1 INTRODUCTION
Discourse marker (DM) can be defined as a lexical item whose removal does not affect the semantic integrity of the sentence. Its absence only changes the pragmatic implications of the sentence. Any lexical item, when used as a DM, no longer belongs to the grammatical category originally assigned to it. From a pragmatic perspective, DMs are of particular interest to many researchers because they are independent of the propositional content of the utterance. However, from a prosodic point of view, DMs attract a lot of attention because they come with various intonational realizations and constitute a distinct prosodic entity, especially in sentence-initial position. Therefore, determining how to circumscribe and differentiate various pragmatic functions of DMs by using prosodic cues becomes critical.
Previous research has demonstrated that native speakers use different prosodic cues, such as pause duration, f0 contour, f0 reset, intensity, etc., to classify DMs (see, for example, Komar, Reference Komar2007; Kleinhans et al., Reference Kleinhans, Farrús, Gravano, Pérez, Lai and Wanner2017; Didirková et al., Reference Didirková, Christodoulides and Simon2018, etc.). These prosodic features, together with their variations in pragmatic functions, make them particularly difficult for non-native speakers to master. Even though researchers in second language acquisition (SLA) have found that non-native speakers, like native speakers, use prosodic cues to realize specific pragmatic functions of certain DM(s) (see, for example, Romero-Trillo et al., Reference Romero-Trillo and Romero-Trillo2012), other studies have found an absence of pragmatic differentiation among L2 speakers. This lack of variation “annuls any possibility of functional specificity and results in the duplication of pragmatic meanings and, as a consequence, the absence of the various pragmatic nuances conveyed by the intonation of native speakers” (Romero-Trillo, Reference Romero-Trillo and Newell2014, p. 220).
Despite the fruitful work done on delimiting the prosodic features used by both native and non-native speakers, much of the existing literature has focused on a specific pragmatic function of one DM or several DMs appearing in the same position in the utterance (either the sentence-initial or the sentence-final position). Less research has been done on how speakers use prosodic cues to distinguish between different pragmatic functions of the same DM appearing in different positions in an utterance. In addition, many of the existing studies conducted on non-native speakers have been restricted to learners of English, and not as much data is available on learners of other languages, such as French.
To our knowledge, no literature has examined the case of the acoustic cues used by Mandarin Chinese L1 speakers learning French DMs. As Chinese is a tonal language, tone is mainly used to make a semantic distinction at the lexical level. In contrast, French is not a tonal language. The tonal pattern in French is used at the sentence level to indicate different types of sentences. It is not used to make lexical semantic distinctions. Difference in how tonal pattern is used between the two languages suggests a potential challenge for Chinese L1 speakers learning to express different pragmatic functions using different prosodic cues. We still do not know whether non-native speakers whose L1 is a tonal language can successfully learn to use prosodic cues to differentiate among different pragmatic functions of a particular DM and/or which prosodic cues they would use. Therefore, in this article, by conducting a corpus-based study on the French DM donc, we attempt to answer the following questions: Do Mandarin Chinese L1 speakers learning French also use the pitch cue to differentiate the pragmatic functions of DM donc? Do they use the same prosodic pattern as native speakers of French to make the distinction? Is there any L1 influence on their use of prosodic patterns? Is there any observable gender difference as documented in previous studies on tonal languages (see, for example, Zhang et al. Reference Zhang, Gong, Sheng, Sun, Bellamy and Ji2022)?
The structure of this article is as follows: first, the relevant literature on discursive functions of donc in French native speech, as well as previous work on the intersection of intonation and pragmatic functions, are reviewed. Second, the methodology, including information on corpora, speakers, tokens, acoustic measurements, and statistical analysis, are presented. Third, the results of the current study are presented and discussed in detail. Finally, the article concludes with a summary of the current work as well as future implications.
2 LITERATURE REVIEW
2.1 Pragmatic functions of donc in native speech
Donc is undoubtedly one of the most studied DMs in French. Researchers tried to classify the pragmatic functions of donc in French native speech over the past several decades. However, depending on the approach taken, the classification of the pragmatic functions of donc diverges from researcher to researcher.
Zénone (Reference Zénone1981) identified five types of donc: recovery donc, discursive donc, argumentative donc, metadiscursive donc, and recapitulative donc.
Recovery donc indicates the resumption of a theme previously addressed during the conversation and then abandoned or interrupted by a digression. For example:
For this type of donc, two criteria can be used for its identification: thematic criteria and substitution criteria. That is to say, on the one hand, donc can reintroduce something either explicitly or implicitly mentioned in the conversation. On the other hand, donc in this usage can be substituted by another marker fulfilling the same function, such as alors.
Compared to the definition of recovery donc, Zénone’s definition of discursive donc seems to be vague. She pointed out that discursive donc can often be found in the structure of X donc p, where X indicates that the antecedent is not constituted by the linguistic content and “p” stands for any syntactic structure that comes after donc. For example,
The third type of donc is called argumentative donc in Zénone’s classification. According to her, the argumentative donc relates an antecedent immediately present in the context with a consequence p in a structure of q donc p, for example:
She argued that the reason why this type of donc is called “argumentative” donc is because it presents, on the one hand, p as the consequence and the conclusion of what it follows; on the other hand, q as the motivation or proof of the validity of what follows it.
The fourth type of donc is metadiscursive donc. In a structure of q donc p, p is a qualification and/or a definition of q. For example:
Lastly, there exists also a recapitulative donc in Zénone’s classification. She pointed out that this is an infrequent usage of donc, especially in oral speech. In this usage, donc merely repeats the conclusion of the previous paragraph without bringing any new information. For example:
Nevertheless, in her later work (Zénone, Reference Zénone1982), she collapsed these five functions into two: co-textual (where donc relates two propositional phrases in the same sentence) and contextual (where donc relates two propositional phrases in different sentences). However, the problem with this classification is that these five functions are not mutually distinct, and the later binary analysis does not seem necessary.
In line with Sperber and Wilson’s cognitive approach, Hansen (Reference Hansen1997) proposed analyzing the discursive functions of donc under the notion of “mutual manifestness” and identified three discursive functions of donc: marking a conclusion or a result, marking a repetition of the earlier statement, and the discursive usage. For example:
However, she admitted that the first and the second usages are not mutually exclusive. In the third usage, donc is more of a modal particle than a discourse marker, as the particle tends to be pronounced without the final /k/.
From a pragmatic point of view, Vlemings (Reference Vlemings2003) tried to distinguish between the conjunctive inferential usage and the adverbial discursive usage of donc. The inferential donc, also called the argumentative donc, marks an inferential connection between the host utterance and the antecedent (Ferrari and Rossari, Reference Ferrari and Rossari1994). According to Vlemings, the inferential usage is the core meaning of donc. It makes explicit a connection to an extralinguistic situation. Pragmatically speaking, the discursive usage contextually derives from its inferential usage. Therefore, the inferential usage of donc could also account for its discursive usage. The distinction between the two is no longer helpful.
Bolly and Degand (Reference Bolly and Degand2009), based on corpus data, investigated the multi-functional status of donc in French. They distinguished four functions of donc: consequential, recapitulating, reformulating, and discursive. They argued that the consequential function is the primary function of donc, while the other three are discursive functions. They also pointed out that the functionally evolutionary transition from consequential function to other discursive functions, “parallels a loss of the specific semantico-syntactic features of consequential donc” (Bolly and Degand, Reference Bolly and Degand2009: 1).
Dostie (Reference Dostie2014) distinguished six functions of donc: to introduce a consequence, to indicate a refocusing toward the principal theme, to present an earlier stated comment, to solicit following information, to… (suspensive), and to indicate a continuum in a narration.
In this usage, donc relates two propositions that are in a cause-consequence relation. This is the function that historically develops from the temporal value of donc. Therefore, this type of donc not only indicates the causal connection between the two propositions that it relates but also the two events are given at the same time. This seems to overlap with the “causal link” function in Zénone’s classification. For example:
Donc in this usage refocuses the discourse on the theme on which it must continue and that the digressions have made us lose sight of it. For example:
Donc in this usage is similar to a reformulative marker, in that it allows the speaker to present, in different words, what has just been said. For example:
When used as a word-phrase and produced with a slightly rising intonation, donc functions as a request for the interlocutor to follow up on what is being said and come to a conclusion.
The suspensive donc announces a potential following, but is eventually left unfinished. For example:
In such a situation, donc does not bridge the propositional content of the statements that precede and follow it. Rather, it serves to make the text fluid, in the sense that it allows the speaker to link the different sequences of the text being developed narratively. For example,
This classification seems less confusing than the others in that it does not try to define the discursive usage and distinguish it from other usages. However, as seen in other approaches, very often, the “discursive usage” is not clearly defined and not really distinguished from other functions. Some believe that it evolves from the original consequential functional, as proposed by Bolly and Degand (Reference Bolly and Degand2009). Nevertheless, they did not tell how a discursive function should be different from recapitulating or reformulating functions since all three evolve from the original consequential function and can all be defined broadly as discursive function. This is also the problem with Zénone’s and Hansen’s classification. Since in all three approaches, the discursive function cannot clearly be defined and distinguished from other functions other than the original consequential function, the overlap between different functions is unavoidable. Meanwhile, the problem with Vleming’s classification is that it includes too few functions. It only distinguishes the discursive usage and the non-discursive usage. Therefore, in the current work, we will adopt Dostie’s classification for coding pragmatic functions. On the one hand, based on the relation between preceding and the following propositions, this classification allows us to better identify instances of each function without too much confusion or difficulty in classifying. On the other hand, since her classification is also based on oral corpus data with a similar structure as ours, her classification should be more applicable for comparison between native and non-native speech of French in the current study.
2.2 Donc in non-native speech
Previous research indicated that non-native speakers could acquire some pragmatic functions of donc used as a DM. However, the frequency of use by non-native speakers might differ from that of native speakers. In addition, some pragmatic functions might be missing from the non-native speech.
By contrasting classroom instruction and immersion, Barnes (1998) investigated the usage of several DMs in narratives, including donc, by 27 adult American learners of French. He noted that in non-native speech, donc could be used to “mark a relation of consequence” (1998, p. 193) and “mark a shift from one level of the discourse to another, for example, from descriptive background or commentary to the main storyline” (1998, p. 193). Pellet referred to the first usage described by Barnes as “referential function” (2006, p. 158), which corresponds to the first function proposed by Dostie (Reference Dostie2014), and the second as “structural function” (2006, p. 158), which corresponds to the third function proposed in the work of Dostie (Reference Dostie2014). Out of the six functions proposed by Dostie in native speech, only two were found in the narratives of non-native speech in Barnes’s study.
Rehner (2002) analyzed the usage of several DMs in the speech of 41 French immersion students in Canada. She distinguished between discursive and non-discursive functions for donc. The discursive functions she identified are organizational/transitional function, clarification/expansion, turn-yield, emphasizer, and punctor. The non-discursive function includes consequence marking. This classification corresponds more or less to Dostie’s classification. However, she did notice that overall, donc was not used frequently by the students. Her findings suggested that compared to classroom instruction, immersion seems to be more effective for students in acquiring the various pragmatic functions of DMs. While the difference is significant, however, the frequency of usage still does not reach the native level.
Hancock (Reference Hancock2004) examined the usage of donc by three Swedish speakers learning French. As for the position of the marker, Hancock (Reference Hancock2004, p. 104) pointed out that donc can be used as a ligator to introduce either the first pre-rheme or a pre-rheme within the idea unit (“Rheme” can be defined as the constituent of a sentence that adds most new information, in addition to what has already been said in the discourse). It can also appear within an idea unit and co-occur with the pre-rheme or within the rheme (Hancock, Reference Hancock2004, p. 104). Meanwhile, donc can also be used as a punctor at the end of the idea unit (Hancock, Reference Hancock2004, p. 105). She found that the last usage was not found in her native data. As for the pragmatic functions of donc, she proposed four categories: to mark a conclusion or a consequence; to index equivalences; to indicate a resumption; and to signal an implicit conclusion. Her results indicated that while some usage is unique to non-native speakers, some functions found in native speech are also missing from the non-native speech.
Deng (Reference Deng2020) then examined 1,245 occurrences of donc in the speech of 40 Chinese speakers learning French. She argued that donc is one of the most frequent DMs used by Chinese speakers learning French. As for the pragmatic functions, she proposed distinguishing only two categories of donc: the argumentative donc, which is in its original usage, and the discursive adverbial donc, which has a broader meaning. Her results showed that in non-native speech, donc is used as an exemplification marker “to introduce some examples to the prior statement” (Deng, Reference Deng2020, p. 7), which is absent from native speech. She suggested that this might be due to the impact of the L1 transfer, since in Chinese, the equivalent of donc could be used as a marker of exemplification.
As shown by these four studies, we see that non-native speakers can acquire the pragmatic functions of donc. However, non-native speakers from different L1 backgrounds demonstrated different patterns of usage both from each other (at the group level) and from native speakers. While the usage of donc is infrequent for some groups of non-native speakers, it is the most frequently used DM by other groups of non-native speakers. This gives us reason to believe that there is some influence of L1 on the acquisition of DMs in non-native speech. However, the question is whether this influence exists only at the pragmatic level.
2.3 Discourse marker and prosody in native speech
Previous studies have confirmed that there is a correlation between prosodic patterns and pragmatic functions. For example, Carter and McCarthy (Reference Carter and McCarthy2006, p. 539) pointed out that prosodic information can distinguish between DMs and other parts of speech since DMs tend to have their own intonational phrase. By examining how the pragmatic function and the distribution of pragmatic markers in oral speech are reflected by prosodic cues in BBC TV series, Komar (Reference Komar2007) demonstrated that the intonation pattern largely depends on the discourse function of a marker in English. His findings confirmed that the prosodic patterns could be used to distinguish the DMs which are more involved in discourse organization and those whose function is more pragmatic. This observation is particularly relevant to our research in that since donc fulfills both a discourse organizational function (for example, when used to indicate the continuum in narration) and a more pragmatic function (for example, when used to solicit following information or as a suspensive marker), speakers would use different prosodic patterns to differentiate between different functions of DM donc.
Kleinhans et al. (Reference Kleinhans, Farrús, Gravano, Pérez, Lai and Wanner2017) investigated the relationship between the discourse structure of a spoken monologue and its prosody in English. Their results showed that prosodic features such as intensity, f0, and/or speech rate are key features with the highest predictability for discourse relations in intrasegmental features (which occur within an elementary discourse unit) and intersegmental features (which capture differences between two elementary discourse units). This means that it is possible to use these cues to distinguish between DMs and non-DMs and among different pragmatic functions.
This point was also noted by other researchers when studying the correlation between prosodic patterns and DMs in different languages. For example, Hasson (1999) reported on the functions and prosodic correlates of the Swedish DM men “but”/ “and” and så “so” in dialogue. She showed that the presence of a previous pause, a preceding or following DM, and the location of or within the turn (initial/non-initial) are factors that are relevant when combined with the prosodic parameters in distinguishing different functions of DMs in dialogue. Van Zyl and Hanekom (Reference Van Zyl and Hanekom2013) found prosodic variation between the neutral and reluctant agreement for the English affirmative cue word okay. Based on large corpora data in European Portuguese, Cabarrão et al. (Reference Cabarrão, Moniz, Ferreira, Batista, Trancoso, Mata and Curto2015) demonstrated that prosodic features can be used for automatic classification of discourse markers and non-discourse markers. Volín et al. (Reference Volín, Weingartova and Niebuhr2016) also reported that, for the Czech DM jasně “right”, different prosodic patterns align with different pragmatic functions. Prieto and Roseano (Reference Prieto and Roseano2021) found that while Catalan used specific intonation patterns to express low and intensified commitment statements, Friulian speakers used only one type of pitch contour to express both types of epistemic commitment. Shan (Reference Shan2021) discussed the prosodic and pragmatic features of the DM ni zhidao “you know” in spoken Chinese. Four functions were explored in his study: initiating a topic when occurring sentence-initially, holding the floor when appearing within clauses, marking coherence when making its presence between clauses, and projecting attitudes and feelings when showing up sentence-finally. The results showed that its prosodic features, including duration, tempo, pre-pause, post-pause, f0, and intensity, significantly relate to its pragmatic functions and thus can be used as indicators for pragmatic functions.
As for French DMs, Didirková et al. (Reference Didirková, Christodoulides and Simon2018) conducted an experimental study on the prosodic features used to distinguish the DMs alors and et and the non-DMs alors and et in French. By analyzing data from 20 adult native speakers of French, they found that the silent pause duration before the DM and the absolute duration of the DM are used by the speakers to differentiate between the DM use and the non-DM use of alors and et in French. Christodoulides (2018) then conducted a complementary study on alors and et by looking at the silent pause before alors and et, articulation rate, pitch reset, prosodic prominence as well as prosodic boundaries. The results showed that the usage of et as a DM was more prevalent in non-planned speech; silent pauses preceded occurrences of alors and et, both as DMs and as non-DMs; the difference in silent pause duration in DM usages vs. non-DM usages was not statistically significant for alors but was statistically significant for et; DMs did not systematically constitute a separate prosodic unit; a strong prosodic boundary differentiates between the usage of et as a DM or as a coordinating conjunction between verb phrases and subordinate clauses, and its other non-DM usages. However, both studies only looked at how to distinguish between the non-DM usage and the DM usage of the markers. Neither examined how speakers use these prosodic features to distinguish the various pragmatic functions of the same marker. Lee et al. (Reference Lee, Jouvet, Bartkova, Keromnes and Dargnat2020), by investigating corpus data in both English and French, compared the prosodic characteristics of French DMs (alors, bon and donc) and English DMs (now, so and well). The results revealed that pragmatic roles of DMs cause specific prosodic behavior in terms of the existence and position of pauses, as well as their f0 articulation in their immediate context. Furthermore, comparable pragmatic roles usually have similar prosodic traits, even across languages.
2.4 Discourse markers and prosody in non-native speech
Researchers in SLA also investigated how non-native speakers use different prosodic features to realize the pragmatic functions of DMs. For example, Romero-Trillo and Newell (Reference Romero-Trillo and Romero-Trillo2012) examined the usage of pragmatic markers mhm and yeah that function as feedback elements in the speech of 50 Spanish speakers learning English. On the one hand, they found that both native and non-native speakers use a combination of the duration of the marker and the pitch to realize the pragmatic functions of the marker. However, the non-native pattern differs from the native pattern in that native speakers tend to have a higher final pitch than non-native speakers. On the other hand, the duration of both markers produced by native speakers is longer than that by non-native speakers. This finding confirmed that even though non-native speakers can use prosodic cues to realize certain pragmatic functions, their pattern differs from the native one. De Marco (Reference De Marco2016) compared the prosodic cues of three DMs, però ‘but’, allora ‘then’, quindi ‘therefore’, used by five Italian L2 speakers from diverse L1 backgrounds. She found that when using DMs, non-native speakers tend to have a slower speech rate and shorter pauses than native speakers. However, at the same time, she argued that this might also be due to the difficulty in managing one’s speech as a non-native speaker.
Similar to studies on the interaction between prosody and DMs in native speech, existing research on non-native speech has focused on the distinction between the use or non-use of DMs, while usage of the same DM for diverse pragmatic functions is ignored. In addition, the non-native speakers investigated were mainly L1 speakers of European languages. We do not know how Asian speakers from a tonal language background would use prosodic features, especially pitch, to differentiate various pragmatic functions of the same DM since the tonal pattern in their L1 can be transferred to their L2. Furthermore, most of the existing studies are mainly based on a small sample. A corpus study based on sociolinguistic interviews could reveal a more authentic usage of the target language by non-native speakers in a natural setting. Therefore, we hope to partially fill the gap by comparing non-native data with native data from two comparable corpora in French.
3 METHODOLOGY
3.1 Corpora
3.1.1 Native corpus
The native speech corpus that we used for the current study is ESLO 2 (Enquêtes Sociolinguistiques à Orléans: http://eslo.huma-num.fr/index.php, Baude and Dugua, Reference Baude and Dugua2011), which is an online, publicly accessible corpus, consisting of sociolinguistic interviews with native speakers of French in Orléans, France. ESLO 2 was initiated in 2008 and is currently under construction. The corpus consists of several modules, ranging from interviews to questionnaires. For the current study, we only used the interview section, including 78 interviews in total. The length of each interview varied from 1 to 2 hours, depending on the speaker. For our final analysis, we randomly selected 20 interviews for comparison with non-native data in our corpus.
3.1.2 Non-native corpus
The datasets used in this study for non-native speakers were extracted from a private corpus consisting of semi-guided conversations in French carried out by one of the principal investigators with L2 French speakers in France between 2014 and 2017. The choice of this corpus is justified by the spontaneous nature of the conversations, which allows for the maximum naturalness of the production of DMs. For the current work, we randomly selected 20 interviews for comparison purposes. We made sure that in each interview, there are enough tokens of donc for further analysis.
The length of each interview varies from 30 minutes to 90 minutes. A list of questions, similar to those asked in ESLO, was prepared before the interview. These questions were used merely as a medium to mimic the natural conversation flow and were loosely based on a script including (1) questions designed to gather personal information from the interviewees as well as to familiarize them with the interview format; (2) questions on memorable events in life such as childhood memories, theft-related experiences, first days in Paris, situations where their lives were in danger, etc.; (3) a series of questions on their French learning and their opinions on the usage of French by native speakers around them. All interviews were recorded and transcribed entirely in Praat (Boersma & Weenink, Reference Boersma and Weenink2016). Since the speakers in our corpus are all non-native speakers of French, there might be some grammatical errors in their oral production. For the transcription, these errors were not corrected. All interviews were transcribed as produced.
3.2 Speakers
3.2.1 Native speakers
The 20 native speakers come from Orléans, France, and include 10 female and 10 male speakers. The age range of these 20 speakers is between 20 and 80 years old (M = 40.1 ± 16.9). Their socio-economic status also varies, with participants’ professions ranging from engineers to part-time workers. This diversity in demographic profiles allows our sample to be more representative. Table 1 presents the detailed information on each speaker in this study, including their gender, age, and profession at the time of the interview.
Table 1. Detailed demographic information of the 20 native speakers
3.2.2 Non-native speakers
The 20 non-native participants are all Chinese L1 and French L2 speakers. All were born in mainland China and only came to France after reaching puberty. All of them were living in Paris, France, at the time of the interview. These speakers are not L2 learners in a classroom setting, but rather the actual L2 users living in Paris. They were recruited through the interviewer and interviewees’ social networks, and their participation in the interviews was entirely voluntary.
These speakers, 10 women and 10 men, are between 23 and 51 years old (M = 28.9 ± 6.4). While some have lived in Paris for more than 10 years, others have recently arrived in Paris. Their length of residence in France ranges from 1 to 12 years (M = 4.7 ± 3.2). Accordingly, their linguistic competence also varies. They hold a variety of jobs, from student to engineer. As mentioned earlier, these non-native speakers did not come from a conventional language classroom setting. This diversity in professional status better represents the actual language users in the community. Table 2 presents the detailed information on each speaker in our corpus:
Table 2. Detailed demographic information on the 20 non-native speakers
3.3 Tokens
For coding and identification of tokens, first, all instances of donc produced by the native and non-native speakers in the two corpora used in the current study were identified by the two principal investigators and extracted on one tier in Praat for further analysis. For this step, all tokens produced by the interviewees during the entire interviews were included without any exclusions. Occurrences interfered by the interviewer’s backchanneling during the interview were manually excluded by listening to the audio files. In total, 2,120 occurrences of donc in native speech and 876 occurrences of donc in non-native speech were extracted. Second, the two principal investigators coded separately the pragmatic functions on a second annotation tier in Praat. Each token of donc was assigned a function label based on Dostie’s (Reference Dostie2014) classification: as donc 1: introduce a consequence; donc 2: indicate a refocusing toward the principal theme; donc 3: introduce a comment on what was stated earlier; donc 4: solicit the following information; donc 5: suspensive donc, and donc 6: indicate a continuum in a narration. Third, after the separate coding, the two investigators compared their coding for functions. In this step, the occurrences on which the two investigators disagreed on their function were excluded from the final analysis. Therefore, for the final analysis, in total, in native speech, we had 1,733 occurrences of donc; for non-native speech, we had 743 occurrences of donc.
Examples for each function in non-native speech in our corpus are provided below:
-
1) Introduce a consequence
bon après comme on est pas une ville moderne comme Shanghai Pékin donc les gens sont tous gentils (RZ_2016)
-
2) Indicate a refocusing toward the principal theme
Intervieweuse : croyez vous que savoir parler anglais est très important ici?
Interviewée : hmmm dans la vie quiditienne non dans des entreprises ça dépend si c’est des entreprises un peu internationales c’est quand même un atout de parler anglais puisuqe y a des/des filias un peu partout dans/dans le monde entier y a des contacts des collègues anglais je me souviens que quand j'étais en stage j’étais dans le banque c’était une société générale on a fait des fin des/des échanges avec les collègues en Grande Bretagne par écrit par email bon ça suffit ça va c’était ok mais euh et réunion par téléphone en anglais c’était dur moi je/je peux comprendre une moitié c’est XXX mais si il faut que je parle quelque chose c’était/c’était vraiment dur donc je pense que c’est un/c’est un atout de parler anglais (LK_2014)
-
3) Introduce a comment on what was stated earlier
C’est une ville très difficile à vivre…euh…en général alors c’est une ville pendant l’été hyper chaud/pendant l’hiver hyper froid donc très très difficile à vivre pour les… (HQ_2016)
-
4) Solicit the following information
Interviewée : je connais des Français mais ça veut dire pas amis amis vraiment amis donc…
Intervieweuse : mais vous sortez souvent avec eux (ZDZ_2016)
-
5) Suspensive donc
moi je suis en troisième cycle c’est le la fin d’études oui donc mais le diplôme va certifie non de valider (LYY_2014)
-
6) Indicate a continuum in a narration
euh en fait je fais ma licence français euh donc à Pékin à l’université euh c’est une information de quatre ans (XSS_2014)
The distribution of donc by pragmatic function in both native and non-native speech is provided in Table 3.
Table 3. General distribution of donc by pragmatic function in two corpora
3.4 Acoustic measurements
The word donc in the sound files was first identified and labeled in Praat by the two principal investigators with extensive phonetic training. The annotation boundaries of the word donc were adjusted according to its spectral features. Subsequently, the Praat script ProsodyPro (Xu, Reference Xu2013) was used to extract the time-normalized f0 values of the word donc. For better and more accurate extraction of f0 values, the vocal pulse markers during the voiced part of the word were scrutinized and added if missing. For each time interval, we initially extracted 50 measurement points with equal distance. We transformed the time-normalized f0 values into semitones using the formula st = 12 * log2(f0/reference f0) and only selected 10 time-normalized f0 values for statistical analyses (Zahner-Ritter et al., Reference Zahner-Ritter, Einfeldt, Wochner, James, Dehé and Braun2022). Based on our visual inspection, we set 90 Hz and 165 Hz as the reference f0 for male and female speakers, respectively, ensuring that most of the converted f0 values were positive in semitones (M = 2.73 ± 4.86) (Zahner-Ritter et al., Reference Zahner-Ritter, Einfeldt, Wochner, James, Dehé and Braun2022). To satisfy the statistical requirement of normal distribution, we only selected f0 values from −5 to 10 in semitones, and thus a total of 2,277 occurrences of donc were used for the statistical analyses.
3.5 Statistical analysis
In this study, we used generalized additive mixed models (GAMMs) to examine the differences in the pitch contour of the word donc across pragmatic functions (Wood, Reference Wood2011, Reference Wood2017). The advantage of using GAMM on time-normalized f0 values is that it allows us to model the non-linear relationship between f0 values and a set of predictors (i.e., pragmatic functions, language group, and gender) over time with pre-specified base functions.
The f0 values at 10 equal-distance time points were the response variable of the GAMM. To determine the best-fitting model, we fitted the models using the method of maximum likelihood estimation (Sóskuthy, Reference Sóskuthy2021; Wieling, Reference Wieling2018). For model comparisons, we started with a GAMM model without interactions between fixed factors, and then added interactions between fixed factors (Zahner-Ritter et al., Reference Zahner-Ritter, Einfeldt, Wochner, James, Dehé and Braun2022). Factors or interactions were included or excluded depending on their contribution to the model fit. To evaluate the model fit of each model, we adjusted the number of base functions. To reduce the effect of tailed residuals, we performed the best-fitting model with the scaled t distribution (van Rij et al., Reference Van Rij, Hendriks, van Rijn, Baayen and Wood2019).
To evaluate the differences in pitch contours of donc 1-6 between native and non-native speakers, we split the data by donc type and ran the same GAMM with the ordered language group factor for each partitioned dataset (Chuang et al., Reference Chuang, Fon, Papakyritsis and Baayen2021). To further evaluate the within-group differences for each language group, we additionally separated the partitioned datasets by language group and repeatedly performed the same GAMM with the ordered gender factor (Chuang et al., Reference Chuang, Fon, Papakyritsis and Baayen2021). For non-native speakers, the duration of stay was also included in the model as one of the random effects. The GAMM formulas are available in the Appendix.
4 RESULTS
4.1 Native vs. non-native speakers
Figure 1 shows the averaged f0 contours of functions 1-6 of the word donc produced by different groups of speakers (i.e., native vs. non-native speakers in Figure 1a; male vs. female native speakers in Figure 1b; male vs. female non-native speakers in Figure 1c). As shown in Figure 1a, the pitch contours of donc 1, 3, 5, and 6 were approximately parallel between the two groups. These pitch contours were falling, and the differences between the two groups were no more than two semitones. Donc 2 and donc 4 demonstrated more differences between the two groups than other uses of donc. As for donc 2, there was a difference in the general pitch contour between the two groups, with native speakers having a relatively rising pitch contour but non-native speakers maintaining a falling pitch contour. Regarding donc 4, although the two groups shared a relatively falling pitch contour, the non-native speakers had a small rise after the onset, and the difference in the pitch contour between the two groups was over three semitones.
Figure 1. Averaged f0 contours (in st) of the word donc 1-6 produced by native and non-native speakers (a: upper panel), by native speakers only (b: middle panel), and by non-native speakers only (c: lower panel). The upper dashed line represents the sum of mean and standard error at each time-normalized point. The lower dashed line represents the difference between mean and standard error at each time-normalized point.
Figures 1b and 1c show the pitch contours of donc 1-6 among male and female speakers within the native and non-native speaker groups, respectively. As shown in Figure 1b, small differences were observed regarding the pitch contours of donc 1-6 between male and female native speakers. Specifically, while pitch contours of donc 2, donc 3, and donc 6 are similar across both genders, male native speakers tended to have a more prominent rising offset than female native speakers for donc 5. For donc 1, although the second half of the pitch contours were almost overlapping, the first half of the pitch contour was high-falling for male native speakers but relatively flat for female native speakers. More differences can be found in donc 4 between male and female native speakers. Specifically, the male native speakers had a steeper falling onset and slight rise until the middle of the word and then fell in a small range, followed by a rising offset, while the female native speakers exhibited a less steep fall from the onset to the middle of the word and then rose gently with a falling offset.
The pitch contours of donc 1-6 of non-native speakers presented more interesting results in Figure 1c. Like the native speakers, the general patterns of the pitch contours of donc 1-6 were similar between male and female non-native speakers. In detail, non-native speakers had almost overlapping pitch contours of donc 5 and 6 irrespective of gender. Regarding donc 1, donc 2, and donc 3, although there was a noticeable difference in pitch between male and female non-native speakers, the pitch difference was less than two semitones. The pitch contour of donc 4 produced by female non-native speakers demonstrated more individual differences than the contour for male non-native speakers. One finding inviting discussion is that male non-native speakers tended to have a higher pitch than female non-native speakers.
Table 4 summarizes the parametric coefficients (part A) and smooth functions (part B) of the generalized additive mixed-effects model. Each predictor has one reference level that is not shown in Table 4 (i.e., donc 1, female, and native). The results of the parametric coefficients indicate the effect of the predictors in the model, which is similar to those in the generalized linear model, while the results of the smooth functions evaluate the significance of the smooth terms. The explained variance of the model was about 66%. When it comes to the word donc with different pragmatic functions, the results indicated that donc 2 had a significantly higher pitch than donc 1, and donc 5 had a significantly lower pitch than donc 1. In addition, the results also suggested that the pitch of male speakers was significantly higher than that of female speakers, and the pitch of non-native speakers was significantly higher than that of native speakers. As shown in Table 4, all the smooth functions were significant. Specifically, all non-linear trajectories associated with different uses of donc were significant. In addition, by-token random intercepts and by-speaker smooth for the normalized time were also found to be significant. To better understand the interaction terms in the model, the data was first partitioned by donc type and then by language group. A set of GAMMs was performed with the partitioned data using ordered factors.
Table 4. Summary of the generalized additive mixed-effects model fitted to the time-normalized f0 values of the word donc (*: p < 0.05; **: p < 0.01; ***: p < 0.001)
Figure 2 presents the predicted f0 values of donc 1-6 generated by the model. As shown in Figure 2, all the predicted pitch contours of donc 1-6 based on the predicted f0 values were falling from the onset to just before the offset, but with slight differences in the falling slopes. A rising offset was predicted for all types of donc. Except for donc 4 and 5, the predicted pitch contours of donc 1, 2, 3, and 6 were similar to those in Figure 1 such that they were falling in general. Regarding the pitch contour of donc 4, native speakers had a slight rise in the middle part of the word (see Figure 1b), while non-native male speakers had a rising onset (see Figure 1c), which were different from the falling pitch contour predicted by the model. In addition, regarding the pitch contour of donc 5 produced by non-native speakers (see Figure 1c), its falling slope was smaller than that of the predicted pitch contour.
Figure 2. Predicted f0 values of the word donc with different pragmatic functions generated by GAMM.
Figure 3 presents the predicted difference curves between native and non-native groups and the predicted pitch values of the two groups for the word donc 1-6. As shown in the plots of difference curves in Figure 3, we can conclude that there were significant differences in the predicted f0 values between the native and non-native groups in terms of donc 1, donc 3, donc 4, and donc 5 (see Figure 3a, 3e, 3g, and 3i). For donc 2, the two groups still differed significantly in the first half of the predicted pitch contour, while the significant difference was not maintained for the second half of the predicted pitch contour (see Figure 3b). No significant difference was found between the two groups regarding donc 6 (see Figure 3k). The predicted pitch contours can be summarized as follows: the predicted pitch contour of donc remained falling with a slight rise at the offset, except for donc 2 of native speakers (which slightly rises, see Figure 3d), regardless of pragmatic functions and language background. The differences in predicted pitch between native and non-native speakers were relatively small, around one or two semitones, and the predicted pitch of native speakers was generally lower than that of non-native speakers.
Figure 3. Predicted difference curves between native and non-native groups (left panel) and predicted pitch values of the two groups (right panel) for the word donc 1-6. For the predicted difference curves, the green shaded region represents the 95% confidence interval of the predicted mean differences, and the difference is significant when the 95% confidence interval region does not include zero. For the plots in the right panel, the red line represents the predicted pitch values of the non-native group, and the blue line represents the predicted values of the native group.
4.2 Native speakers only
Figure 4 presents the predicted difference curves between male and female speakers in the native group and the predicted pitch values for native speakers for donc 1-6. Based on the plots of the difference curves in Figure 4 (left panel), it was noted that there were no significant differences between male and female native speakers for each usage of donc, suggesting an absence of gender effect on the production of donc with different pragmatic functions. Notwithstanding, there were discernible variations in the predicted pitch values for male and female speakers, as illustrated in the panel of Figure 4, in a descriptive sense. Except for donc 4 (falling-rising, see Figure 4h) and 5 (falling, see Figure 4j), the predicted pitch contours of other donc of female native speakers were relatively flat. In contrast, male native speakers exhibited a different pattern of the predicted pitch contours of the donc such that all usages of donc had a falling pitch contour, except for donc 2 (rising, see Figure 4d)
Figure 4. Predicted difference curves between male and female speakers in the native group (left panel) and predicted pitch values for native speakers (right panel) for donc 1-6. For the predicted difference curves, the green shaded region represents the 95% confidence interval of the predicted mean differences, and the difference is significant when the 95% confidence interval region does not include zero. For the plots in the right panel, the red line represents the predicted pitch values of male speakers, and the blue line represents the predicted values of female speakers.
4.3 Non-native speakers only
Figure 5 presents the predicted difference curves between male and female speakers in the non-native group and the predicted pitch values for native speakers for the word donc 1-6. Based on the plots of the difference curves in Figure 5 (left panel), it was found that there were no significant differences between male and female non-native speakers for donc 1, 4, 5, and 6 (see Figure 5a, 5g, 5i, and 5k). One notable aspect of donc 2 was the significant differences in pitch contours between male and female non-native speakers (see Figure 5c), and male non-native speakers had a significantly higher predicted pitch than female non-native speakers. Regarding donc 3, there were significant differences in the first half of the predicted pitch contours, with a higher predicted pitch for male non-native speakers, but not in the second half (see Figure 5e).
Figure 5. Predicted difference curves between male and female speakers in the non-native group (left panel) and predicted pitch values for non-native speakers (right panel) for the word donc 1-6. For the predicted difference curves, the green shaded region represents the 95% confidence interval of the predicted mean differences, and the difference is significant when the 95% confidence interval region does not include zero. For the plots in the right panel, the red line represents the predicted pitch values of male speakers, and the blue line represents the predicted values of female speakers.
5 DISCUSSION
5.1 Using the pitch cue to differentiate the pragmatic functions
Going back to the questions we raised at the beginning of this article: Do Mandarin Chinese L1 speakers learning French use the pitch cue to differentiate the pragmatic functions of DM donc? Do they use the same prosodic patterns as native speakers of French to make the distinction? As shown by our results, the pitch seems to be an important feature in differentiating various functions of donc in non-native speech. Pitch contours in non-native speech mostly match the native pattern, especially when donc is used in function 1: introduce a consequence, function 3: introduce an earlier stated comment, function 5: suspensive donc, and function 6: indicate a continuum in a narration. For these functions, the native pattern is falling. For function 2: to indicate a refocusing toward the principal theme, and function 4: to solicit the following, non-native speakers demonstrated a different pattern from native speakers. Function 2, which indicates a refocusing of the topic, tends to be associated with a rising pitch contour in native speech, while non-native speakers tend to realize it with a falling pitch contour. It seems that non-native speakers use falling pitch contours to realize all six functions, while native speakers tend to associate the rising and falling pitch contours with different functions. The rising pitch contour is completely missing from the non-native speech.
Based on these results, it seems that non-native speakers do not necessarily vary in pitch contour for every individual pragmatic function fulfilled by one DM in an utterance. In other words, pitch contour alone is not sufficient to distinguish all pragmatic functions of a DM. Native speakers must use other prosodic cues in combination with f0 to make the distinction between all functions. Nonetheless, the native speakers’ f0 pattern shows some variations associated with different pragmatic functions. On the other hand, even though non-native speakers rely on f0 as a prosodic cue to signal pragmatic functions, the variations of pitch contour patterns between various functions of the DM are much smaller in non-native speech than in native speech. Our results also confirmed previous findings on non-native speech in general: non-native speakers do not vary much in producing the prosodic cues of DMs (see, for example, Romero-Trillo, Reference Romero-Trillo and Newell2014). Therefore, part of the distinction made by native speakers associated with this variation is not retained in non-native speech. However, this seems to be problematic if non-native speakers are expected to be able to identify particular pragmatic functions of a marker by using f0 as the acoustic cue.
These findings could further our understanding of the acquisition of prosody in L2. As pointed out by Xu (Reference Xu, Behr, Yue and Myers2017), intonation in Chinese is mainly used for expressing focus, modality (statement/question), topic, and boundary marking. Shan (Reference Shan2021) also confirmed that f0 is an important prosodic feature that native speakers of Chinese use to distinguish between different pragmatic functions in Chinese native speech. That is to say, non-native speakers of French with Chinese as L1 should be able to use f0 to distinguish between these pragmatic functions in French. But why they do not behave as expected? One reasonable hypothesis is that maybe in their L1, suoyi, which is the Chinese equivalent of discourse marker donc, fulfills a unique pragmatic function in Chinese. However, Wang and Huang (Reference Wang and Huang2006) argued that suoyi can fulfill four different pragmatic functions in Chinese: 1) consequential connective, 2) paraphrasing use, 3) resumptive opener, and 4) topic-initiator. These four functions correspond to donc 1, donc 3, donc 4 and donc 5 according to Dostie’s classification, respectively. However, the donc 2 and donc 6 functions are missing from Chinese usage of suoyi. Even though the Chinese suoyi fulfills fewer pragmatic functions than French donc, f0 should still be in the disposition of Chinese speakers for function distinction.
A second reasonable explanation is that the syntactic coherence of oral discourse is acquired before prosody in L2 acquisition. Given the fact that none of our non-native speakers have extensive contact with native speakers, except for those who use French in the working environment, it is possible that the acquisition of prosody in L2 requires not only the contact with native speakers but also the type and/or intensity of interaction with native speakers in informal situations. This point has already been supported by many previous studies on L2 acquisition (see, for example, Bardovi-Harlig and Bastos, Reference Bardovi-Harlig and Bastos2011; Taguchi Reference Taguchi2013; Bardovi-Harlig, Reference Bardovi-Harlig2013; Taguchi and Roever, Reference Taguchi and Roever2017, Magliacane and Howard Reference Magliacane and Howard2019, etc.). This suggests that the acquisition of socio-pragmatic competence might be more complex than we thought and that acquisition of competencies at different linguistic levels might follow a certain order. However, the current data do not allow us to examine and test this hypothesis. A longitudinal study in the future might shed more light on this topic.
5.2 L1 impact
Our results also indicated that overall, non-native speakers use a higher f0 than native speakers for all functions of DM donc, regardless of the gender of the speakers. This finding has not been previously documented in speech from other L1 backgrounds. Why do Mandarin speakers behave differently? One possible explanation is the impact of the tonal patterns in their L1.
As mentioned earlier in the literature review, Chinese is a tonal language, and pitch is used at the lexical level to make semantic distinctions (Duanmu, Reference Duanmu2007). In general, there are four different tones: the first tone (a flat tone), the second tone (a rising tone), the third tone (a dipping tone), and the fourth tone (a falling tone). The first tone is a level tone maintaining a higher pitch. The second tone starts from a lower pitch and ends at a slightly higher pitch. The third tone starts at a neutral tone and then dips to a lower pitch before ending at a higher pitch. The fourth tone starts at a slightly higher than neutral pitch and then goes downwards quickly. Since French is not a tonal language, non-native speakers are aware that tones are not used to distinguish between different lexical items in French. Accordingly, they may use one tonal pattern as the default. Evidence showed that Mandarin-speaking children acquire the level tone and falling tone before rising and dipping tone (see, for example, Li & Thompson, Reference Li and Thompson1977). This suggested that the level tone and falling tone are more likely to be the default tone in Mandarin Chinese than the rising and the dipping tones. However, Liu et al. (Reference Liu, Lai, Singh, Kalashnikova, Wong, Kasisopa and Burnham2022), found that native Mandarin listeners weigh pitch height over pitch direction cue in their perception of Mandarin tones. In addition, Wayland et al. (Reference Wayland, Chen, Zhou and Hong2019) found that a change from Mandarin tone 1 to all other Mandarin tones is easier to detect than the reverse direction among both Mandarin and Cantonese listeners. Further, since only the first tone is unchanged from the starting point to the ending point, it is possible that non-native speakers, when learning French, use the first tone as the default tone for all lexical items in a non-tonal language. If this is the case, it would explain why Mandarin speakers in our study, in general, have a higher pitch than French native speakers: the level tone in L1 Mandarin Chinese, used as the default tone, is transferred to L2 French.
This observation answers our earlier question about the impact of L1 on the acquisition of DMs. Even though the target language is not tonal, speakers whose L1 is a tonal language would still use one tonal pattern in their L1 as the default tonal pattern and apply it to all lexical items in the target language at the lexical level, knowing that the semantic distinction at the lexical level is no longer made by tonal contrasts.
5.3 Gender difference
As for the question of any difference between female speakers and male speakers, the answer is affirmative. Male non-native speakers, in general, have a higher pitch than female non-native speakers when producing donc. Compared with the native patterns, we noticed that non-native female speakers produce a pattern closer to the native pattern with few variations between different pragmatic functions, while non-native male speakers conform less to the native pattern and have more significant variations between functions. As shown above, the tonal patterns in Mandarin Chinese seem to greatly influence the pitch contour pattern of donc in learners’ L2. In combination with the gender difference observed here, it seems that this impact is more pronounced for male speakers than for female speakers. Even though previous work has demonstrated that male non-native speakers are better than female non-native speakers in acquiring the informal DMs (see, for example, Deng, Reference Deng2016), in the current study, when it comes to prosodic cues, female speakers are more accurate in grasping the pitch contour patterns than male speakers. How can we explain this? Why do female speakers behave differently from male speakers?
This may be explained by earlier observations made on gender differences in language acquisition between males and females. As Ortega (Reference Ortega2009) pointed out, gender is one of the most important factors in language learning. Subsequent studies further demonstrated how female and male learners differ from each other in language learning. For example, Burman, Bitan & Booth (2008) used functional magnetic resonance imaging (fMRI) during several spelling and writing tasks and found that male learners rely more on their senses while female learners prefer a more holistic approach to language learning. Females and males differ significantly in skills they use during language acquisition. Arabski and Wojtaszek (Reference Arabski and Wojtaszek2011) provided further evidence that males outperform females in methods that use senses, such as videos, reading and writing exercises, and auditory exercises. They perform better on oral exams and exams with visual aids. In contrast, females are better than males in speech production, grammar exercises, and exercises involving abstract thinking. DMs, especially the informal ones, are lexical items mostly used and learned in informal interactions. As previous studies have already proven that males are better at auditory exercises, this explained why male speakers master the usage of informal DMs better than female speakers in L2 speech. Meanwhile, the fact that female speakers are better than male speakers in speech production also explains why female speakers’ pattern is closer to the native pattern. Hence, our findings provide further evidence on the gender difference in language acquisition.
6 CONCLUSION
In this article, by comparing native and non-native speech data, we investigated the pitch contour of the French DM donc associated with various pragmatic functions of the French DM donc. On the one hand, our results partially confirmed previous findings that non-native speakers could also use the pitch cue to realize certain pragmatic functions. Their patterns differ from the native pattern. On the other hand, we also found that the tonal pattern in speakers’ L1 significantly impacts their interlanguage. Women seem to be better than men at acquiring the native prosodic patterns in using DMs.
Our study sheds new light on how non-native speakers from a tonal language background use prosodic cues to realize different pragmatic functions in a non-tonal language. This provides new insights into the default prosodic pattern in Mandarin Chinese speakers’ speech when acquiring a foreign language. It also offers a new reflection on the acquisition of socio-pragmatic competence. Socio-pragmatic competence should not be limited to knowing how to use the proper form in an appropriate pragmatic context. We also need to know the prosodic cues associated with these functions since these cues are often important decoding features for listeners in real-life communication with native speakers.
There are several avenues for future studies. First, since we only looked at the pitch contour of the DM, it is advisable to look at other acoustic cues, such as duration of the marker, pre-segment pause, etc., to see how non-native speakers combine these features to distinguish between various functions, if at all. Second, as shown by our results, the pitch contour patterns of donc do not show much variation in native speech. It would be preferable to look at other markers that exhibit distinctive pitch contours among different pragmatic functions to see how the non-native pattern differs from the native pattern. Third, it would also be relevant to compare non-native speakers with common L1 backgrounds acquiring different L2s to see if the L1 impact is consistent.
Acknowledgement
We want to thank Rachel Meyer for her careful proofreading.
Competing interests declaration
The author(s) declare none.
Appendix
We performed GAMMs using the bam() function in the package mgcv in R (R Core Team, 2022; Wood, Reference Wood2011, Reference Wood2017), and the model outputs were visualized using the basic plotting functions in R and the plot_smooth() in the R package itsadug (van Rij et al., Reference Van Rij, Wieling, Baayen and van Rijn2015). An autocorrelation rate of 0.8 was used for all models to correct the autocorrelation in the f0 data that resulted from the dependency of the vibration of the vocal folds at consecutive time points (Chuang et al., Reference Chuang, Fon, Papakyritsis and Baayen2021). The formula of the best-fitting GAMM is provided below:
The GAMM formula used with the subsetted data for each donc type:
The GAMM formula used with the subsetted data by language group for each donc type:
Open access
