model fit. The non-significant predictors were removed, resulting in the final model in Table 17. (Plots are provided in Figures B5 and B6 in Appendix B.) The marginal $r^2$ of this model is 0.376, and the conditional $r^2$ is 0.437. The probability of the nasal variant is predicted to be above zero only for those lexical items which most closely resemble PG-origin items in terms of lexical stratum, distance between target and trigger of nasalisation, and stress. Specifically, the Spanish-origin items which most closely resemble PG-origin items are those which include few, if any, violations of PG phonotactics, bear stress on a syllable containing a nasal consonant, and have little to no intervening material between that syllable and a prefix.

Table 17 Summary of model of progressive nasalisation triggered by Spanish-origin items.

Looking at all contexts of progressive nasalisation, all morphemes nasalise at significantly different rates from one another, and Spanish-origin roots undergo nasalisation at a significantly lower rate than PG-origin roots. The interaction between root origin and morpheme identity is significant for all morphemes. When we consider only nasalisation triggered by Spanish-origin items, lexical stratum has a significant impact: the more phonotactically well-formed a root is in PG, the higher the rate of nasalisation. In fact, only those Spanish-origin roots which share their phonological properties with PG-origin roots are predicted to trigger progressive nasalisation at all. Roots in which the trigger of nasalisation appears within a stressed syllable nasalise at a significantly higher rate than those in which the trigger is in an unstressed syllable, signifying that the language-specific connection between stress and nasality in PG is maintained for Spanish-origin roots as well. An increase in distance between trigger and target significantly negatively correlates with nasalisation rate.