3.1. Coding

The data for the present analysis were coded for the following predictors:¹⁴

(i) Person-number: The effect of subject person-number on pro-drop has been reported in numerous studies. These include partial pro-drop languages such as Spanish and Brazilian Portuguese, and radical pro-drop languages such as Mandarin Chinese. The effects observed display differing patterns. In a number of studies of Spanish varieties, plural person-numbers favor the null form, whereas singular person-numbers tend to favor the overt pronoun (Carvalho & Child, 2011, p. 19; Orozco & Hurtado, 2021, p. 14; Otheguy & Zentella, 2012, p. 790).¹⁵ Yet, in Afro-Brazilian Portuguese, Lucchesi (2009, p. 177) found plural person-number and second-person singular to favor the overt pronoun, whereas the rest favored the null form. Duarte (1995, p. 48), studying Rio de Janeiro Portuguese, found a similar result, with the exception that third-person plural also favored the null form. Finally, in studies of the radical pro-drop language Mandarin Chinese, the null form is favored by second person plural and third person plural, whereas singular person forms and first person plural generally favor the overt pronoun (Jia & Bayley, 2002, p. 110; Li & Bayley, 2018, pp. 151–153; Li, Chen & Chen, 2012, pp. 103–105).

In the study of PK subject pronoun expression by Silva (2020, pp. 89–91), person-number was not selected as statistically significant, although its profile appeared suggestive of a conditioning effect. The present study re-considers person-number with the smaller and more orthogonal database in order to clarify the constraint status of person-number, and re-assess the hypothesis of Neeleman and Srendöi (2007) that agglutinating pronominal morphology and pro-drop are connected. In the PK pronoun system, the plural forms of the second and third person may be considered non-fusional. These are, respectively, bolotu (<bo(s) + otru ‘other’) and elot(r)u (>eli + otru), and are derived from the corresponding singulars, bos and eli by addition of the segment otru, that might be interpretable as a plural marker. In the present analysis, person-number values were coded as singular versus plural forms.

(ii) Structural location of the variation: The effect of structural position of the subject has been noticed in a number of studies on different languages, including for example Hebrew (Melnik, 2007), Russian (Bizarri, 2015), and in variationist studies of Spanish (Guy & Orozco, 2008; Otheguy & Zentella, 2012, Torres-Cacoullos & Travis 2019, among others), Portuguese (Genuíno, 2017; Menon, 2000, among others) and Cape Verde Creole (Rodríguez-Ricelli, 2021). Differential effects are reported for subjects of main clauses, and coordinate and subordinate clauses, with the null variant frequently reported as prominent in the latter two structures.

Torres-Cacoullos and Travis (2019) have drawn attention to a graded effect of different types of clause-linking on null subjects in Spanish and English, noting differential roles for syntactic and prosodic linking in both languages. In the present study, re-assessment of this factor group investigates the possibility of a graded effect, and is further motivated by the exclusion of tokens coding instances such as those of examples (1) and (2), involving non-finite clause direct objects or purposive clauses of bitransitive verbs. The effect of linking is assessed by means of the following factors: (i) antecedent in an independent sentence, as in (3); (ii) conjunction by formal link, as in (4); (iii) first juxtaposed clause, conjunction by prosodic link alone, as in (5); (iv) complement subordination by prosodic link only, as in (6),¹⁶ and adverbial subordination, as in (7):

(iv) Priming – Previous mentions of the subject and form of the reference: Since the early 1990s, many studies have noted that repetition is a significant factor for recall maintenance (Lucchesi, 2009; Paredes Silva, 2003; Scherre & Naro, 1991; Travis, 2007). Although in variation studies of subject pronoun expression in Portuguese and Spanish co-reference is rarely the first selected independent variable, variation studies of Mandarin-Chinese are unanimous in this respect (Jia & Bayley, 2002, p, 109; Li & Bayley, 2018, p. 150; Li, Chen & Chen, 2012, pp. 103–104). Yet, these studies do not consider the actual form of the antecedent. They merely report that antecedents favor the null subject, whereas a new reference favors a pronoun (Li & Bayley, 2018, p. 150; Li, Chen & Chen, 2012, p. 104).

Where the form of the prior reference is concerned, while variation studies of Spanish and Portuguese report subject pronouns or null subjects to be favored, respectively, by the prior same representation, Lucchesi (2004) and Novaes (2007) both concerning varieties of Brazilian Portuguese, found a prior NP or a prior null subject to favor a subsequent null subject pronoun. For Cape Verde Creole, Rodríguez-Ricelli (2021, p. 150) also confirmed this effect.

Similar findings to those of Lucchesi (2004) and Novaes (2007), were reported for PK in Silva (2020, pp. 73–76). Nevertheless, in that study, <form of antecedent> and <occurrence of an antecedent> were treated as separate predictors. In contrast, in order to improve orthogonality, the current study coded for a single predictor addressing both antecedence and form of antecedent according to whether the subject was previously mentioned, a pronoun, example (8), as an NP, example (9), or as a null subject, example (10), or whether not mentioned previously, example (11):

(v) Verb semantic class: The effect of verb semantic class on subject pronoun expression has been widely observed in studies of Spanish (e.g. Bentivoglio, 1989; Erker & Guy, 2012; Orozco, 2015; Travis, 2007) and Brazilian Portuguese (e.g. Carvalho & Child, 2011; Silveira, 2012)¹⁷ Speech verbs, epistemics, perception and stative verbs favor the subject pronoun, whereas all other verbs favor the null variant (Silveira, 2012, pp. 95–100; Orozco & Hurtado, 2021, pp. 3–4).¹⁸ Silva (2020, pp. 83–85) confirmed these broad findings for PK with a predictor containing six verb classes, some of which registered similar weights. Following the revisions to the database, it was decided to re-run the analysis limiting the range of classes.

The present study codes verb classes following the approach of Erker and Guy (2012), reducing this factor group to three factors: a mental activity class (e.g. lembrá ‘to think, to imagine’, intendé ‘to understand’, keré ‘to want, to desire’, a stative class (e.g. teng ‘to have’, fiká ‘to stay, to live’, gostá ‘to like’, sabé ‘to know’) and an external activity class (e.g. kuré ‘to run’, komprá ‘to buy’, skribé ‘to write’, prendé ‘to teach’). This coding takes into account the broad effect of the stative/dynamic opposition observable in PK across the more complex factor group analyzed by Silva (2020), plus the notion that verbs of the mental activity type may promote individualization of the subject via the pronoun (Enríquez, 1984; Erker & Guy, 2012).

(vi) The presence of a TMA marker: In variation research of pro-drop in Spanish and Portuguese, a number of studies have found that different tense-aspect forms effect pronoun variation, with the following frequently observed ranking profile conditioning the null subject: preterite > present > imperfect > conditional/subjunctive (e.g. Bayley & Pease-Álvarez, 1997; Carvalho, Orozco & Shin, 2015; Silva-Corvalán, 1997; Silveira, 2012). Silva-Corvalán (1997) proposed that Spanish TMA forms play a conditioning role by virtue of their pragmatic function in discourse: both the preterite and present are factual and assertive, but the preterite is dynamic, event-focussed, and foregrounded, whereas the present tense is not always dynamic or focal. In contrast, other TMA forms, such as the imperfect, conditional and subjunctive are non-assertive, non-dynamic, irrealis and backgrounded. It is hypothesized that it is in the foreground that more null subjects are expected, where the event is focused, yet in the background, where the event is not focused, more subject pronouns are expected (Silva-Corvalán & Enrique-Arías, 2017, pp. 180–181).

Whereas Spanish is a tense-prominent language, PK is aspect-prominent. The Kristang verb functions primarily in terms of aspect, represented by four pre-verbal TMA markers,¹⁹ which perform six basic functions (Baxter, 2012b, pp. 124–125). Two markers have unique aspect functions: ja ‘perfective’, as in (12), ta ‘progressive’, as in (13); and the other two have double functions: (i) lo(gu) ‘future-irrealis’ and ‘habitual aspect’, the latter is shown in (14); and, (ii) Ø ‘imperfective continuous’ and ‘perfective’, as in (15). The markers ta, ‘progressive’, lo(gu) ‘habitual’, and Ø ‘imperfective continuous’ are transparent for past and present. However, in negated sentences, the aspectual markers are replaced by the negator ngka occurring with verbs representing perfective or imperfective aspect. In example (10) above, the sentence Ø ngka tomá buku prendé ‘I didn’t take the book to learn’, presents one such instance, where the aspectual value of the verb, as perfective or imperfective, would be recuperated from the prior discourse.

Silva (2020, pp. 85–88) observed a conditioning effect in PK null subject pronoun use according to the four forms of the TMA marker, potentially related to TMA function and discourse structure. In order to clarify whether the variation is due to the presence of an overt TMA marker, or whether it is functionally based, the present study adopts a slightly different approach, and considers the individual TMA marker functions plus the effect of the negator ngka.

(vii) Individual speaker: This factor group assesses the stability of the predictors among individual speakers regardless of the potential predictions of <age-group> or <gender> (Gorman, 2010; Johnson, 2009). In the study of PK pro-drop by Silva (2020, pp. 92–93), it was found that the effect of speaker sex was not statistically significant, although, numerically, women deleted marginally more than men, whereas speaker age-group was found to exert a very weak conditioning effect suggestive of an apparent-time change towards less use of the null subject.

4.1 Co-reference and prior form

The effects of priming are presented in Table 2, where it is seen that two factors favor the null subject: a null antecedent is quite favorable (pr. 0.814), whereas an NP is moderately favorable (pr. 0.623).

However, a prior pronoun is slightly unfavorable (pr. 0.452), perhaps because of the possibility of ambiguity of reference. In contrast, the absence of a prior mention is quite unfavorable to the null subject pronoun (pr. 0.144), as it is expected that the subject of a first mention would be realized so that the interlocutor can properly identify the reference, unless this identification is facilitated by pragmatic considerations.

On the one hand, these results confirm the role of priming in conditioning the dependent variable, with null antecedents leading to subsequent null subjects and pronominal antecedents leading to subsequent pronominal subjects. This coincides with the findings of variation studies by other researchers. Unfortunately, existing variation studies of subject pronoun expression in Chinese have not addressed the specific prior subject forms in priming. However, in studies of Brazilian Portuguese, the null antecedent favors a subsequent null subject, whereas an antecedent subject pronoun registers a weak disfavoring of a subsequent pronominal subject or registers a neutral value (Lucchesi, 2009, p. 180; Novaes, 2007, p. 170). Furthermore, variation studies of varieties of Spanish reveal a similar result, with some researchers reporting parallelism for both factors: null antecedents prime subsequent null subjects while pronominal antecedents prime the subsequent pronominal subjects (Carvalho & Bessett, 2015, p. 158; Orozco, 2015, p. 29; Torres-Cacoullos & Travis, 2019, p. 17; Travis, 2007).

Basically, the subsequent subject is licensed on the basis of the identity strength of the previous subject, constituting a referential sequence. Having already activated a particular representation of the subject, the speaker, accessing the previous reference, has no need to activate a new full (i.e. phonetic) representation for the subsequent identical subject, unless clarification of the reference were required.

Inspection of the prior NPs in the corpus reveals that there are only three instances in which an NP is not a first reference. Hence, the overwhelming majority of prior NPs constitute switch references. While we noted above that the prior NP may favor a subsequent null subject, it is obvious that references subsequent to an NP may also be represented by pronouns. Thus, in PK, the initial stage of a chaining effect anchored by an NP can be immediate, licensing the null subject, or prolonged, licensing first a pronoun and subsequently a null subject.

4.2 Structural location relative to the antecedent

Table 3 presents the results of the analysis of the five factors of the predictor addressing location of the subject. Two factors are favorable to subject pronoun omission: asyndetic coordination, registering a strong relative weight of 0.788, and the presence of the antecedent in an independent sentence, registering a weak favoring effect (weight 0.546).

In the instances of asyndetic coordination, the antecedent is in the first clause immediately preceding the subject of a first juxtaposed clause, within the intonation contour of the complex clause. The tight linkage between the two clauses leads to a high degree of accessibility (pr. 0.788). However, when the preceding reference is in a different macro clause (i.e. clause separated by a pause and clause-final intonation), the effect is weaker (pr. 0.546). In this case, a previously referenced subject, if still fully accessible, could lead to a subsequent Ø, whereas a prior reference not entirely accessible (for example, moderated by factors such as the presence of a competing referent, or a change of action involving the same referent) could lead to a subsequent pronoun.

The remaining three factors are unfavorable to the null subject. Indeed, the results for subjects in coordinate and subordinate clauses suggest that omission may be restricted by structural considerations. In syndetic coordination, a null subject is disfavored with a relative weight of 0.409 when the subject is located in the first clause after a coordinator. Where complement and adverbial subordinate clauses (with the antecedent in the matrix sentence) are concerned, a weight moderately more unfavorable to null subject is recorded: 0.296. In this case, the syntactic embedding involving a subject with the same person-number as the subject of the matrix may be affected by the need to clarify reference.

There are certain similarities with the results of other studies. Research on Portuguese reports pronouns most likely to be expressed in nominal subordinate clauses, and either close to neutral or disfavored in coordinate clauses (Carvalho & Bessett, 2015, p. 158; Genuino, 2017, pp. 105–109; Novaes, 2007, p. 169; Silveira, 2012, p. 80). Some studies of Spanish varieties also report subordinate clauses as favoring pronoun expression, while coordinate clauses favor omission (Carvalho, 2015, p. 158; Orozco, 2015, p. 24). The preference for null subject in asyndetic coordination has also been noted in a variety of Brazilian Portuguese (Novaes, 2007, p. 169). Where independent clauses are concerned, a number of studies of Brazilian Portuguese and Spanish several studies also report independent clauses as registering a result close to neutral (Carvalho, 2019, p. 158; Carvalho & Bessett, 2015, p. 158; Genuino, 2017, pp. 105–109; Menon, 2000, pp. 170–171; Novaes, 2007, p. 169; Orozco, 2015, p. 24; Orozco & Guy, 2008, p. 77; Silveira, 2012, pp. 100–101).

4.3 Verb class

The results of the analysis of verb semantic class, presented in Table 4, reveal only a mild effect on the subject pronoun variable, with dynamic verbs slightly favoring the null subject, statives close to the neutral point²⁰, and mental activity verbs mildly unfavorable. The PK weighting profile is somewhat similar to that observed by Torres-Cacoullos & Travis (2019:671) for Colombian (Cali) Spanish,²¹ and Bouchard (2018:17) for São Tomé Portuguese, with the null subject occurring most with dynamics (Cali: pr. 0.57 (67%); São Tomé pr. 0.55 (72.1%)) and statives and mental activities unfavorable, respectively, in that order (Cali: pr. 0.42 (53%), pr. 0.31 (35%); São Tomé pr. 0.53 (64.3%), pr. 0.43 (61.8%)).²²

We noted earlier that researchers have recently begun to question the validity of a predictor based purely on verb semantic class. Torres-Cacoullos and Travis (2019, pp. 21–23) take the discussion a step further, pointing to the effect of verb semantics in discourse structure. In relation to temporal sequencing, they propose that the favorability of dynamic verbs to null subjects involves subjects referring to temporarily sequential or simultaneous central events or situations relating to foreground discourse, whereas statives and epistemics relate more to background discourse (Torres-Cacoullos & Travis (2019: 21–23).²³ This proposal gains support from research on narrative structure and lexical aspect in relation to tense-aspect morphology (Bardovi-Harlig (2000; among others), that reveals the natural predominance of activity, achievement and accomplishment aktionsart classes in narrative foreground, as opposed to statives.

4.4 TMA marking

Initially, five factors were included for analysis according to TMA marker functions, yielding the results shown in Table 5. The future-irrealis function of lo(gu) was excluded because of low token numbers in the revised data-base.

The profile in Table 5 lacks clear indications of a discourse constraint on TMA relative to subject pronoun expression. For example, the result for the habitual aspect marker logu seems anomalous in terms of the proposal by Silva-Corvalán (1979) that imperfective values are more expected in narrative background where the overt subject pronoun would be preferred (Silva-Corvalán & Enrique-Arías, 2017, pp. 180–181).

However, comparing the overt TMA marked instances with the Ø-marked instances, the close proximity of values is suggestive of a structural effect, whereby overt filling of TENSE favors a null subject and covert filling favors a pronominal subject.²⁴ As such, a second analysis was conducted, reducing the group to two factors only: ‘overt’ versus ‘covert’. The result is shown in Table 6:

On comparison of the log-likelihood values of the two analyses, by chi² test, it was found that the difference between the two analyses is not significant.²⁵ In this case, the more economical analysis is preferred, whereby it appears that the recuperation of the prior referent of the null subject could be aided by the presence of the overt aspect marker. In the latter case, the listener may expect that the prior reference is continued.

4.5 Person-number

In the study of PK pro-drop by Silva (2020, pp. 89–91), person-number was not a significant conditioner of PK pro-drop. However, in the current analysis, person-number coded as singular versus plural was selected, yet it is the least influential of the linguistic variables considered. The results in Table 7 reveal only a slight favoring of null subject by plural person forms:

The profile is reminiscent of some of the findings of the Brazilian and Chinese studies noted in section 3.1, where singular person-numbers disfavor the null subject, whereas plural person-numbers favor it. Nevertheless, this result for PK does not provide clear support for the Neeleman and Srendöi (2007) hypothesis that a language only allows radical pro-drop if its personal pronouns manifest agglutination or some other nominal feature. Although PK presents agglutination in the contrast between both second and third person singular and plural pronouns, the second-person plural forms were few (15 tokens, yet all with pronominal subjects) and were excluded from the analysis. Thus, only non-agglutinating first-person plural nus and agglutinating third-person plural olot(r)u were considered.

4.6 The individual speaker

The results from the analysis of this predictor clarify the curiosities noticed by Silva (2020, pp. 88–89, 92–93) concerning the weak tendencies in the distribution of PK pro-drop according to speaker sex and age-group. Figure 1 presents the profile of the individual speaker weights:

Figure 1: Individual speaker, effect on null subject pronoun.

The individual rate of null subject pronoun expression ranges from 0.282 (F1C, female, age-group 1)²⁶ to 0.667 (FM3, female, age-group 3). There are 11 speakers with scores above 0.5 and 7 with scores below this point. Looking more closely at the sample, several interesting tendencies emerge, as do also certain notable individual differences.

Scores above 0.5: Among the 11 speakers above the 0.5 mark, there are four speakers of age-group 3: the three women of this group, and one man. The three women had lives restricted to the home domain. Speaker F3M, with the highest pro-drop rate, was born into a traditional fishing family in the Banda Praya community, and had no formal education. During interview, she commented that she had scant knowledge of English. PK was clearly her dominant language, although she also knew local Malay.²⁷ The other two women in this group (F3L and F3E) had fathers who were in employment connected with the British colonial presence, and had lived for short periods in other locations in East Malaysia. They had minimal education primary level, as did the male MT3 (considered below). For this age-group, it is important to bear in mind that all speakers, men and women, were born in other locations in Malacca, and principally in Praia Lane and Trankera, traditional residential locations where Kristangs lived in contact with Baba Chinese, speakers of Baba Malay, a pro-drop language (Lee, 2014, pp. 257, 263).

Also, within the above 0.5 range, there are four age-group 2 speakers: F2S, M2V, F2X and M2B. F2S and M2V stand out as they are ‘professionals’ (primary school teacher, and nurse) with secondary and specialist education, yet their pro-drop scores are above the 0.6 point. Both F2S and M2V had always lived in the Ujong Pasir Kristang community. F2S is from a traditional fishing family. M2V’s father was employed by the British administration as a lighthouse attendant, and was often absent from the home setting. The other two speakers in age-group 2, F2X and M2B, are from fishing families. F2X had a traditional home-domain existence and M2B had worked as a fisherman and as a gardener in government employ in Independent Malaysia. Both had minimal primary education. The fact that there is minimal difference in the scores of these members of age-group 2 might be explained by the fact that they grew up in a period in which PK was more widely spoken in the community, when there was greater community network cohesion and when there were more Kristang-dominant speakers.²⁸

Of the remaining two members of the above 0.5 group, M1F has the second highest pro-drop score in the data. This speaker was from a traditional fishing family and resided in a house with 16 members of extended family in a street that was known to most speak PK.²⁹ At the time of the interview, he worked in an electronics factory. The second age-group 1 speaker is M1U, from a traditional fishing family, with a father versed in Kristang oral traditions. He had primary education to level 3, was employed in a soft-drink company, yet also worked as a fisherman.

In sum, in the set of speakers above the 0.5 cut-off, there are some suggestions of an age-related tendency, as 8 of the ten speakers are in age-groups 2 and 3.

Scores below 0.5: Now, turning to the 8 speakers in this range, two facts stand out: (i) two of the three men of age-group 3 (M3P, M3W) score below 0.5, whereas MT3 scores just above the neutral point; and, (ii) four of the six speakers of age-group 1 are in this range, including all three females of this age-group. In age-group 3, the three men had primary education to level 3 or 4. However, they had once had employment with colonial institutions and businesses (M3T – rubber plantation clerk; M3P health inspection assistant; M3W – attendant, vehicle import company) where they interacted with a wider community and in use of English was important. M3T and M3W were also musicians, and occasionally worked away from the community. M3T and MP3 also participated in the British colonial military volunteer battalion during the Japanese occupation of former British Malaya and Singapore.

In age-group 2, speaker F2I (daughter of M3P) completed education to level 4 secondary,³⁰ subsequently married a Kristang government clerk and was thereafter occupied in the home domain. In contrast, M2D completed level 4 primary, spent 7 years in the army (under the British) and subsequently worked in the Malacca Public Works department before turning to full-time fishing.

Among the speakers of age-group 1, while F1K had only primary level 3 education, the educational level of the others is relatively higher, with F1A, M1O, F1C attaining secondary school level. F1A and F1K were occupied in the home domain, whereas M10 (government clerk) and F1C (teaching assistant, and sportswoman) were active beyond the community.

Overall, among the speakers with scores below 0.5, some facts are suggestive of a conditioning of pro-drop by age-difference and sex: the strong presence of age-group 1 speakers, especially females, whereas the presence of two men of age-group 3 (but absence of women of this age-group) might relate to a time when the role of Kristang women was very much restricted to the home-family and immediate community domains. On the other hand, exposure beyond the community, notably in occupations involving interaction and requiring language skills are suggestive of a negative influence on pro-drop.