2.1 1930s Nishnaabemwin

This section illustrates key aspects of Nishnaabemwin as it was spoken in the late 1930s. Crucially, adult speakers in the 1930s had phonological representations with unstressed vowels, but reduced them phonetically. Furthermore, though the severe reduction was not quite total deletion, it was liable to be perceived as such, and the subsequent generation evidently did perceive reduction as deletion.

The precursor to rhythmic syncope, the severe phonetic reduction of unstressed vowels, was first documented for Nishnaabemwin by Bloomfield in the late 1930s. Bloomfield (Reference Bloomfield1957: 5) reports that ‘the vowels of … odd-numbered’ light syllables were reduced, where the count goes from left to right, restarts after long vowels and ignores the final syllable. The alternating reduction matches the uncontroversial iambic stress pattern common in Ojibwe dialects (Valentine Reference Valentine1994), which places odd-numbered light syllables in the weak branch of the foot. Specifically, the typical Ojibwe dialect has iteratively assigned left-aligned iambs, with otherwise unparsable final syllables placed in a degenerate foot. While syllables with long vowels are always heavy, closed syllables are not. With reduction grafted to this stress pattern, a word like the name for the language, which had been [(ʌˈnɪ)(ʃɪˈnaː)(ˈbεː)(mʊˈwɪn)], was now pronounced as [(^əˈnɪ)(ʃ^əˈnaː)(ˈbεː)(m^əˈwɪn)].

Importantly, person prefixes were part of the domain for footing. This gave rise to paradigmatic alternations, which provided evidence of an active reduction-cum-deletion process, as shown in (1). Note that phonological mappings are presented in rule-based derivations because they are a compact and comprehensible way to illustrate phonological mappings, not because derivations are assumed to be theoretically desirable (see Bowers & Hao Reference Bowers and Hao2018 for a formal characterisation of alternating vowel deletion). To clarify which stage of Nishnaabemwin is under discussion, examples are labelled either 1930s or Modern.Footnote ¹

1. (1)

   

Such alternations were quite well attested, with approximately 40% of the words listed in Rhodes (Reference Rhodes1985a) beginning historically with a light syllable, and around 25% with more than one light syllable. Similar figures obtain for more recent dictionaries of Nishnaabemwin (Corbiere & Valentine Reference Corbiere and Randolph Valentine2016), as well as dictionaries of non-syncopated dialects of Ojibwe (Nichols & Nyholm Reference Nichols and Nyholm1995, Nichols Reference Nichols2015).

2.2 Modern Nishnaabemwin (post-1930s)

Reports written after Bloomfield (Reference Bloomfield1957) indicate that Nishnaabemwin has substantially changed. Rhodes (Reference Rhodes1985b: 467) reports the rise of an ‘innovative’ dialect that utilises new prefixes and that has levelled out the alternations found in 1930s Nishnaabemwin (see also Kaye Reference Kaye1974, Rhodes Reference Rhodes and Cowan1975 and Piggott Reference Piggott1980, discussed in §2.3). Recall from (1) that ‘kick someone’ alternated between [d^əŋgɪʃk^əw-aː-t] ‘if he kicks him’ and [n^ədʌŋg^əʃkʌw-aː] ‘I kick him’. In the innovative dialect, ‘I kick him’ is [ndoː-dŋgɪʃkw-aː], with a novel prefix and vocalism for a prefixed form. Strikingly, with this innovation, the paradigm has lost the 1930s rhythmic syncope alternations, since ‘if he kicks him’ is still [dŋgɪʃkw-aː-t]. This combination of levelled paradigms with innovative prefixes is the ‘regular’ inflectional pattern in the innovative dialect (Rhodes Reference Rhodes1985b: 466), as is robustly reflected in the dictionary published in the same year (Rhodes Reference Rhodes1985a), from which some representative forms are given in (4).Footnote ²

1. (4)

   

The innovative prefixation strategy is slightly more complicated, as three prefix allomorphs have emerged (all three prefix allomorphs are found with the words in (4), but only one is shown, for brevity). In addition to [ndoː-], 1st person may also be realised as [ndɪ-] or [ndʌ-]. According to Rhodes (Reference Rhodes1985b: 467), [ndoː-] and [ndʌ-] predominate in Odawa communities, while [ndɪ-] is favoured in Eastern Ojibwe.

The adoption of a consistent vocalism in place of the 1930s alternations is a sharp departure from the expected alternating deletion pattern discussed in §2.1.3. The origin of the consistent vocalism does not warrant discussion here, as it can be straightforwardly traced from the unprefixed allomorph used in the 1930s (see Bowers Reference Bowers2015). In contrast, at first blush the innovative prefixes seem to have no precedent from the 1930s, since the syncopated realisation for the prefix [nɪ-] is simply [n-]. The following section discusses the likely origin of the Modern Nishnaabemwin prefix system.

2.3 Restructuring coincided with deletion interpretation

Thus far it has been shown that adult speakers of 1930s Nishnaabemwin had extreme phonetic reduction of unstressed vowels (§2.1), that reduction was reinterpretable as deletion (§2.1.2) and that Nishnaabemwin now contains innovative prefixes and levelled paradigms (§2.2). This section reviews evidence that restructuring was initiated at the same time that severe reduction was documented, and that restructuring went hand in hand with the reinterpretation of reduction as deletion. §2.3.1 discusses evidence that restructuring may have begun slightly before the reinterpretation of reduction as deletion, while §2.4 discusses claims that restructuring may not have been adopted by all members of the cohort.

The first reports after Bloomfield (Reference Bloomfield1957) contain brief indications that a major change was underway in Nishnaabemwin. A remarkably precise date for the initiation of restructuring can be inferred from Piggott (Reference Piggott1980: 2), who writes that ‘the elaborate inflectional system of Ojibwa [Nishnaabemwin] which is well attested in the grammar of older speakers has undergone (or, rather, is undergoing) considerable modification in the grammar of speakers who are in the mid-thirties and under’. The changes Piggott mentioned most likely include the prefix reanalysis and stem levelling discussed above (Glyne Piggott, personal communication). This is buttressed by the fact that Rhodes (Reference Rhodes1985b) and Valentine (Reference Valentine1994, Reference Valentine2001) identify only the novel prefixation system plus levelling as the main morphological innovations in Nishnaabemwin.

The age of the leading edge of the cohort suggests that the 1930s pattern began to be abandoned towards the end of the decade. Piggott (Reference Piggott1980) describes field data from 1968 to 1974. Importantly, the speakers who were in their mid-thirties and younger between 1968 and 1974 were children during the period of extreme reduction documented by Bloomfield in 1938.

Where Piggott suggests a concrete date for the initiation of restructuring, Rhodes (Reference Rhodes and Cowan1975, Reference Rhodes1985b) affirms that the prior generation displayed severe reduction, and that the restructuring cohort reinterpreted this as deletion. Rhodes (Reference Rhodes and Cowan1975: 130) notes that, for older speakers, ‘deletion seems to be a … casual speech phenomenon’, echoing Bloomfield's description of a low-level reduction process. Rhodes further reports that older speakers had an intuitive awareness of absent vowels, stating that ‘the vowels can be easily resupplied (with only some ambiguity of quality)’. This contrasts with younger speakers, for whom absent vowels are ‘totally abstract’, since ‘they cannot resupply the [missing] vowels except by reference to appropriate allomorphs’ (Reference Rhodes and Cowan1975: 130). Such distinct types of awareness should match distinct linguistic characterisations, like the difference between a gradient phonetic process and a categorical one.

Somewhat later, Rhodes (Reference Rhodes1985b: 464–466) provides concrete evidence that missing vowels are truly absent for younger speakers. Recall from §2.1.2 that in 1930s Nishnaabemwin severely reduced vowels could still surface as short vocoids. In Modern Nishnaabemwin, short vocoid segments do not correspond to the reduced vowels described by Bloomfield. That is, both [n^ʊwεː] ‘he speaks’ and [aːn^ʊwɪ] ‘despite’ have the same vocoid, but the vocoid is innovative in [n^ʊwεː] (cf. Western Ojibwe [ɪnwεː]; Nichols Reference Nichols2015), while the vocoid corresponds to a historical vowel in [aːn^ʊwɪ] ‘despite’ (cf. Western Ojibwe [aːnʌwɪ]; Nichols Reference Nichols2015).

Rhodes further shows that the presence and quality of excrescent vocoids can be predicted by the identity of flanking consonants. This is standardly interpreted as evidence that the vocoids are the result of phonetic interpolation between CC transitions. Interpolation between consonants presupposes that the consonants are adjacent prior to phonetic implementation, even if they were historically separated by a vowel. The neutralisation of the distinction between originally adjacent and formerly separated consonants provides further evidence that etymological vowels have been categorically removed in Modern Nishnaabemwin. On a similar note, Rhodes (Reference Rhodes1985b: 468–469) and Valentine (Reference Valentine2001: 73–89) enumerate various neutralisations in consonant clusters that are derived by reduction-cum-syncope, which presupposes that formerly separated consonants are now adjacent phonologically. Importantly, these data diagnose only categorical absence, but leave unresolved the key question of whether the absent vowels were removed by synchronic rhythmic syncope or were lost outright from underlying representations (see §7).

Hence the contemporary reports suggest that the recent development of Nishnaabemwin is divided into two stages: a severe reduction stage in the 1930s, and a subsequent restructured stage. Furthermore, the onset of restructuring coincided with the interpretation of reduction as deletion.

2.4 Possible heterogeneity in the restructuring cohort

While the early reports established that restructuring emerged with the reinterpretation of reduction as deletion, later work indicates it may not have been the only analysis. Valentine (Reference Valentine1994: 150) stops short of claiming that all younger speakers restructured the phonology, affirming only that restructuring has occurred ‘at least for some speakers’. Sometime later, he substantiates the implied restriction by limiting the restructuring to younger and/or less proficient speakers (Reference Valentine2001: 65–67). The limitation to younger speakers is certainly consistent with the reports from the 1970s, though the original 1970s cohort would have been around 60 years old in 2001, and thus unlikely to qualify as young. Note, however, that even if Valentine is referring to a still younger cohort, it is unlikely that they would have been much younger than 30, since intergenerational transmission in even the most robust Nishnaabemwin-speaking communities ceased in the 1960s.Footnote ⁶

More importantly, Valentine (Reference Valentine2001: 66–67, citing unpublished pedagogical materials by Mary Ann Corbiere) indicates that restructuring is limited to less proficient or even attrited speakers, and that highly proficient speakers continue to use the conservative 1930s system. Such highly proficient conservative speakers may even include members of the restructuring cohort (Rand Valentine, personal communication). However, Valentine (Reference Valentine2001: 66) acknowledges that there has been no systematic study of how widespread innovative prefixes and levelled paradigms are in Modern Nishnaabemwin.

The current descriptive literature thus does not provide a clear picture of learners’ responses to perceived rhythmic syncope. While the earliest reports suggest a uniform rejection of the rhythmic syncope alternations, sufficiently expert speakers may prefer the conservative pattern, or perhaps even use it exclusively. Which scenario is actually attested has significant potential implications for the learnability of rhythmic syncope. If all learners failed to acquire the original rhythmic syncope pattern, a rhythmic syncope grammar may be highly dispreferred, or perhaps even impossible for learners. On the other hand, if at least some learners succeed in acquiring rhythmic syncope, a rhythmic syncope grammar must be attainable at least some of the time. If those conservative speakers are generally non-attrited, whether a rhythmic syncope grammar is acquired may depend on the amount of exposure to the pattern.

The surveys in this study seek to address these questions. The first two surveys measure speakers’ intuitions regarding prefix choice (innovative vs. conservative) on stems that either did not (Survey 1) or did (Survey 2) alternate in 1930s Nishnaabemwin. The key concern is whether some or all speakers will show a preference for conservative 1930s patterns, and thus demonstrate that rhythmic syncope is acquirable. The third survey directly tests the theory that the restructuring was due to insufficient exposure to 1930s forms by measuring the accuracy with which speakers differentiate 1930s forms from non-existent forms, henceforth ‘foils’. Moreover, the participant sample deliberately included a range of proficiencies within the target cohort, so that any effect of proficiency on conservatism could be measured.

2.5 Further desiderata

The descriptive literature is incomplete in other respects as well. Rhodes (Reference Rhodes1985b: 466) claims that new prefixes did not spread across the entire lexicon, but that the new prefixes are limited to ‘exactly’ stems that had vowel–zero alternations after the first stem consonant in 1930s Nishnaabemwin. That is, words that had no left-edge alternations under prefixation in the 1930s, as in [n-gaːsknʊs] ‘I whisper’ (unprefixed form [gaːsknʊzʊ-t] ‘if he whispers’), or words where the only alternation could be attributed to prefix allomorphy, as in [nd-ʌgoːʤɪn] ‘I hang’ (unprefixed form [goːʤɪn]), reportedly take only etymologically correct prefixes. In contrast, words that alternated after the first stem consonant under prefixation, as in [n-mɪskwiːwnɪkεː] ‘I have a bleeding arm’ (unprefixed form [mskwiːwnɪkεː] ‘he has a bleeding arm’), do take the innovative prefixation pattern, as in [ndʌ-mskwiːwnɪkεː] (Rhodes Reference Rhodes1985b: 467). This pattern of prefixation is fairly robust in Rhodes (Reference Rhodes1985a), where non-alternating stems are often not listed with innovative prefixes. However, the force of the claim that new prefixes are limited to historically alternating stems is vitiated by Rhodes’ (Reference Rhodes1985b: 467) observation that for ‘these same speakers’ who have innovated Modern Nishnaabemwin, novel prefixation has spread to many formerly non-alternating words. This can be seen in Modern Nishnaabemwin [zaːmsɪniː] ‘he overeats’ and [ndʌ-zaːmsɪniː] ‘I overeat’ (1930s Nishnaabemwin prefixed form [nd-oːzaːmsɪniː]).

Finally, according to the description of the data in Rhodes (Reference Rhodes1985a: xx), new prefixes may appear on verbs, but not on nouns. The validity of this is unclear, as Valentine (Reference Valentine1994: 146–147) adduces nouns with innovative prefixes. Furthermore, there were editorial considerations that may have resulted in the suppression of innovative prefixes on nouns in Rhodes’ dictionary (Richard Rhodes, personal communication).

In sum, within the broader goal of measuring the degree of innovation after incipient rhythmic deletion, this study seeks to address the empirical goals in (15).

1. (15)

   

Foreshadowing the main results, we can say that Modern Nishnaabemwin was uniformly adopted by participants of all ages and proficiencies, while innovative prefixation occurred whether or not stems historically alternated, and whether or not stems were nouns or verbs. Furthermore, in the final survey participants reliably distinguished historical forms from foils. These results suggest that, despite substantial exposure to the historical deletion pattern, the cohort in question uniformly converged on a changed language.

3.1 Speaker proficiency and word familiarity

In this study, it is especially important to assess the effects of speaker proficiency and word frequency. Speaker proficiency must be tracked so that we can (a) ensure that the sample of participants was not mistakenly filled with heavily attrited individuals, and (b) establish whether proficiency is significantly correlated with a speakers’ use of conservative forms. Word frequency is a perennial variable of interest in both synchronic and diachronic studies (see for instance Bybee Reference Bybee2001, Hay et al. Reference Hay, Pierrehumbert, Walker and LaShell2015), and in this study we would expect highly frequent words to correlate with a higher degree of conservatism. Unfortunately, standardised proficiency tests and large text databases for calculating frequency do not exist for Nishnaabemwin. However, the design of the experiment allows vocabulary size and word familiarity to be estimated. These estimates are proxies for proficiency and frequency.

Word familiarity was calculated by dividing the number of participants who recognised a word by the total number of participants who were shown that word. Because words were drawn randomly in order to sample more of the lexicon, an individual word was typically shown only to at most a handful of participants. To avoid claiming that a word is universally recognised when it had been shown to few participants, familiarity estimates were Laplace smoothed, with a constant value of 2. Thus, if a word had only been shown to and recognised by a single participant, this measure estimates that one third of speakers would recognise it, while a word that was shown to and recognised by twenty participants is estimated to be recognised by (20 / (20 + 2)) = 91% of speakers.Footnote ⁸

To measure speaker proficiency, the percentage of words recognised by each speaker across all surveys was calculated. On average, speakers recognised 70.8% of words (standard deviation 13.7%). This rate of vocabulary recognition is comparable to that of English-speaking adults (Brysbaert et al. Reference Brysbaert, Stevens, Mandera and Keuleers2016).Footnote ⁹ Note that participant age is not correlated with proficiency (r ² = 0.002), although the correlation does improve substantially if the two participants over 80 years old are excluded (r ² = 0.169).

4.1 Results of Survey 1

The prefix survey results strongly indicate that [ndʌ-] and [ndoː-] are the default allomorphs in Modern Nishnaabemwin. The results also indicate that historical forms have not been entirely supplanted by default forms. To see the interweaving of historical attestation with regularisation, consider the data in Table II, which summarises the results of the forced-choice component of the survey. Each cell contains the percentage of times a prefix was selected for a category (column).

Table II

Percentage of times speakers selected a prefix–stem combination as the best option, within historical classes (columns). Values for historically attested prefix–stem combinations are in bold.

Recall that Rhodes (Reference Rhodes1985b) and Valentine (Reference Valentine2001) raise the possibility that speakers are conservative with respect to non-alternating words, and will consistently select the historical prefixed form. The effect of historical attestation can be most clearly seen by reading across the rows, where each prefix stands the best chance of being selected with its historical host, i.e. the bolded cell is the largest in each row.

Despite this boost, the historical prefixes are not the clear favourites. Within each class of stem (the columns in Table II), the historical form is the most commonly selected form only for the recut allomorphs ([ndʌ-, ndoː-, ndɪ-]), and only [ndoː-] is selected more than 50% of the time. That is, the bolded cell is not categorically selected; it is in fact rarely the majority of its column, and it is not even consistently the most popular option.

Rather than providing evidence of a conservative system, the forced-choice responses point to strong regularisation, as at least one of the top two prefix allomorphs in every column is [ndʌ-] or [ndoː-]. Furthermore, in three out of the five stem categories (ʌC, ʊC, VV), both [ndʌ-] and [ndoː-] are the top two prefix allomorphs, together making up 74% to 50% of selected forms in any category.

Recall that Valentine (Reference Valentine2001) indicates that more skilled Nishnaabemwin speakers use conservative 1930s prefixation. The measure of speaker proficiency (the percentage of items recognised in all surveys) is not correlated with the rate at which speakers select historical forms (r ² = 0.006), as shown in Fig. 1a. Perhaps more importantly, speakers generally select historical forms only around 40% of the time, while only five speakers select the historical form 50% or more of the time, and only one speaker selects historical forms over 75% of the time. Contrary to previous suggestions, none of the participants employs 1930s Nishnaabemwin prefix allomorphy.

Figure 1

Plots of prefix survey results: (a) speaker-by-speaker plot of percentage of historical forms selected, by proficiency score; (b) speaker-by-speaker plot of rating given to historical forms, by proficiency score. Dark dots indicate speaker means; light dots show raw ratings (jittered).

This is not to say that more proficient speakers are insensitive to historical attestation. The rating data (Fig. 1b) shows that more proficient speakers give modestly higher ratings to historical forms than less proficient speakers do.

4.1.3 Model criticism

The fixed effects of the maximal model of the data are shown in Table III. Predictors that are not present in the best model are italicised, and are displayed in order to facilitate comparison between surveys. Examination of the estimates for the fixed effects emphasises that innovative prefixation in Modern Nishnaabemwin is now widely preferred, because effects favouring conservative prefixation are swamped by those indicating prefix regularisation. Four prefixes are shown in Table III; the fifth, [n-], is the reference level included in the intercept.

Table III

Fixed-effects table for maximal model of prefix-rating results. Italicised factors are excluded in the best-fitting model.

As in the forced-choice data, the rating data clearly shows that [ndʌ-] and [ndoː-] are generally favoured, as their ratings are respectively 1.4 and 1.6 points higher than the intercept value. The remaining allomorphs have smaller, non-reliable effects (absolute t-values greater than 2 indicate reliable effects). The non-default allomorphs [n-, nd-, ndɪ-] only approximate the high ratings of [ndʌ-] or [ndoː-] if they are the historical allomorph, the speaker is highly proficient and the word is widely familiar.

To be specific, the model estimates that non-default prefixes on their historical host words are rated 0.4 to 1.6 points lower than [ndʌ-] and [ndoː-], all else being equal. In order to gain even an additional 0.4 points, the effects of proficiency (β̂ = 0.37) and familiarity (β̂ = ―0.26), and their interactions with historical attestation (β̂ = 0.19 and 0.40 respectively) must be included.Footnote ¹³ Only if a word is one standard deviation more familiar than average and a speaker is one standard deviation more proficient than average do the interaction terms contribute approximately 0.4 rating points. While highly proficient speakers are clearly more sensitive to historical attestation, they do not routinely prefer historical forms. Rather, the historical form is at best favoured to roughly the same extent as the modern regularised forms.

5.1 Results of Survey 2

The levelling survey results affirm that historically alternating stems now take the innovative prefixes, as suggested by Rhodes (Reference Rhodes and Cowan1975, Reference Rhodes1985a, Reference Rhodesb) and Piggott (Reference Piggott1980). Participants overwhelmingly selected levelled forms (e.g. [ndoː-mkɪzɪn]) over alternating forms ([n-mʌkzɪn]). Contrary to what Valentine (Reference Valentine2001) reports, highly proficient speakers did not reliably select historical forms, as seen in Fig. 2a.Footnote ¹⁴ Furthermore, the two speakers who selected historical forms nearly three-quarters of the time also chose foil items at a comparable rate, suggesting that they had adopted a strategy of merely selecting conservative-looking forms, regardless of whether these were actually found in the language. Increased word familiarity increased the likelihood of the historical form being selected, as can be seen in Fig. 2c. Figure 2b illustrates that, contrary to the prefix survey, older speakers chose historical forms less often than younger speakers. Finally, in a striking result, Fig. 2d shows that, as the number of alternating vowels increases, participants are less likely to select historical forms.

Figure 2

Plots of levelling survey results. Dashed lines mark 50% (chance) historical selection rate. (a) Speaker-by-speaker plot of percentage of historical forms selected, by proficiency score; (b) speaker-by-speaker plot of percentage of historical forms selected, by age; (c) percentage of times historical forms were selected, by wordfamiliarity score, for words that were sampled four or more times; (d) percentage of historical forms selected, by number of alternating vowels. In both (a) and (b), a lighter shade indicates that the speaker accepted foil forms at a higher rate.

5.1.3 Model criticism

Table V shows the fixed effects of the maximal model, where negative coefficients indicate that a predictor improves the odds of the innovative form ([ndoː-mkɪzɪn]) and positive coefficients improve the odds of the historical form ([n-mʌkzɪn]).

Table V

Fixed-effects table for maximal model of levelling preference results. Italicised factors are excluded in the best-fitting model.

The key confirmation is that historical forms are strongly dispreferred, as can be seen in Fig. 2 above, where the historical form is typically chosen less than 50% of the time. That is, in addition to the baseline tendency to accept innovative forms represented in the intercept, there is also a large effect favouring the selection of innovative forms when there would be multiple vowel–zero alternations in the conservative form. Historical forms stand the best chance of being selected if the speaker is younger than average, or if the historical form is locally ambiguous with an innovative prefix. Notably, the effect of age is smaller than the effect of the presence of multiple alternating vowels. That is, in order for a historical form to be favoured when multiple vowels alternate, the speaker must be more than one standard deviation younger than the survey average. The dispreference for historical forms underlines the major result of the prefix allomorph survey: the cohort has effectively abandoned 1930s Nishnaabemwin forms.

6.1 Results of Survey 3

All speakers preferred attested 1930s Nishnaabemwin forms over foils. This can be seen in Figs 3a and b, which plot the rate at which individual speakers chose attested 1930s Nishnaabemwin forms by proficiency and age. Furthermore, word familiarity, but not the number of alternating vowels, is positively correlated with the rate at which 1930s forms were selected (Figs 3c, d).

Figure 3

Plots of alternation accuracy survey results. Dashed lines mark 50% (chance) historical selection rate. (a) Speaker-by-speaker plot of percentage of historical forms selected, by proficiency score; (b) speaker-by-speaker plot of percentage of historical forms selected, by age; (c) percentage of times historical form was selected, by word-familiarity score, for words that were sampled four or more times; (d) percentage of historical forms selected, by number of alternating vowels.

6.1.3 Model criticism

Table VI shows the fixed effects for the maximal model (containing proficiency, familiarity and age as predictors). Positive coefficients correspond to an increase in the likelihood that the attested form is chosen.

Table VI

Fixed-effects table for maximal model of alternation accuracy results. Italicised factors are excluded in the best-fitting model.

The model estimates that historical forms are selected at a much higher rate than foils, as reflected in the intercept value. This is significantly augmented by the effect of word familiarity, and whether there were multiple alternating vowels.

7.1 Restructuring analysis

A restructuring analysis is quite simple. Underlying forms, to a very good first approximation (Bowers Reference Bowers2015: ch. 5), correspond to 1930s Nishnaabemwin unprefixed forms, and the person prefixes [ndʌ-, ndɪ-, ndoː-], and even [n-, nd-], are probabilistically deployed according to a speaker's preference for particular allomorphs (Bowers Reference Bowers2012). There is no need for the grammar to build and destroy iambic feet under this analysis, because all of the missing etymological vowels are already absent from underlying representations. Hence levelled paradigms in the modern Odawa varieties are generated via the identity map, so that /mkɪzɪn/ → [mkɪzɪn] and /ndoː-mkɪzɪn/ → [ndo:mkɪzɪn]. Other prefixed forms, like *[nmkɪzɪn] or *[ndmkɪzɪn], are ruled out by presumably high-ranked phonotactic constraints, as the word-initial clusters in such forms were unattested in 1930s Nishnaabemwin.

Central to the analysis of Modern Nishnaabemwin is the continued acceptance of 1930s prefixed forms. Under a restructuring analysis, 1930s Nishnaabemwin alternations are irregular. That is, the 1930s form for ‘my shoe’ is stored whole as /n-mʌkzɪn/, and produced without alteration. Deriving 1930s alternations via a general, regular rule is impossible if etymological vowels have been lost from URs. For instance, from a restructured UR like /mkɪzɪn/, vowels that appeared only in 1930s Nishnaabemwin prefixed forms, like the [ʌ] in [nmʌkzɪn], would need to be epenthesised. With three potentially epenthetic short vowels, [ʌ ɪ ʊ], such a procedure is not generally applicable.

There is some independent evidence for 1930s forms being listed irregulars, as many speakers in the study reported that regular innovative forms had compositional meanings, while 1930s prefixed forms had idiosyncratic meanings. For instance, one participant reported that the innovative form [ndoː-goːʤɪn] means ‘I hang’ (cf. [goːʤɪn] ‘he hangs’), while the conservative form [ndʌ-goːʤɪn] means ‘I fall’. This is reminiscent of the state of affairs in Tongan with respect to regularised and etymological suffix allomorphs (Kie Zuraw, personal communication); see also Hale (Reference Hale and Sebeok1973, Reference Hale1991: 100) on Maori.

A restructuring analysis, because it merely selects prefix allomorphs and does not alter the underlying representation, is compatible with any modern theory of grammar. Only the variability in prefix allomorph choice requires any departure from default grammatical mechanisms. A full analysis could be couched in current theories such as Partial Order OT (Anttila Reference Anttila, Hinskens, van Hout and Wetzels1997), Stochastic OT (Boersma Reference Boersma1997, Boersma & Hayes Reference Boersma and Hayes2001), Noisy Harmonic Grammar (Boersma & Pater Reference Boersma, Pater, McCarthy and Pater2016) or the Maximum Entropy model of Harmonic Grammar (Goldwater & Johnson Reference Goldwater, Johnson, Spenader, Eriksson and Dahl2003).

7.2 Recapitulationist analysis

A recapitulationist analysis of Modern Nishnaabemwin posits that underlying forms have not restructured, and that vowels continue to delete in an alternating pattern (Valentine Reference Valentine2001: 65–66, 79–80; see also Valentine Reference Valentine1994: 147–148 on Potawatomi). For convenience, the recapitulationist grammar is assumed to enforce a serial interaction between stress and syncope (McCarthy Reference McCarthy2008), though a subregular transducer obeying certain properties can also compute the mapping (Bowers & Hao Reference Bowers and Hao2018).Footnote ²⁰

In order to generate rhythmic syncope alternations, [n-] must be part of a foot that contains stem material, thereby shifting foot boundaries leftwards. Meanwhile, in order to condition non-alternation, innovative prefixes must be followed by a foot boundary. The prefix [ndoː-] has a long vowel and thus is always followed by a foot boundary. However, [ndʌ-] and [ndɪ-] can only be followed by a foot boundary if they are disyllabic. This can be ensured by assuming that person prefixes contain syllabic consonants (see Valentine Reference Valentine2001: 74), yielding URs like /n̩-/ or /n̩dʌ-/.Footnote ²¹ Under the recapitulationist theory, the absence of overt alternations between [mkɪzɪn] and [n̩doː-mkɪzɪn] is the result of a new person prefix being followed by a foot boundary. Thus, whether a stem is preceded by a new person prefix or appears in isolation, its vowels will always be parsed into the same foot position, as in (m_ˈkɪ)(ˈzɪn) and (n̩ˈdoː-)(m_ˈkɪ)(ˈzɪn). This is illustrated in the derivation in (17).

1. (17)

   

Because of the neat correspondence between foot boundaries and innovative prefix boundaries, under the recapitulationist account, the lexicon and grammar of Modern Nishnaabemwin remain substantially the same as 1930s Nishnaabemwin. The only major differences are that phonetic reduction is now phonological deletion, and that innovative prefix allomorphs have been created.

Note that the development of innovative prefix allomorphs must be stipulated in a recapitulationist account. Under a rhythmic syncope grammar, the presence of the stem-initial vowel in (n̩d-ˈʌ)(ˈgoː)(ˈʤɪn) ‘I hang’ or its absence in (_ˈgoː)(ˈʤɪn) ‘he hangs’ has a perfectly sound phonological explanation, where the stem vowel remains part of the stem, yielding /ʌgoːʤɪn/, and deletes following the general rhythmic syncope pattern. No innovation would therefore be expected in the language, and 1930s Nishnaabemwin would have been truly phonologised. An important apparent benefit of a recapitulationist analysis is that it need not resort to listing 1930s Nishnaabemwin prefixed forms, since it generates them directly from the UR when an appropriate prefix is used. For instance, forms like [n̩-mʌkzɪn] are analysed as the result of /n̩-mʌkɪzɪn/ →(n̩ˈmʌ)(kɪˈzɪn) → [n̩mʌkzɪn]. However, §7.3 shows that continued syncopation is a liability for the recapitulationist analysis.

7.3 Distinguishing restructuring from recapitulation

Wider descriptive facts weigh against the recapitulationist analysis of Modern Nishnaabemwin. As seen in (17), the recapitulationist analysis generates modern levelled paradigms by attaching an innovative prefix to a historically alternating word, as in /n̩doː-mʌkɪzɪn/ → (n̩ˈdoː)-(mʌˈkɪ)(ˈzɪn) → [(nˈdoː)-(m_ˈkɪ)(ˈzɪn)], but it can also attach a conservative prefix to produce a conservative prefixed form, as in /n̩-mʌkɪzɪn/ →(n̩-ˈmʌ)(kɪˈzɪn) → [(n̩-ˈmʌ)(k_ˈzɪn)] ‘my shoe’. However, this result only holds for consonant-initial alternating words. The recapitulationist analysis predicts that alternating words that began with a short vowel (VCV words) will overapply deletion. To see this, first consider the 1930s derivation of /ʌnʊkiː/ ‘work’ in (18).

1. (18)

   

Recall from §2.2.1 that stem-initial short vowels were parsed into the prefix, thereby creating the innovative prefix allomorphs. Concretely, learners exposed to 1930s Nishnaabemwin analysed [ndʌnkiː] as [(n̩ˈdʌ)-(n_ˈkiː)] ‘I work’. As a result, the underlying representation for ‘work’ is /nʊkiː/, [ʊ] being supplied from the unaffixed form [nʊkiː] ‘he works’. The recapitulationist analysis thus predicts that modern speakers should produce the paradigm in (19).

1. (19)

   

According to Rhodes (Reference Rhodes1985a, Reference Rhodesb: 467), modern speakers produce four forms. These are [nʊkiː] ‘he works’, with no deletion, [ndʌ-nʊkiː] ‘I work’ (where [ndʌ-] stands in for any innovative prefix), and, less frequently, the conservative deletion form [ndʌnkiː] ‘I work’ and the form [n-nʊkiː], with /n-/ attached to the stem. The recapitulationist analysis must include the original stem-initial vowel in the UR to prevent the vowel [ʊ] from obligatorily deleting in isolation forms and after innovative prefixes. That is, with all original vowels included, the recapitulationist analysis produces /ʌnʊkiː/ → [(_ˈnʊ)(ˈkiː)] and /ndʌ-ʌnʊkiː/ →[(n̩ˈdʌ)-(_ˈnʊ)(ˈkiː)].Footnote ²²

Adopting this analysis requires an internally inconsistent view of the rise of the innovative prefix allomorphs (see Valentine Reference Valentine1994: 148 for early acknowledgement of the inconsistency). An underlying representation like /ʌnʊkiː/ can only exist if the morphological boundary between prefix and stem is not reanalysed from [nd-ʌnkiː] to [ndʌ-nkiː]. However, the most salient aspect of the modern language is that the innovative prefix allomorphs have cascaded across the lexicon, making it all but certain that the reanalysis occurred. The recapitulationist analysis must therefore propose that a rhythmic syncope grammar and the correct morphological affiliations were found by learners, but that learners nonetheless changed the language away from the rhythmic syncope system.

In contrast, the restructuring analysis states that the UR is /nʊkiː/, and the modern paradigm is generated seamlessly by the identity map, as in /nʊkiː/ → [nʊkiː] ‘he works’, /ndʌ-nʊkiː/ → [ndʌ-nʊkiː] or /n-nʊkiː/ → [n-nʊkiː]. Continued use of the historical form [ndʌnkiː] is due to lexical listing.

In sum, a recapitulationist analysis either misgenerates or results in contradiction when innovative prefixes on VCV words are considered. A restructuring account does not suffer from this problem.