2.1. Prosodic status of pronominal clitics and domains for sandhi phenomena

The debate was fuelled by Zwicky and Pullum’s (1983, pp. 503–504) criteria advanced to circumscribe divergent properties of English not and n’t. Some of their points prove also relevant for EP clitic pronouns: (i) unlike clitics, which are moderately selective about the category of their host, affixes are rigorous in selecting the category of the stem to which they attach; (ii) morphophonological properties that cannot be captured by general phonological rules in a given language are easily found in affixed words, but on a lesser scale in clitic-host combinations; (iii) unpredictable gaps are more characteristic of affix-stem than of clitic-host combinations (see also Miller & Monachesi, 2003, pp. 92–93). These observations seem to lend support to the affixhood of EP object pronouns. Zwicky and Pullum (1983) also point to the recursive addition of clitics, but not of affixes, to material already containing clitics, and to the fact that stem-affix combinations outrank host-clitic groups in that they are more often semantically irregular.

Yet, the affixhood hypothesis clashes with some of the distributional properties of clitic pronouns. Therefore, the idea of their word-internal insertion must be taken with reservations, if not jettisoned altogether (the point is extensively discussed in Vigário, 2003, pp. 131–155). Stress patterns are usually referred to in order to corroborate the non-affixal status of pronominal clitics. In EP, word stress can travel maximally to the third syllable from the right edge of the word (three-syllable window; Luís & Kaiser, 2016, p. 214). If clitic pronouns were parts of verbal forms, their attachment would coerce the stress into moving onwards, so that its placement would be compliant with the three-syllable window. Meanwhile, postverbal clitic pronouns do not alter the locus of the word stress in EP.

Aside from lexical stress, in numerous other cases word-internal allomorphy fails to come about if clitic pronouns, rather than inflectional or derivational affixes, are added to the stem. An intriguing case involves the insertion of the non-back glide between two heterosyllabic vowels, the first of which is the stressed [e]. It can be found in the first-person present indicative, the present subjunctive and most of the imperative forms of verbs whose infinitive ends in -ear (despentear ‘to mess up one’s hair’ – despenteio, chatear ‘to annoy’ – chateio, manusear ‘to handle’ – manuseio, etc.) or, for some Latinate verbs, in -iar (incendiar ‘to set on fire’, comerciar ‘to trade’, negociar ‘to have business affairs’).¹ This process is blocked if the segment [e] is followed by a direct object pronoun. For example, lê-o ‘read it’ surfaces as [le.u], but never as [le.ju]. Thus, unlike the inflectional ending -o, accusative pronominal clitics do not act as word-internal syllables with respect to this sound change.

The affixhood vs. wordhood debate can be associated with the theoretical construct known as Strong Lexicalist Hypothesis (SLH): No syntactic rule can refer to elements of morphological structure (Dalrymple, 2001, pp. 83–84, 97–98). If speech production proceeds sequentially (the prerequisite to which is a set of hierarchically ordered modules; see Levelt, 1989; Labelle, 2001, pp. 156, 161–162), clitic attachment deviates from how ‘normal’ multi-word combinations are built. If clitic pronouns (and affixes) are parts of another expression, their processing must start very early. The syntactic module cannot operate until its input has been fully specified. This is tantamount to saying that all of the irregular properties of verb-clitic/clitic-verb combinations are word-internal and arise in the lexical module. Even more radically, SLH requires two separate entries to be stored in the mental lexicon of EP users: one for a bare verb and another one for sequences combining verb forms and clitic pronouns. The use of the latter entry becomes mandatory if arguments have not been planned as fully-fledged NPs in an earlier module. Therefore, clitic-verb/verb-clitic sequences must be memorized on a one-by-one basis, just as some stem-affix combinations are claimed to be.

In other proposals, remedial measures are introduced to obviate the need for positing an overgrown lexical module. One of them relies on the assumption that, instead of operating on raw sequences generated in the morphological and the syntactic component, the phonological interpretation has an indirect character. It is mediated by the prosodic structure (Frota, 2013, pp. 4–5), which involves a set of hierarchically organized domains: syllable (s), prosodic word (PW, or ɷ), phonological phrase (PP, or f), intonational phrase (IP, or I) and phonological utterance.

Syllable consists minimally of a nucleus, and maximally of an onset, nucleus, and coda. Taken together, the nucleus and the coda constitute the rhyme. The nucleus represents the peak of sonority; less prominent segments that precede the nucleus belong to the onset. Those that follow the nucleus constitute the coda. The onset and the coda may remain empty (alternatively, the term onsetless is used for empty-onset syllables). If no segments are present in the coda, the syllable is open. For stress computation purposes, onset dominates the sound segments directly, whereas for the nucleus and the coda, the dominance is mediated via mora constituents. Therefore, the rhyme is the moraic component (Collischonn & Wetzels, 2016, p. 87).

Other prosodic elements – prosodic word group, and clitic group – are sometimes added to this list. These are the domains for sandhi phenomena in EP. For example, in Vigário’s (2003, p. 125) typology of phonological processes, all of the postlexical phenomena make reference to the PW. By contrast, in word-internal phonological changes, some processes are active on this level as well (initial vowel realization, heterosyllabic [e]-centralization, hiatus-breaking glide insertion; see also Frota, 2013, pp. 83–88, for the experimental study of the last of these phenomena), whereas others (vowel reduction, theme vowel deletion, and final nasal diphthongization) are not.

Essentially, prosodic domains need not overlap with previously generated morphosyntactic units (their boundaries are not coincident).² Likewise, in the rules operative in a given domain, reference to other grammatical modules (or to semantic information) is allowed. This approach to clitic-verb/verb-clitic sequences does away with the need to combine expressions of both categories in the lexicon. Even if some phonological changes affect the form of the verb and the form of the clitic in an unpredictable way, compelling evidence is also found for the non-lexical derivation of verb-clitic sequences in EP.

One of the models stemming from Prosodic Phonology involves precompiled phrasal allomorphy. Precompilation lists allomorphs (in the lexicon) alongside contexts for their activation.³ An alternative procedure involves the list of lexical rules for the instantiation of particular allomorphs. These rules are paired with suitable syntactic environments. Analyses conducted using this theoretical apparatus show that EP clitic pronouns are defectively prosodized morphosyntactic words. Despite occupying syntactic terminal nodes, they do not form autonomous PWs (Santos & Vigário, 2016, p. 130).

The domain instrumental in accounting for the behavior of phonological clitics in EP is the PW. It usually includes a stem plus any of the adjacent affixes. If any of them forms an independent domain for lexical stress assignment, it constitutes a PW on its own. In other words, each PW bears exactly one lexical stress, which is allocated at its right edge (three-syllable window). PWs can be clustered together to form a higher level domain, namely Prosodic Word Group. That such cumulative units consist of successive PWs can be seen in certain compounds and abbreviations.

As phonological clitics are inherently unstressed, they must be part of one of the PWs onto which the adjacent expressions are mapped. In EP, clitic words are not all alike with respect to the PWs to which they belong. Proclitics prosodize with the host PW, thereby elongating it by one or two syllables at best. These syllables are simply adjoined at the PW’s left edge. Phenomena revealing this type of attachment include the alternation in height of non-central non-high vowels, prosodic word initial stress, and emphatic stress assignment. With respect to all of them, prosodically deficient function words (including definite articles) behave like PW-initial segments (Vigário, 2003, p. 196). By contrast, postverbal clitic pronouns are incorporated into the preceding PW. The case is discussed by Vigário (2003, pp. 187–188) and Santos and Vigário (2016, p. 131). The postlexical sandhi phenomenon, deemed to best grasp the nature of incorporation, involves non-back final vowel deletion. It comes about at the PW’s right edge. To be deleted, the non-back vowel must be PW-final, and occur in IP-medial position. For example, in TOme-o de maNHÃ ‘Take it in the morning’, the enclitic pronoun prevents the host final vowel from occupying the PW-final position. Therefore, instead of being deleted, it must surface as a non-back glide. By contrast, in CAbe-te a TI (limpar a casa) ‘it’s up to you (to clean the house)’, the clitic final vowel becomes PW-final and is deleted [‘kab.ta.‘ti], whereas the IP continues.

Optional [ǝ] deletion in sentences like JA te aconteCEU ‘(It) has already happened to you’ (another realization involves the non-back glide [j]) is argued to be indicative of the fact that clitics are not enclitically attached to the preceding PW and that preverbal pronominal clitics are proclitic by nature (Santos & Vigário, 2016, p. 131). Besides advancing our understanding of the variationist dimension of present-day EP dialectal speech, nasal epenthesis in preverbal accusative clitic pronouns is significant in that it poses a challenge to this opinion, at least as far as EP dialects are concerned. If no, na, nos, nas variants are precompiled and listed as allomorphs alongside vowel-initial forms (e.g., Vigário, 1999, p. 589), then the conditions for their instantiation must indicate that they prosodize with the preceding diphthong-final PW. The non-verbal category of the expression corresponding to the hosting PW must likewise be disclosed. In sum, despite occurring in the preverbal position, accusative clitic pronouns are assumed to be enclitically incorporated in the preceding expression. Rather than prosodically adjoining to the ensuing verb, they form a PW along with the left-hand-side proclisis trigger.

2.2. Nasal diphthongs

It is a long-established idea that nasal vowels represent an underlying tautosyllabic VN sequence. It dates back to at least the Prague School (Piggott, 1992, p. 38). This assumption was later fine-tuned by Mattoso Câmara Jr (1971, p. 33) to account for the status of nasal diphthongs in Portuguese. He posited a sequence that consists of an oral diphthong followed by a nasal archiphoneme. The nasality of the latter spreads into the preceding segment. Bisol (2008, p. 2) sustains the idea that both nasal vowels and nasal diphthongs rely upon VN in their underlying form, except that the initial sequence is VN for nasal vowels and VN-V for nasal diphthongs. According to her line of reasoning, the intervocalic nasal is incorporated into the vowels. This is why the V(nasal)V sequence fails to produce a hiatus. The intervocalic nasal is either lost or realized as the onset of the next syllable: luna > lũa > lua ‘moon’; um one.masc – ũa > uma one.fem (see also Martins 2016a, pp. 7, 12). Moreover, Bisol (2008, p. 11) makes a distinction between two types of diphthongs in Portuguese: true diphthongs (portão ‘gate’, põe ‘puts’, capitães ‘captains’) correspond to two input vowels, one of which becomes a glide by syllabification. By contrast, false diphthongs (viagem ‘trip’, homem ‘man’, som ‘sound’) have only one vowel in the input. In present-day standard EP, verb final nasal diphthongs surface as nasal glides (e.g., viram [’vi.ʀɐ̃w̃]; Vigário, 2003, p. 76). The nasality of the vowels in nasal diphthongs is assumed to arise via the assimilation from the following tautosyllabic nasal consonant. Thus, the nasality of the glide carries over into the preceding vocalic segment. The two segments share the same value for the feature [back].

The ongoing debate over Portuguese nasal diphthongs revolves around the structure of the syllable they realize. For some scholars, these segments represent complex (or ramified) nuclei of open syllables (Mateus & d’Andrade, 2000; Martins, 2016a, p. 11, “Outras mudanças ocorreram que afetaram os ditongos decrescentes sem lhes alterar a natureza de núcleos ramificados” ‘Other changes came about and affected falling diphthongs without altering their nature of ramified nuclei’ [emphasis mine]). For others, falling nasal diphthongs are rhymes with both their nucleus and coda constituents realized. The glide is located in the syllable coda in this approach (Collishonn & Wetzels, 2016, pp. 89–90). The latter position is corroborated to some extent by word stress patterns of EP. Filled codas in nasal diphthongs are posited on account of the sameness of effect on stress assignment exerted by both closed syllables and diphthongs. This line of reasoning hinges on how successfully the notion of syllable weight is mapped onto present-day Portuguese (many of the analyses conducted in this vein take into account Brazilian Portuguese).⁴

Finally, word-final diphthongization is argued to be a word-internal process (Vigário, 2003, pp. 77–78). Two apparent counterexamples – irmã ‘sister’ and orfã ‘orphan.fem’, which end in a simple nasal vowel and where no nasal glide is present – are assumed to arise by the coalescence (Sampson, 2016, p. 669) of the stem-final vowel and the feminine class marker -a. No underlying final nasal segment is present in these words. Therefore, no nasal glide surfaces. In verbal morphology, a nasal diphthong is steadily upheld in the closing syllable, no matter whether a given verb form stands alone or is followed by other expressions, including clitic pronouns. This stability shows that the relevant domain for word-final nasal diphthongization is the morphological word rather than the prosodic word.

Historically, the [ɐ̃w̃] diphthong is known to have developed from word-final simple nasal vowels [õ] and [ã]. By contrast, [ẽj̃]/[ɐ̃j̃] diphthongs derive from the word-final [ẽ]. Generally, these evolutions do not stimulate much controversy in the literature (Sampson, 1983, pp. 34–37; Fagan, 1992, pp. 383–385). An exception to this is Mattos e Silva’s (2006, p. 73) reconstruction, where three stages, rather than two, are singled out. The intermediate phase in the development from [õ] to [ɐ̃w̃] involves the [õw̃] nasal diphthong. Three late Latin endings – -ane, -one, -anu – are presumed to have produced the [õw̃] diphthong. The syncope of the intervocalic [n] is instrumental in triggering the nasal resonance of the preceding vowel. Mattos e Silva posits the following sequence of events to explain the advent of the [õw̃]: i) the [+ nasal] feature of [n] carries over onto the preceding vowel, i.e., [on] > [õn]; ii) rather than being simply deleted, [n] turns into a glide, i.e., [õn] > [õw]; iii) the [+ nasal] feature is inherited by the resulting semi-vowel, eventually producing [õw̃]. The emergence of [w̃] to replace [n] in the ending -one is traced back to the fact that both segments – the vowel and the resulting glide – share [+ back], [+ round] features.

Besides phonetic reconstructions, attempts were made to establish correspondences between the mutations of particular Latin endings and the spelling variants representing them in Galician-Portuguese documents. For example, Maia (1986, p. 602) identifies the following graphic developments for -one: -õ, -on, and -om (e.g., condiçom, condiçõ). Finally, the chronology of the emergence of word-final nasal diphthongs is variously reconstructed. Martins (2016a, pp. 11–12) localizes the advent of [ɐ̃w̃], [ẽj̃] and [ɐ̃j̃] in Old Portuguese. In so doing, she contradicts the opinion of Ramos (1985, p. 95), who concludes that at that stage the convergence of word-final simple nasal vowels [õ] and [ã] towards the nasal diphthong [ɐ̃w̃] was still under way. Essentially, these reconstructions are indicative of the fact that some of the CORDIAL-SIN realizations of the negative marker with a simple nasal vowel (i.e., {nũ=não}) in the rhyme should not be approached in terms of monophthongization. On the contrary, from the historical perspective, in present-day standard EP falling diphthongs represent the development from simple nasal vowels.

In the history of Portuguese, word-final nasality also had a direct bearing on clitic allomorphy in the preverbal position, which is the principal point of interest in the present paper. Pronominal clitics were subject to different forms of liaison with both the preceding and the ensuing sound material. In the latter variant, the encounter of pronouns’ vocalic nuclei with vowel-initial verb forms resulted in vowel deletions. Elided forms in this configuration were optional. They helped circumvent the formation of heterosyllabic vocalic clusters. The point is discussed by Martins along with relevant examples (Martins, 1994, pp. 217–218), e.g., sobligarõ (< se obligaram ‘(they) took an obligation onto themselves’), todo lhera prazer (< tudo lhe era prazer ‘everything was a pleasure for him’), etc. By contrast, fusions with the preceding sound material were exclusive to third-person direct object pronouns. In such configurations, o, a, os, as surfaced with an [n] onset following proclisis triggers ending in a nasal segment.

Finally, in medieval Portuguese, word-final nasality triggered liaisons of definite articles with the preceding expressions. The array of forms following nasal-final function words (in particular, the prepositions com ‘with’, sem ‘without’, and em ‘in’, and the negative nem) involved no, na, nos, and nas alongside lo, la, los, and las. Thus, in the manuscripts of Galicia and northern Portugal, aside from cũ lo noſſo caſal ‘with our farmhouse’, there are occurrences of cõ no conuento ‘with the monastery’ (Maia, 1986, p. 648). This is an intriguing case where allomorphy and free variation play along with each other. Whereas [n]-initial and [l]-initial forms to the exclusion of the vowel-initial one are an instance of complementary distribution, the selection of either of these two consonant-initial realizations was made on a free-choice basis.

3.1. Pronominal clitics in present-day EP

Enclisis is the default pattern in non-dependent finite clauses. Proclisis is conditioned by the elements that have the property of attracting the pronoun to the pre-verbal position. They constitute the first element of the sequences subject to analysis in this paper. They include: i) negation; ii) preverbal negative indefinite pronouns and adverbs (ninguém, nunca); iii) some other adverbs placed preverbally, such as já ‘already’, ainda ‘still’, among others; iv) preverbal quantified subjects. Proclisis occurs also in: v) clauses with a filled CP, e.g., qu-questions; and vi) subordinate clauses with overt complementizers. These distributional properties illustrate the syntactic behavior of clitics as heads, since they require a host of a specific grammatical category.

In standard EP grammar nasal epenthesis in accusative clitic pronouns is obligatory in enclisis only if the preceding verb form ends in a nasal diphthong. This constraint is nearly always respected in the speech of CORDIAL-SIN informants (see 1a). Some scholars impose an additional requirement on the verb saying that it must encode the [+ plural] feature (Luís & Kaiser, 2016, pp. 221–222). In the corpus, singular verb forms tend to combine with a vowel-initial accusative clitic pronoun even if their final syllable contains a suitable diphthongal segment (see 1b).

Moreover, nasal onsets are clitic-specific in today’s standard EP. As discussed in 2.2, unlike in early stages of Portuguese, postverbal definite articles do not undergo allomorphy in present-day EP. They keep their form intact in spite of being homonymous with 3^rd person accusative pronouns and no matter what the phonological profile of the syllable preceding them is (see 2a). Likewise, preverbal accusative pronouns are always vowel-initial in the standard grammar (see 2b).

3.2. Clitic placement in diachrony

In the history of EP, clitic placement had been different from what it is nowadays. In some of the structural environments, which were subsequently reanalyzed as obligatory enclisis contexts (from the first half of the 19^th century on; some authors locate the change slightly earlier; see Galves et al., 2005; Namiuti, 2008, p. 53; Sandalo & Galves, 2013), pronominal clitics occurred either pre- or postverbally. Galves et al. (2005, pp. 42–43) single out five positions where the choice was made on an optional basis: root clauses with an overtly expressed preverbal subject, clitic left dislocation, sentences with a clause-initial non-VP adverb or non-argumental prepositional phrase. Additionally, the position of pronominal clitics varied in verb-first root clauses following a hierarchically subordinate clause and in verb-first root clauses following coordinating conjunctions e ‘and’, pois ‘then’, mas ‘but’. Thus, if verb-clitic/clitic-verb variation contexts are considered to have been a separate mode of placement, pronominal clitics were able to occur in as many as five different positions with respect to the verb: enclisis, mesoclisis, non-obligatory proclisis, obligatory proclisis and interpolation (clitic-verb non-contiguity). The negative marker não had a special (albeit not exclusive) role in instantiating the last of these linear arrangements. Importantly, not only did the interpolation of não show up in obligatory proclisis contexts in negated clauses, but also in verb-clitic/clitic-verb ones (Namiuti-Temponi, 2006, p. 173). Obligatory proclisis contexts remained essentially unchanged throughout history. By contrast, the only structural position where enclisis has been intact involves the so called Tobler-Mussafia context (Hirschbühler & Labelle, 2000, p. 165): a definite ban on sentence-initial pronominal clitics. Indeed, violations of this constraint are really exceptional.

Finally, verb-clitic/clitic-verb variation contexts had varying outcomes in history. In alternate periods, preference was given either to the preverbal (its peak was reached in the 15^th century) or to the postverbal placement. It seems that the revival of enclisis in free-choice environments had a popular inspiration. In her study of the speech of Gil Vicente’s popular characters (16^th century), Martins (2012, p. 98) shows that they tended to use an enclitic grammar.

6.1. Local distributions and individual behaviors

The first variable is designed to measure how the interview sites are different in the number of informants employing nasal epenthesis. The result is indicative of whether nasal allomorph is a socially widespread mode of realization in a given interview site or is rather isolated (see Seiler, 2006, p. 167). It also measures the degree of uniformity among the local speakers in the processing of the configuration at work. To this end, the percentage of informants producing nasal-initial pronouns needs to be calculated. The count in Table 2 takes into consideration the overall number of informants producing the TCP sequence.

On average, no snowball effect can be found. This is to say that seldom does the presence of a consonant-initial preverbal accusative pronoun in the speech of an informant carry over to the rest of the speakers in the same locality. The higher the number of the interviewees producing the TCP sequence, the lower the percentage of onset-inserting ones is. The three localities boasting the 100% score involve maximally two (clh, flf) informants producing the relevant configuration. In those cases where the nasal allomorph is present in the speech of the majority of interviewees or where the two groups are equal (ctl, cov, grc), the overall number of speakers is likewise low, reaching maximally four persons (ctl). Besides, individual behaviors tend to be balanced, with there being nearly as many individuals who only rely on canonical sequences as there are those who produce at least one consonant-initial preverbal accusative pronoun.

The second variable is designed to measure the strength of the tendency to employ nasal onsets across interview sites, once a TCP sequence is present there (Seiler, 2006, p. 168). In other words, the ratio of consonant-initial preverbal accusative pronouns to vowel-initial ones needs to be calculated for every locality. As said above, in none of the places where the interviews were conducted are nasal-initial preverbal accusative pronouns employed as the unique variant. Yet, in some places (Table 3), they outnumber the corresponding canonical forms.

All four localities (clh, crv, flf, grc) with prevailing non-canonical realizations are clustered in the archipelago of the Azores. At the same time, the number of TCP sequences documented in the interviews conducted there is low, thus calling for some caution in drawing conclusions. In two interviews (exb, min), a single occurrence of consonant-initial variant invites even more caution. Tellingly, in the localities where the highest numbers of TCP sequences are found (cov, grj, mld, pic), the vowel-initial variant is clearly preferred.

Another variable is speaker-oriented. It characterizes individual grammars of CORDIAL-SIN speakers by measuring the position of TCP sequences amid the contexts where phonological conditions are met for nasal epenthesis in prosodically deficient words. Thus, particular informants’ speech will be checked for the same non-canonical forms in syntactically different conditions (mainly verb – definite article plus rare occurrences of verb – demonstrative pronoun and adverb – definite article). This procedure is believed to be more justifiable than describing localities in terms of the presence of the nasal epenthesis in other contexts. The uniqueness of TCP sequences in being the locus for onset insertion can be taken as a proof for a special status of clitic pronouns in the grammar of CORDIAL-SIN informants. Special status is meant here to refer to speakers’ awareness that expressions belonging to other categories, even if homonymous with accusative clitics, are precluded from undergoing the epenthesis. Conversely, the presence of nasal onsets in other phonologically comparable settings lends support to the hypothesis of a phonology-driven individual grammar, with an apparently indiscriminate treatment of clitic words of any category in various syntactic contexts.

In four of the 15 localities where the relevant nasal epentheses are found, the only context where they appear is the TPC sequence, i.e., proclisis trigger ending in a nasal segment followed by 3^rd person direct object pronouns (clh, exb, flf, grc; in sum, six informants with nasal-initial preverbal accusative pronouns, one – with canonical forms only). These epentheses fail to appear in other configurations with clitic words. In the remaining eleven localities, 44 informants produce the TPC sequence. Twenty of them produce at least one nasal-initial preverbal clitic pronoun and 24 do not. These two groups show divergent behaviors towards nasal onsets in other configurations where favorable phonological conditions for the nasal epenthesis are met. The respective figures are as follows: 13 speakers in the former group employ the nasal onset also in other contexts (ratio 1: 0.538; see 5a–b above repeated as 9a–b below), and only six in the latter (ratio 1: 3; five interview sites: grj, out, pft, sta and pic with two such interviewees).

These figures show an overall avoidance of nasal onsets in other contexts in the speech of the informants who only use vowel-initial clitic pronouns in TCP sequences. By contrast, consonant-initial preverbal accusative pronouns are moderate attractors for the selection of the nasal-initial allomorph in other combinations involving clitic words and exhibiting suitable phonological structure. The 26 informants of this group are evenly divided: one group extends the alveolar nasal to other prosodically deficient words, whereas the second does not. Be that as it may, no implicational hierarchy emerges in either direction (i.e., nasal epenthesis in TCP sequences > nasal epenthesis elsewhere, or the other way round). Table 4 represents the variation in phonology-syntax mappings in CORDIAL-SIN speakers established on the basis of the above calculation.

With respect to vowel-initial phonological clitics, in the 15 localities subject to the count, an entirely canonical individual grammar is the best represented option (group 3), even if it hardly exceeds one third of the interviewees. This group does not lend itself to further description. The speech of its members exhibits no variation whatsoever. As far as vowel-initial clitic words are concerned, everything is compliant with standard EP grammar. Conversely, the behavior of the speakers belonging to the three other groups evinces an interesting interplay of methods of processing prosodically deficient monosyllables. These groups of informants instantiate what Harris and Campbell (1995, pp. 97–119) and Andersen (2001, p. 230, 2006, pp. 69–70) refer to as extension, i.e., a more or less intentional change for which no particular pragmatic motivation can be defined. One of the contexts in which it may take place is where a given morphophonemic rule is projected onto a new morphosyntactic environment and the innovation is immediately actualized in usage (see also Bybee & Torres Cacoullos, 2008, p. 400). The description that follows relies on the assumption that each time when speakers treat clitic pronouns and other clitic words differently with respect to nasal epenthesis, they must be aware that they represent distinct morphosyntactic categories (Nkollo, 2020, p. 294). Likewise, it should be borne in mind that in the individual speech production of CORDIAL-SIN informants canonical and non-canonical variants are systematically interwoven. Thus, the presence of an irregular nasal onset in a given clitic word must not be taken as a proof of an outright relinquishment of the standard grammar.

In group 2 nasal onsets are extended onto preverbal accusative pronouns. At the same time, other similar-sounding clitic words are consistently vowel-initial. Neither favorable phonological conditions nor the homonymy of accusative pronouns and definite articles prove sufficient to compel the informants into extending onsets beyond the TCP sequence. Therefore, a tacit assumption to the effect that only clitic pronouns are allowed to take a nasal onset must be part of this group’s internalized grammar. Nasal epenthesis in preverbal accusative clitics is thought to have come about by analogy with the alternations in the postverbal domain in phonologically equivalent conditions. Thus, the only impaired segment of standard EP clitic grammar in this group of informants pertains to the preverbal vs. postverbal divide. Class-specific treatment of clitic pronouns is maintained throughout.

At first glance, the utterances in group 1 were produced as though its members had been oblivious of morphosyntactic distinctions between various classes of clitic words, all of which are treated alike with respect to nasal onsets. In accusative clitic pronouns, initial nasal segments are introduced haphazardly, irrespective of whether they are postverbal or not. In other monosyllables, the nasal diphthong – oral vowel sequence seemingly acts as a sufficient cue to proceed with nasal epenthesis as well. Still, this method of processing clitic words must not be necessarily equated with a purely phonology-based grammar or the lack of any grammar whatsoever. As said above, besides the TCP sequence, it is the verb – definite article combination that is the most frequent locus for nasal onsets across words. The postverbal position is assumed here to be the overriding criterion. Thus, the linear ordering of clitic words induces speakers to process definite articles the same way as postverbal accusative pronouns. Obviously, the homonymy is another contributing factor to the category-blind nasal epenthesis.

Group 4 is peculiar in that its members behave as if they were aware of the ban on nasal epenthesis in pronominal clitics in the preverbal domain. Thus, nasal-initial allomorphs in 3^rd person accusative pronouns are rigorously restricted to enclisis. Curiously, in their speech this parameter does not clash with instances of nasal epenthesis outside TCP sequences. Nasal onsets are unwarily produced only if the preceding word is a verb and contains suitable sound material. Unlike in group 2, where the informants were rigorous in applying a class-specific treatment of clitic pronouns, group 4 members compute the conditions for the use of the nasal-initial allomorph on the basis of the postverbal placement of clitic words. If a verb form happens to be diphthong-final, the nasal segment appears and all other parameters are played down.

The last variable characterizes individual phonology-syntax mappings with respect to TCP only (see Table 5). It relies on the simple count of occurrences of nasal-initial preverbal accusative clitics in the global production of TCP sequences by a given informant. As two informants employ exclusively nasal-initial pronouns, the ratio cannot be calculated. The comparison of scores helps measure the strength (or the degree of entrenchment) of the non-canonical realization in individual grammars. As indicated above, there are 26 speakers producing nasal-initial preverbal accusative clitics. Ten of them produce merely a single TCP sequence, in which the pronoun happens to exhibit a nasal onset. They will be left out of the count, as a single occurrence tells us nothing about variation or its absence.

Again, the preference given to non-canonical forms is paired with a poor attestation of TCP sequences in the speech of particular CORDIAL-SIN informants. Conversely, the more frequent this combination is, the lower is the share of nasal-initial allomorphs. If the numbers of vowel-initial and consonant-initial variants are equal, speakers also produce only a small aggregate number of TCP sequences. Thus, once used, a nasal onset in the preverbal accusative pronoun is not mass-reproduced in CORDIAL-SIN interviews. Only as an exception do nasal-initial clitic pronouns play the role of attractors for nasal epenthesis in other TCP sequences produced by a given speaker. On average, this sandhi phenomenon appears to be no more than a fanciful realization, as it were, in an otherwise standard individual grammar.

6.2. Concomitant phonological processes

Aside from displaying varying shapes of the accusative clitic pronoun, TCP sequences in CORDIAL-SIN are even more markedly subject to a series of concomitant phonological processes. Whereas some of them leave the skeleton of the sequence intact, the others alter it to the point where it becomes necessary to redefine the phonological characteristics outlined in the introduction. One of these processes involves monophthongization. Technically, the glide in the first of the juxtaposed syllables is deleted, leaving the preceding vowel as the only element of the rhyme. Monophthongization is found in 53 out of 117 TCP sequences in the 15 localities with two concurrent realizations of the preverbal pronoun. It is also regular in other configurations in phonologically equivalent conditions. According to the annotations in the verbatim transcription section, monophthongization occurs more frequently in syllables preceding nasal-initial accusative pronouns (23 out of 45 TCP sequences). The corresponding result for sequences with vowel-initial pronouns works out at 30 out of 72 occurrences. The examples below illustrate the point in each of the TCP variants.

Two interpretations are to be envisaged. According to one, if the first syllable has a monosegmental rhyme, more propitious conditions are provided for an onset to be added to the subsequent clitic word. Conversely, if the shape of the pronoun is planned prior to the production of the TCP sequence, then the nasal onset retroactively triggers the deletion of the glide in the preceding syllable. Thus, there is a backward effect of one of the segments onto the realization of the preceding sound material. A non-left-to-right production of TCP sequences is the prerequisite for this scenario.

Further observation of CORDIAL-SIN data demonstrates that in the vast majority of cases monophthongization is paired with another phonological process. Once the glide in the first syllable is deleted, vowel quality tends to get altered, too. The most frequent alteration affects the nucleus of não [nɐ̃w̃], which surfaces eventually as [nũ], as can be seen in (11a). This change is found in both of the variants. In TCP sequences with consonant-initial pronouns, the alteration of vowel quality takes place in 17 out of 45 cases. Before vowel-initial pronouns, it affects 26 out of 72 sequences. The correlation between monophthongization and the shift in the vocalic timbre is unilateral. Monophthongs are produced irrespective of whether the vowel gets altered or not (see 10a-b and 11a-b), but change in vowel quality is nearly always superimposed onto monophthongs. There is, however, a single instance of vowel alteration paired with a diphthongal mode of the realization of the rhyme (see 11c).

The processes affecting TCP sequences discussed thus far have had no erosive effect on the phonological conditioning. Two concurrent realizations of the sequence remain essentially independent of monophthongization and alteration of vowel quality. By contrast, the third process alters the sequence to the point where an adjustment of preliminary conditions may become necessary. Four TCP sequences are subject to the loss of the nasal resonance in the nucleus of the first syllable. This shift is invariably paired with the nasal epenthesis in subsequent clitic pronouns. In addition, the loss of nasality in the rhyme of the first syllable is always concomitant to monophthongization and alteration of vowel quality. This mode of realization is found only in the mainland of Portugal. The districts are Aveiro (2 occurrences), Vila Real and Bragança (see 12a–d). In each of the cases, it is the negative marker não that is distorted. On occasion, the loss of the nasal resonance is found also in other syntactic configurations (see the cleft sentence in 13).

This cumulative phonological shift poses a challenge to the previously defined phonological side of the use of the alveolar nasal. Neither of the prerequisites (the diphthongal rhyme or its nasal resonance) is met in the first of the two juxtaposed syllables. Morphosyntactic constraints notwithstanding, a simple oral vowel ending the preceding word does not warrant the alveolar nasal any more than any other consonant. Proof of this is two-word sequences with a clitic element in which CORDIAL-SIN speakers insert intervocalic glides or liquids across words (see 14a–b).

Ergo the selection of the alveolar nasal must have been planned beforehand, i.e., before preparing words in speech production. This is not in keeping with the view that sound segments are activated sequentially, from left to right. One might wonder, then, if the production of TCP sequences in (12a–d) is really a feedforward process, with neatly staged phases: lexical selection, morphological and phonological encoding, phonetic encoding, and neuromuscular control of articulators (Levelt et al., 1999, pp. 3–4).

Two explanations for the nasal epenthesis after a simple oral vowel in TCP sequences can be advanced. These explanations are not mutually exclusive, although one of them would call for a thorough experimentally based confirmation. Generally, CORDIAL-SIN offers data produced by elderly speakers. Research over accuracy and speech rate point to substantial differences depending on the speaker’s age. In the research task designed and carried out by Bilodeau-Mercure and Tremblay (2016), speakers’ orofacial physiology – lip and tongue muscular flexibility and endurance, as well as tactile sensitivity – has been investigated. Older informants (66–85 yo) were demonstrated to have a reduced speech rate compared to younger speakers. In articulation, they were shown to run up against particular difficulties producing nasal vowels. They were trying to remedy this by either slowing down or producing an oral vowel – nasal consonant sequence instead of nasal vowels. This effect has been blamed on an age-related motor disorder, namely the decline of lip endurance.

As for CORDIAL-SIN data, the age of the remaining denasalizing informants is not revealed, except for the informant Arquibaldo. Yet many of the interviewees are around 60 years old or older. If the result of the experiment is correct, then the use of the nasal-initial allomorph can be seen as an attempt to make up for the loss of nasal resonance in the rhyme of the preceding syllable. Investigating this effect in CORDIAL-SIN speakers at present is probably an unfeasible task. Yet, examining other elderly informants is a possible direction for further research.

Another explanation for the selection of the alveolar nasal rather than of any other consonant is cue-based. Nasal onsets are obligatory in enclisis after plural verb forms ending in a nasal diphthong. Their perceptual salience in combinations involving clitic pronouns reaches such a degree that they can be extended over the preverbal domain even after a monosegmental and denasalized rhyme.

7.1. Historical continuity gone astray

In accordance with the competing grammars approach (Kroch, 2001; Galves & Kroch, 2016, pp. 487–488), individuals displaying variation in the realization of the TCP sequence, must have instantiated at some point different phonology-syntax mappings in a kind of internalized diglossia. Competition arises when speakers have been exposed to both a vernacular (with the nasal-initial allomorph) and a learned grammar (vowel-initial variant). A sufficient amount of evidence is a prerequisite to developing two simultaneous and alternative rankings of parameters. Conversely, if a single realization of either of the forms is steadily maintained in individual speech production, no diglossia can be applied. In this approach, language acquisition (parameter setting; Battye & Roberts, 1995, p. 8) is the decisive phase in bringing about grammatical change. Individual grammars are computed on the basis of both human biological endowment and a multitude of cues available to new generations of language users via their parents’ speech (Roberts, 2007, p. 124). Innovations can be equated with parameter resettings, i.e., micro-scale transitions between stable successive individual grammars. The principal mechanism prompting the change is the faulty imitation (by abductive reasoning) of the older generation’s speech habits. As the internalized grammar of their parents is not directly available to children, its reconstruction is unavoidably imperfect. Thus, infant grammar moves away from the original one. Observable change is explicitly declared to be piecemeal, gradual, and chaotic (Lightfoot, 1999, p. 105), which portrays well the actual usage of preverbal accusative clitics by CORDIAL-SIN informants. At present, the viability of this approach will be matched against the interplay of vowel- and consonant-initial allomorphs documented in history.

In her work devoted to Galician-Portuguese written texts, Maia (1986, p. 671) describes nasal-initial accusative pronouns as forms that always occur after a nasal-final word (poor attestation of vowel-initial forms after the preceding negative marker, e.g., nõ as aguardando ‘without waiting for them’ is noted accordingly). An assimilatory process, actuated by the nasal segment in the final syllable of the preceding word, is declared to have affected lo, la, los, las pronominal forms rather than empty-onset ones (see Martins, 1994, p. 214, “lo, los, la, las são as formas do pronome acusativo átono que, do ponto de vista evolutivo, precedem as formas o, os, a, as” ‘from the evolutionary point of view, lo, los, la, las are atonic accusative pronominal forms that are earlier than o, os, a, as’). Thus, two consonant-initial variants used to exist in parallel before the advent and the subsequent spread of vowel-initial forms.

Tellingly, Maia’s analysis contains no reference to the syntactic category of the preceding expression. This is tantamount to saying that the choice of either consonant-initial or vowel-initial realization must have been made in the phonological module. Morphosyntactic conditioning played virtually no role in the selection of the onset. The only incidence of syntax involved the linear ordering of expressions containing suitable sound material. Thus, inflected verb forms were but one of the possible types of hosts for nasal-initial accusative clitics. Examples quoted by Maia (1986, p. 672) leave no doubt as to the category-blind attachment of clitic pronouns with nasal onsets: e nõnaſ vẽderdes nẽ doardes ‘and you will neither sell them nor give them as a dowry’, nẽ nas dar a noble caualeyro ‘nor to give them to the gentle knight’; dalos o abbade a quen nos laure ‘the abbot gives them to whoever can cultivate them’, poſerõno todo ‘they put all of it’ (see also Martins 1994, pp. 217–218). This analysis appears, then, to provide a compelling evidence of historical ties between old speech habits and their present-day outcomes in the North of Portugal and in Galicia.

Still, the scriptural production of the subsequent centuries reveals the decline in the number of diphthong-final or nasal-final words that allowed nasal epenthesis in ensuing preverbal pronouns. In the 16^th-century Crónica dos Reis de Bisnaga, concurrent realizations are interspersed only after the relative/interrogative quem and the negative marker não. The query for the right-hand side context of sem and nem has brought hits involving vowel-initial accusative pronouns only. These results are indicative of an increasing role of morphosyntactic conditioning in the processing of preverbal clitic pronouns. Conversely, phonological criteria ceased to be felt as crucial. The examples below, retrieved from the Corpus Informatizado do Português Medieval (Xavier, 1998, coord.), illustrate the point.

The primacy of the morphosyntactic conditioning at the expense of the phonology-based one reached its peak in the 17^th century.⁶ The query of the resources of the Biblioteca Nacional de Portugal – Biblioteca Nacional Digital (BND), encompassing 11 integral texts published between 1614 and 1697,⁷ brought a single occurrence of nasal-initial 3^rd person pronoun in the preverbal domain. It comes from a compilation of judicial regulations, the Regimento dos Juizes das Aldeas … (1639), which displays an antiquated grammar (multiple interpolations, i.e., more than one constituent separating the clitic from the verb; see Martins 2016b, pp. 421–423, the inflected infinitive used with an injunctive value; Carvalho 2015, pp. 669, 673, and variable past participles in compound tense forms).

Thus, dwindling preverbal contexts in which nasal epenthesis used to be documented pose several problems to the continuity hypothesis. Obviously, the disjunction between the speech production of the inhabitants of remote areas of the Entre Minho e Douro region and the grammar of the then urban centers might not be excluded. Likewise, a given grammatical habit is not unlikely to resume after a period of dormancy (see D’Alessandro, 2014 for the unexpected advent of annə, borrowed from the Italian compound tense auxiliary, to replace an almost extinct impersonal pronoun nomə < Lat. homo in Abruzzese, a southern Italian variety spoken in the central part of the region). Be that as it may, the scarcity of attestations of the nasal-initial allomorph in the scriptural production of the 17^th-century EP must not be overlooked. For the time being, an uninterrupted transmission of the nasal epenthesis in preverbal pronouns over successive generations of speakers is no more than a guesswork.

Another hindrance has to do with the archipelago of the Azores, the second area with a noteworthy presence of nasal-initial preverbal clitics. Written records of the history of the vernacular clitic grammar on particular islands are badly missing. Starting from the 15^th century, the Azores are known to have been populated by settlers coming prevailingly from mainland Portugal. “A maioria dos povoadores seria proveniente de Portugal continental, merecendo particular referência os contributos do Algarve, Alentejo e Minho; no entanto, tais contributos não terão ocorrido de igual modo nas diferentes ilhas” (Lima, 2009, p. 229, ‘the majority of settlers appear to have come from the mainland of Portugal, with Algarve, Alentejo and Minho having provided most significant numbers of immigrants; nevertheless, newcomers from these regions are thought to have not been evenly divided among different islands’; translation – MN). The historical continuity hypothesis would require the speech habits of the majority of migratory waves to have been superseded by the grammar of the settlers from the Minho region. As for now, no evidence for this direction of grammatical change is available.

Putting aside the lack of sufficiently probative documentation across various places and periods, the representativeness of the historical sources proves an even more serious obstacle. What is more, the analyses highlighting language acquisition as the primary driver of language change and conducted in terms of internalized diglossia and parameter reranking require samples of infant-directed speech in successive generations of speakers to be systematically compared with one another. In an attempt to remedy the lack of necessary data, Kodner (2019, pp. 4–12) tries to infer the characteristics of infant-directed speech on the basis of modern and historical literary corpora. His results rely on the sameness of frequency values in both types of sources. In particular, high token frequency elements are expected to be similar irrespective of the addressees and communicative circumstances. Moreover, the extension of learning algorithm sampled from the two kinds of data consistently points to the same learning outcomes. Therefore, the core cues acquired in childhood are thought to be plausibly reconstructible on the basis of non-infant corpora. Meanwhile, Kodner’s pioneering study seems to be concerned primarily with the acquisition of the lexicon. It remains for future investigation to see whether these results can successfully grasp the evolution of TCP sequences in previous generations of EP learners.

7.2. Nasal epenthesis as an externally-motivated analogical change

Instead of capturing variation documented in CORDIAL-SIN in terms of historical continuity, a more moderate solution is advanced here, making use of the concept of markedness. Rather than positing the transmission of nasal epenthesis in preverbal accusative clitics from one generation to another, it emphasizes the sameness of the processing of TCP sequences at various moments in the history and in various communicative circumstances. This theoretical apparatus has been inspired by H. Andersen’s actualization theory.

For Andersen, grammatical innovations are either internally (from below) or externally motivated (from above). “Typically in internally motivated change, an innovation is extended to unmarked contexts earlier than to corresponding marked contexts” (Andersen, 2008, p. 36). If speech act circumstances are considered, unmarked contexts correspond to spontaneous communication and informal text genres close to orality. In morphosyntax, unmarked contexts are routinely basic and most frequent ones, e.g., main clauses as opposed to subordinate clauses, simple vs. compound tenses. Solutions initiated in these contexts can subsequently gain ascendancy in marked situations and marked morphosyntactic settings, thus ceasing to be felt as informal or colloquial. Innovations from above, on the other hand, depart from marked contexts, both pragmatically and grammatically. They subsequently exert pressure (i.e., gain currency and become largely adopted in a given speech community) on spontaneous communication. This scenario materializes in a situation of language contact (e.g., transfers from the standard language into the dialects, syntactic calques from a foreign language of prestige; see Andersen, 2006, pp. 68–70). In morphosyntax, more elaborate solutions tend to spill over simpler contexts. With the passing of time, the elevated character of these innovations is no longer perceived. At this point, they become unmarked (Andersen, 2001, p. 238).

In this paper, communicatively unmarked contexts correspond to the speech production of CORDIAL-SIN informants. Onset insertion in preverbal accusative pronouns is likewise a feature proper to an oral, unmarked register. By contrast, the absence of nasal-initial forms is thought to be a learned variant (externally-motivated innovation), induced by contact with standard EP. Present-day Portuguese dialects must not be seen as repositories for phonological layers characteristic of Galician-Portuguese or Middle Portuguese texts. Instead, the reverse is thought to be the case: nasal epenthesis in preverbal accusative pronouns documented in medieval and Renaissance texts is thought to have reflected speech habits characteristic of the oral register. Obviously, the degree of communicative markedness of Galician-Portuguese and Portuguese texts prior to the 17^th century outranks the markedness of spontaneous dialectal speech. Still, since these texts contain instances of nasal epenthesis, their authors must have been exposed to two divergent grammars, with empty onsets alongside nasal onsets. Thus, pragmatically, the use of nasal-initial allomorphs in Galician-Portuguese or Middle Portuguese is considered an innovation advancing from below.

To conclude, it remains to ascertain whether consonant-initial preverbal accusative pronouns are unmarked also in morphosyntax. The answer depends on how variation in clitic placement in today’s EP is characterized, and that is not a consensual issue. Some authors claim that proclisis is triggered by the properties of the (higher) left periphery hosting complementizers and topicalized or focused elements (Madeira, 1992; Ledgeway, 2016, p. 1013). Other scholars point to proclisis triggers as the elements that have the property of attracting the clitic to the pre-Infl domain (Duarte & Matos, 2000). As a result, for some scholars proclisis is a less marked pattern and enclisis corresponds to a more complex derivation, with the V movement to a higher functional category. In other proposals, enclisis is primary and proclisis is computationally costlier (a last resort Movement; see Duarte & Matos, 2000, pp. 135–137), because it is activated by specific syntactic elements.

The puzzle can be solved if data from EP L1 acquisition are looked into. Enclisis is the dominant placement in children until approximately forty-two months of age (Duarte et al., 1995). In that period, enclisis is nearly indiscriminately applied, regardless of the presence of proclisis triggers in a sentence (Costa et al., 2015, pp. 14–17; Costa et al., 2016, pp. 440–446). Therefore, enclisis, alongside phonological shifts accompanying it, is thought to be the default solution. By contrast, proclisis is the marked pattern. Nasal epenthesis in preverbal accusative pronouns is an overgeneralization of a sandhi phenomenon from enclisis (unmarked context) in syntactically more complex environments (proclisis; marked context). Thus, in morphosyntax the innovation progresses from below as well. Syntactic conditions under which vowel-initial preverbal accusative pronouns must be used after diphthong-final words require an explicit learning. This variant appears to be problematic for some CORDIAL-SIN informants.