PREVIOUS LITERATURE

In an attempt to better understand the diachronic trajectory of Portuguese past participles, Chagas de Souza (Reference Chagas de Souza2007, Reference Chagas de Souza2011, Reference Chagas de Souza, Alves and Ribeiro de Jesus2015) traced the development of the thematic (regular) and athematic (irregular) participles from Latin to modern Portuguese. He claimed that a three-phase evolution has occurred, yielding the present forms. First, Portuguese inherited and maintained the doublets already found in Latin, such as aceitado - aceito (‘accepted’). Second, this pattern was extended, via analogy, to other first conjugation verbs where the 1SG present indicative form was adopted as the irregular participle form (and as an adjective with copular ser/estar), as, for example, chegado - chego (‘arrived’). Finally, innovative forms such as pedido - peço (‘requested’) have appeared recently with other verbs that normatively are not considered to be members of the class of verbos abundantes. Footnote ⁶ Simões (Reference Simões2008:356) also recognized, at least implicitly, the importance of the 1SG present indicative form to the formation of the irregular participle.

Nevins and Rodrigues (Reference Nevins and Rodrigues2014; see also Nevins, Reference Nevins, Aboh, Schaeffer and Sleeman2015) conducted several “wug” (cf., Berko Gleason, Reference Berko Gleason1958) experiments to test how argument structure affected participle choice. In order to avoid “token frequency in the language or other ungodly factors,” they invented Portuguese nonce verbs, as well as their corresponding nonce regular and irregular participles, and created contexts that would make each verb unambiguously unaccusative, transitive, or ditransitive (Nevins, Reference Nevins, Aboh, Schaeffer and Sleeman2015:193). Participants were presented with a discourse context and given a sentence with the option of inserting a regular or irregular participle of the nonce verb. The authors found that regular participles are greatly preferred overall, but the 1SG present indicative irregulars are used slightly more with unaccusatives as opposed to the other verb types. It is difficult to know, however, which specific properties of the verb the participants were reacting to in the experiment: the type of verb, as the authors intended, the form that the irregular took, or other factors such as unintended similarity to existing verbs/participles in the language. In the present paper, we therefore have chosen to confront the “ungodly factors” head-on with corpus data.

Powers (Reference Powers1985) investigated the variation between regular and irregular participles via a questionnaire study with 16 native Brazilian Portuguese speakers. With respect to perfect constructions specifically, he found that his respondents chose the irregular forms a mere 18% of the time. Crucially, however, Powers included morphologically irregular participles that do not have a regular counterpart (e.g., aberto, escrito); only four of the irregular verbs in perfect constructions that respondents deemed acceptable (eleito, entregue, pago, gasto) had a regular counterpart found in the language. Powers concluded, much like we do here, that existing prescriptive and descriptive characterizations of participle choice are inaccurate and unreflective of actual usage, despite the fact that he did not investigate naturally-occurring data. Below, we show that the rate of irregular participles in naturally-occurring spoken/written Portuguese is much higher than reflected in the results of Powers’ questionnaire study, and shows clear constraints on variation that it left unexplored.

Additionally, several previous variationist studies have examined naturally-occurring usage of these different participle forms and the factors governing their selection. Miara (Reference Miara2013) examined newspaper articles from the Diário Catarinense in Florianópolis in southern Brazil, extracting 843 total tokens. However, in her analysis, Miara included verbs showing no variation between regular and irregular participle forms. Once these invariant tokens are removed from her dataset, 566 tokens showing variation remain, coming from seven different verbs. From these verbs, Miara found a very high percentage (83%) of irregular participle use. Throughout her study, she suggested possible explanations for differences in usage from participle to participle, claiming that the alternation among different auxiliaries is gradient. Our results below confirm this view but with a larger dataset and more comprehensive multivariate analysis.

In another variationist study, Teixeira da Silva (Reference Teixeira da Silva2008) analyzed approximately 670 tokens from a corpus of spoken southern Brazilian Portuguese. Her results show that the regular participle is favored over the irregular form; however, only eight of these tokens, coming from only two verbs, displayed variation between regular and irregular participles with ter/haver. As a result, it is impossible to draw any firm conclusions from this study. Lastly, Mendes (Reference Mendes2005, Reference Mendes and Walker2010) analyzed the overlap between perfect versus progressive forms from a variationist perspective. Looking at synchronic and diachronic use in two corpora of spoken Portuguese from São Paulo and Rio de Janeiro, he found that the perfect form (ter + participle) is becoming more restricted both linguistically and socially when compared to the progressive (estar + present participle). Mendes did not, however, look at variation between regular and irregular participles within the perfect construction.

The most recent study of variability between regular and irregular participles is Thomas (Reference Thomas2018), who focused on European Portuguese. Analyzing the rate of irregular use in 10 different verbs in a corpus of written internet data, she found that seven favor the irregular form compared to three that favor the regular participle. More importantly, however, she also conducted a study of acceptability judgments and found that speakers heavily prefer the irregular forms over the regular forms for six verbos abundantes. Our results below suggest that this preference for the irregular forms may also play a role in the analogical spread of the irregulars, based specifically on the 1SG present indicative form.

RESEARCH QUESTIONS AND HYPOTHESES

Based on the findings of previous research on participles in Brazilian Portuguese, we formulated a number of research questions to guide our analysis. The overarching questions posed in this study are as follows: (1) What is the distribution of irregular/regular participles in constructions with ter/haver + past participle?, and (2) What are the factors that condition the variation between these two participle variants? In this light, we put to the test the normative rule—found in both prescriptive and descriptive grammars, including some authored by linguists—which claims that only regular participles ending in -do should occur in perfect constructions with the auxiliaries ter or haver (and the irregular form appearing exclusively in passive constructions and as adjectives).

We also hope to shed light on possible dialectal differences between Brazil and Portugal, as well as on diachronic differences to see whether there has been a shift toward more irregular participle use. We expect to see dialectal differences between Brazil and Portugal in light of the finding from Thomas’ (Reference Thomas2018) acceptability questionnaire that speakers of European Portuguese consistently preferred irregular over regular forms, as well as the prescriptive promotion of irregulars in school (Patrícia Amaral, personal communication). Additionally, we predict diachronic shift toward more irregular usage, especially when the irregular is identical to the 1SG present indicative form, given Chagas de Souza's (Reference Chagas de Souza2007, Reference Chagas de Souza2011, Reference Chagas de Souza, Alves and Ribeiro de Jesus2015) discussion of innovative irregular forms recently attested in Brazilian Portuguese, which found that these new forms are created via analogy with the morphological basis of the 1SG present indicative, even when the indicative forms themselves are irregular (see also Thomas, Reference Thomas1969:223). In view of the prior linguistic accounts summarized in the previous section, the results of our quantitative analysis of the data lead us to side with Chagas de Souza's qualitative arguments that analogy with the 1SG present indicative best explains the variation. We conclude more generally that probabilistic constraints are operating on analogical processes in grammar in both synchrony and diachrony (cf., Albright, Reference Albright, Blevins and Blevins2009).

OVERALL RESULTS: BRAZIL AND PORTUGAL

A total of 1077 tokens were included in the analysis, with all of them coming from verb lemmas showing variability with regard to participle form in the Corpus do Português. The initial results are contrary to the claim made in most grammars that the auxiliaries ter and haver combine with regular participles only. In fact, more than half of the examples in our data contain irregular participle forms in perfect constructions, as shown in Table 1.

Furthermore, dialectal differences show that there is an even greater preference for irregular participles in Portugal than in Brazil, as illustrated in Table 2. The use of irregular participles in Portugal is 15% higher than that found in Brazil, a difference that is significant by a chi-square test.

Table 1.

Overall frequencies of Portuguese participle form

Table 2.

Frequencies of participle form across dialects in the 20^th century data (n = 711)

Figure 1 below shows the 21 verbs (of the grand total of 58 that we searched for, based on lists from grammars and language textbooks) that showed variation in participle form within the twentieth century data in the Corpus do Português.

Figure 1.

The 21 variable verbs in data set (listed by infinitive) with rates of irregular participle selection, from highest to lowest.

As Figure 1 illustrates, there are enormous differences in the relative frequencies of irregular participle choice depending on the particular verb in question. Indeed, the rates of irregular participles span nearly the whole range of possibilities from a mere 4% for empregar ‘to employ’ and envolver ‘to wrap’ to a near-categorical 97% in the case of pagar ‘to pay.’ Perhaps most notable is that the three verbs with the highest percentage of irregular participles, pagar, ganhar, and gastar, display a combined rate of 94% irregular participles (377/401). Moreover, these three verbs actually provide 64% (377/584) of all of the irregular tokens in the full dataset. This means that the other 36% (207/584) of the total set of irregular tokens are distributed over 18 different verbs. But even in this latter set of tokens there is a decided lack of balance, inasmuch as three other verbs (entregar, salvar, matar) account for 57% (119/207) of the remaining irregular participles, and therefore 15 verbs account for the remaining 88 tokens. A grand total of six verbs (i.e., 28% of the 21 verb types showing variation), therefore, account for 85% (496/584) of all the irregular participles in the dataset. Such large disparities in the overall frequencies of verbs—not only in terms of irregular versus regular participle choice, but also of the widely varying frequencies with which they occur in perfect constructions—have gone essentially unnoticed in the prior literature on participle choice.Footnote ¹¹ Another point to notice about the three verbs with the highest rate of irregular participle usage is that each has an irregular participle that is identical in form to the 1SG present indicative form of the same verb. This coincidence in form will prove important in the analysis below and for our argument that analogical processes are at work in regulating the variation.

In order to see beyond the strong dominating effect of the individual verb in the variation, we performed a mixed-effects regression analysis in R (Bates, Reference Bates2010), including a random slope for lexeme by dialect.Footnote ¹² The best-fitting model of the twentieth century data (as determined by a log likelihood comparison of nested models), is presented in Table 3 below.

Table 3.

Results from the mixed-effects logistic regression model (20^th century data, both dialects) predicting irregular Portuguese participle forms (n = 711). Also included: Auxiliary verb, Person, Number, Polarity

There is a significant difference between Brazil and Portugal, with Portuguese speakers/writers selecting significantly more irregular participles, as expected given the statistically significant difference between the two dialects reported in Table 2. No other fixed effect was significant. In order to more closely examine the hierarchical rankings of constraints on participle choice, as well as the interactions between those constraints, we generated a conditional inference tree (Figure 2). Here, we follow Tagliamonte and Baayen (Reference Tagliamonte and Baayen2012:164), who argue that “conditional inference tree[s] offer an ideal tool to be used in tandem with the mixed-effects logistic model to come to a full understanding of the quantitative structure of a dataset and, as a result, an optimal interpretation of the variation.”

Figure 2.

Conditional inference tree showing the constraint hierarchy for participle choice (20^th century data) (n = 711).

As Figure 2 shows, the form of the irregular participle emerges as the top-ranked constraint: irregular participles are significantly more probable when they match the form of the 1SG present indicative form of the verb. Note that, in Figure 2, we include type of irregular participle rather than lexeme for ease of presentation, since the former is binary whereas the latter consists of 21 different verbs. In the mixed-effects regression reported in Table 3, however, lexeme was included as a random effect and therefore type of irregular participle could not be included without issues of collinearity.

On the right-hand first branch of the conditional inference tree, where the verb's irregular participle form is distinct from the 1SG present indicative, dialect emerges as a significant predictor, with significantly higher rates of regular participles in Brazil than in Portugal. Put another way, Portugal shows greater use of irregular participles even when the form of the irregular participle is not identical to that of the 1SG present indicative. Regarding the selection of regular or irregular participle forms based on the irregular participle form's shape (i.e., whether or not the participle coincides with the 1SG present indicative), the data are distributed as follows (Table 4):

Table 4.

Distribution of Portuguese irregular and regular participles based on the irregular participle's shape

These results reveal that, when the form of the irregular matches that of the 1SG present indicative, the distribution of irregular and regular participles is the exact opposite of when the irregular form's shape is of some other type. Thus, when the irregular form shares the shape of the 1SG present indicative, it has an overall usage rate of 69%, but when it is does not share that shape it only occurs 32% of the time. The actual morphological shape of the participle, then, exerts a strong influence on whether speakers/writers will choose the irregular form.

DIACHRONIC RESULTS: PORTUGAL

Turning to our diachronic analysis of European Portuguese across the nineteenth and twentieth centuries, we provide the results for the overall distribution of the regular and irregular participles in Table 5. Here we only examined the participles that occurred variably in the twentieth century data (see Figure 1) and compared them with their rates in the nineteenth century.Footnote ¹³

Table 5.

Distribution of irregular and regular participles across the 19^th and 20^th centuries (European Portuguese)

As these numbers show, there is a clear and statistically significant shift toward higher rates (28% higher) of irregular participle use from the nineteenth to the twentieth century. We see a more fine-grained reflection of this change when we examine the individual verbs, as illustrated in Figure 3, which shows the seven most frequent variable verbs in the European Portuguese data.

Figure 3.

Rates of irregular participle usage in 19^th and 20^th centuries among the seven most frequent verbs.

As Figure 3 shows, some of the verbs remained essentially unchanged from the nineteenth to the twentieth centuries, as in, for example, gastar ‘to spend’ and pagar ‘to pay,’ both of which already displayed high rates of irregular participles in the nineteenth century. Other verbs, however, showed drastic increases in the overall usage of the irregular form (e.g., aceitar ‘to accept’ and salvar ‘to save’). Just as impressively, none of the verbs showed an increase in the use of the regular participle across the two centuries. There was either an increase in the irregular and corresponding decrease in the regular form, or there was little to no change, as in the case of those verbs which already had high relative rates of the irregular participle in the nineteenth century. The overall trend, then, is clearly toward greater use of the irregular form no matter which verb is being employed.Footnote ¹⁴

When we included these predictors in the fixed-effects model (see Table 6 in the appendix), we found that both century and type of irregular participle were significant predictors of choice between regular and irregular participles in the European Portuguese data.Footnote ¹⁵ The conditional inference tree (Figure 4) provides a visual representation of the hierarchical relationship among these independent variables.

Figure 4.

Conditional inference tree showing 19^th and 20^th century participle choice results for Portugal (n = 768).

Just as in the overall dataset above, when the irregular participle shares its form with the 1SG present indicative, there are significantly more irregular participles in the twentieth century than in the nineteenth. However, when the irregular participle shape is of some other type, there are significantly fewer irregular participles used in the nineteenth century.

Taken as a whole, the results presented in this section evince a clear change across the nineteenth and twentieth centuries in Portugal toward greater use of irregular participles in perfect constructions with ter and haver. Additionally, a comparison of those participles identical in form to the 1SG present indicative with those that are not shows that the increase in use of irregular participles was faster for the latter group; that is, irregulars identical to 1SG increased by only about 20% (comparing the light gray areas in nodes 6 and 7 of Figure 4) from the nineteenth to twentieth centuries, while irregulars not identical to 1SG increased appreciably more (nodes 3 and 4). This could indicate that irregular participles began with forms identical to 1SG and then spread quickly to other verbs without this formal overlap, a typical pattern for actualization/analogical generalization (Malte Rosemeyer, personal communication). More detailed diachronic analysis would be necessary, however, to corroborate this process.