Determining the functions of second-person pronoun forms cê / ocê / você in Brazilian Portuguese through acoustic analysis

This paper pr


Introduction
Brazilian Portuguese (BP) seco nd-person singular pronoun is reali zed in t hree forms: the so-called standard one, você, and the reduced forms ocê and cê.
These three forms and thei r functions have been l argely discussed in the literature, but only now, with the p ublication of the C-ORAL-BRASIL spontaneous speech corpus (Raso & Mell o 2012), is it possible to verify their actual usage.
Previous studies maintain that the reduced forms must be considered as products of g rammaticalization and, particularly regarding the form cê, that it must be considered as a cliticization of você (Vitral 1996;Ramos 1997;Vitral & Ramos 2006, 2008).In those studies, the atonicity of cê is determined through morphosyntactic criteria and it is assumed that cê is always realized as subject.
There are few corpus-based studies, none of them analyzing spontaneous speech data.
This paper proposes a new approach to the study of those forms that utilizes spontaneous speech data retrieved from the C-ORAL-BRASIL corpus.Our goal is (a) to de termine the t onicity or at onicity of the for ms cê/ocê/você based on prosodic criteria and (b) to establish the use, and hence the functions, of the three forms, with particular attention to the form cê.
We argue that tonicity must be examined in relation to prosody.Several studies indicate that the main correlate of lexical and phrasal stres s in BP is duration (Massini 1991;Barbosa 2000Barbosa , 2002;;Fontes 2013).We therefore assume that there is a strong correlation between duration and tonicity.Thus, we consider that longer phonetic syllables [ˡse] in the for ms cê/ocê/você, as compared to the mean durations of the phonetic syllables of their tone units, correspond to stressed syllables, while shorter phonetic syllables [se] in the same forms, as compared to the mean durations of the phonetic syllables of their tone units, equate to unstressed syllables.Our proposal is that the preverbal position, associated with subject function, is related to unstr essed forms, while postverbal positions, associated with object function, is related to stressed forms.

Literature review
The singular and p lural forms of BP secon d-person pronouns have been extensively analyzed, as e videnced, e.g., by the afore mentioned literature.Although different methodologies have been used, none of those studies had the opportunity to examine spoken data as it was done for the present study.Before presenting our own proposal, we will brie fly review the state of art c oncerning previous studies on second-person pronouns in BP.
A series of sy ntactic studies argue that cê is a clitic, or at least a form at a certain stage of clitic/expletive formation (Vitral 1996;Ramos 1997;Vitral & Ramos 2006, 2008).These studies are mostly based on examples of unspecified source or transcribed interviews.The idea that they advance is that there would have been a process of grammaticalization that departed from the archaic honorific form Vossa Mercê, passed through você and ocê, and finally reached the form cê, which is therefore considered as a clitic or an expletive form.Some authors claim that cê cannot occupy the same positions as t he forms ocê and você, nor can cê perform the same functions as the other two forms.That is to say, cê cannot be found (a) in post-verbal position, (b) after a preposition as an indirect object, (c) i n topicalization, or (d) in a f ocalized position.As we will soon show, our spoken data do not confirm most of those claims.
Following the tripartition proposed by Cardinaletti and Starke (1999), Petersen (2008) analyzes part of the d ata presented in previous studies from a different standpoint, concluding that cê corresponds to a weak form.
It is clear, t herefore, that there is a deep concern within the field of syntax studies about the definition of the form cê.
Other researchers, however, prefer t o observe the data rather than to define the forms.
A set of acoustic experiments on the reduced pronoun forms (Ciríaco, Vitral & Reis 2004) tested contextual phonetic variables in order to obtain data on the intensity and duration of s uch forms.According to the authors, results were neither normalized nor subjected to statistical analysis, for they were considered preliminary.Nevertheless, they confirm the cliticization process.
The phonological, morphological and syntactic tests proposed by Zwicky (1985) have been used in order to establish whether cê is a clitic or a full word (Barbosa 2005).The accentual groups observed in introspective data and transcribed interviews indicate that cê carries primary accent and is therefore a full word.Other tests, as morphological, syntactic and prosodic ones, indicate that cê does not show the behavior of a clitic; therefore, it should be considered as a tonic syllable.
From within a sociolinguistic perspective, Peres (2 006), based on corpus data, analyzes the usage of cê/ocê/você both synchronically and diachronically.The corpora that the author use d for comparisons consisted of transcri bed interviews and w ere processed with the Goldvarb 2001 1 program (Robinson, Lawrence & Tagliamonte 2001) in order to det ermine relevant linguistic and extra-linguistic elements.The author finds that the form cê (1) shows an increase in us age over time and (2) occ urs in cont exts in which i t cannot be considered as a cli tic form, such as i n post-verbal position and following prepositions.

Methodology
In order to study the actual usage of the pronoun at issue, we extracted data from a corpus of Brazilian Portuguese spontaneous speech, the C-ORAL-BRASIL (Raso & Mello 2012).This corpus was chosen for three reasons: it is made up of 139 interactions recorded in natural context, showing great diaphasic variation (monologues, dialogues and conversations in a wide variety of situations, for a total of 208.130 words), which ensures, as much as possible, a faithful representation of the universe that the corpus intends to represent (Cresti & Moneglia 2005: 11); -the transcriptions of th e recordings are segmented into tone units and utterances, following the theoretical framework of th e Language into Act Theory (henceforth, L-AcT; Cresti 1995Cresti , 2000)), which com prises the empirical, inductive theory at the base of the C-ORAL-ROM com parable spoken corpora project (Cresti & Moneglia 2005); -its high acoustic quality and the t ext-to-speech alignment through WinPitch (Martin 2004), a computer program for acoustic analysis and measurements that are crucial for our study.
The first step was to retrieve all tokens of the forms cê/ocê/você from the corpus.Transcriptions were used in or der to locate the forms, but a veri fication of each occurrence by l istening to the audio files and observing the spectrum through the Praat software2 was made as well.This p hase was i mportant for screening the utterances that were good candidates for acoustic analysis, since we had to avoid audio portions with overlapping speech and/or too much background noise, factors that could interfere with a correct measurement of the durations.
The selected utterances containing the forms cê/ocê/você were acoustically analyzed in Praat.The utterances were segmented into phonetic syllables in order to allow for their durations to be properly obtained.The durations of the syllables of the form at issue was then compared to the durations of the other phonetic syllables in the same utterance and in the same tone unit.
According to L-AcT, the reference unit for spoken language is the utterance, defined as the smallest stretch of speech showing prosodic and pragmatic autonomy, i.e. performing a speech act.Utterance boundaries are signaled by terminal prosodic breaks, while boundaries within utterances are si gnaled by nonterminal prosodic breaks.In principle, each tone unit corresponds to an information unit.Thus, information units are defined in terms of prosodic, pragmatic and distributional properties.According to L-AcT, the main domain of syntax is the information unit.In fact, different information functions -realized by different tone units -can be found in the same utterance, which can interfere with the prosodic analysis (Cresti 2000(Cresti , 2014;;Raso 2012b;Moneglia & Raso 2014).
To avoid the effects of in trinsic segmental durations, all d urations were normalized in Praat using the SGDetector script (Barbosa 2006).In order to do so, the syllabic segmentation was done in broad transcription in ASCII, which is compatible with the SGDetector.
The normalized durations of the syllable [se] in the forms cê/ocê/você were then compared to the normalized durations of the other phonetic syllables within the same tone unit.
To verify the reliability of the research, results were subjected to statistical analysis with the Wil cox-Mann-Whitney test, in order to validate the difference among the syllable [se] of cê/ocê/você in preverbal and post-verbal positions and to verify whether they were statistically different or not.

Results
In the first search conducted on the corpus, a total of 2.026 tokens of cê, 262 tokens of ocê and 300 of você were collected.Then, they were organized in a spreadsheet in accordance to their occurrence as either preverbal or post-verbal elements, which is an indicator of their probable functions.
As previously mentioned, L-AcT assumes that pra gmatic autonomy is signaled by prosody, so that even elements lacking a full syntactic/semantic structure can be a utonomously realized as s peech acts.This is an im portant point, given that syntactic criteria fail to serve as a rel iable strategy for an adequate identification of reference units for s peech.The pragmatic/prosodic criteria adopted within L-AcT for the identification of reference units, on the other hand, based on the perceptual detection of terminal and non-terminal prosodic breaks, has proven to be much more reliable, as a number of studies have pointed out.In order to justify our methodological choices, we will discuss in more detail some aspects of the L-AcT framework.
Speech flow is segmented into tone units.The tone unit that carries the illocution is, at the informational level, called Comment (COM), the only information unit that is necessary and sufficient for the performance of an utterance, since it is the COM that supplies the utterance with pragmatic autonomy.A sim-ple utterance is constituted by the COM alone, while a compound one is constituted by the COM and one or more other information units.These other units also carry informational functions, and they are divided into textual and dialogic units.Textual units either build the text of the utterance, providing its semantic content, or refer to the text of the utterance, aiding in its interpretation 3 .Dialogic units, on the other hand, regulate the interaction 4 .Each unit of the information pattern is characterized by specific prosodic features.
The evaluation of informational functions compels us to treat differently the cases in which the pronoun itself (cê/ocê/você) forms part of a topic or scanned unit, or is pragmatically motivated, thus presenting specific prosodic features that do not allow us t o compare its dur ation with something (the tone unit) to which the pronoun does not belong (since it makes up the entire locutive content of the tone unit).
Regarding the occurrences of the form cê, 1.974 of them are preverbal and 38 post-verbal.Of the post-verbal group, 36 are preceded by a preposition and 1 makes up the entire locutive content of the hosting utterance, thus carrying illocutionary value.Considering now the t otal number of occurrences of cê, we found 13 of them making up the entire locutive content of the hosting tone unit, of which 9 are preverbal forms in topic units -with or without a scanning unit (i.e. the subject is se parately realized in a to ne unit that has no informational value and is marked by a non-terminal prosodic break) -, and 4 are post-verbal and preceded by a preposition.
Regarding ocê, 177 the occurrences are preverbal and 75 post-verbal, 51 of which are prepositional.We also found 8 cases of ocê isolated within a tone unit, 2 of which preceded by a preposition.
Regarding você, 225 of the occurrences are preverbal and 62 post-verbal, 46 of which being preceded by a prep osition.We foun d 6 cases in w hich você is isolated within an utterance -i.e.carrying illocutionary value -and 6 cases i n which it is isolated within a non-illocutionary tone unit, 1 of which preceded by a preposition.
Table 1 provides a summary of the results presented above.
For more detail, see Cresti (1995Cresti ( , 2000) ) and Raso (2012bRaso ( , 2014)).The examples below, extracted from the C-ORAL-BRASIL corpus5 , show the positions observed for the different forms: ( In (1) the forms ocê and cê are preverbal, in (2) there is a case of post-verbal cê without preposition.In (3) a prepositional post-verbal você is shown.Finally, (4) and ( 5) show, respectively, one example of ocê isolated in a tone unit and an example of você isolated within an utterance.

Acoustic analysis
We selected 133 utterances containing the forms cê/ocê/você from 62 different texts: 52 from familiar/private contexts (22 conversations, 17 dialogues and 12 monologues) and 9 from public contexts (3 con versations, 5 dialogues and 1 monologue).These forms are u ttered by 74 different speakers, 36 men and 38 women representing all sociolinguistic groups in the corpus.The utterances were selected based on the acoustic quality of the sound files and also the communicative situation, as it was neces sary to ens ure diaphasic variation within the sample.
We analyzed 16 utterances separately, as the pronouns carried specific illocutionary or pragmatic functions.In th ese cases, th e duration of th e pronouns could not be compared to the rest of t he tone unit durations because, as already mentioned, they have properties of their own6 .
Table 2 below shows how the other cases are distributed.The acoustic analysis of the 117 form s above was done through Praat.Each utterance was segmented into phonetic syllables, as Figure 1 shows.Example (6) shows the utterance in Figure 1.The z-score standard deviation of [se] is lower than that of the tone unit.
We will now disc uss the ca ses of post-ve rbal ocê.We ana lyzed 14 e xamples, of which 11 showed a higher z-score than that of their tonal unit.Figure 5 and Example (10) illustrate these results.

1
According to Guy & Zilles (2 007: 105), GoldVarb 2001 is a version for Windows of the VarbBrul (Variable Rules Analysis) package.It is a "s et of m ultivariate analysis computer program specifically built for sociolinguistic variation data".The program was developed by Steve Harlow based on the previous Macintosh version GoldV arb 2.0 by Rand & Sankoff (1990).GoldVarb 2001 was developed at York University in a collaborative project between the Language and Linguistics a nd the Computer Science Depar tments (see Robinson, Lawrence & Tagliamonte 2001).

Figure 2 .
Figure 2. Example of phonetic syllable segmentation of preverbal cê Example (7) shows the utterance in Figure 1.

Figure 5 .
Figure 5. Example of phonetic syllable segmentation of post-verbal ocê

Figure 6 .
Figure 6.Example of phonetic syllable segmentation of preverbal você

Figure 7 .
Figure 7. Example of phonetic syllable segmentation of post-verbal você Table1.Occurrences of cê/ocê/você in the C-ORAL-BRASIL corpus according to position