Programme

What do listeners think of disfluencies?

Recent research shows that listeners are sensitive to disfluencies in unfolding discourse, using them to modify predictions about what will be said. But how do they interpret the pragmatic content of the message? I present a series of experiments examining these issues using a "lying game" in which speakers identify the locations of treasure, either fluently or disfluently. Results show that listeners robustly interpret disfluency as a clue to dishonesty, and that this happens early in comprehension, even where other evidence is available. However, when there is a plausible exogenous cause of the disfluency such as speaker distraction, this effect is modulated, suggesting that listeners are causally interpreting the disfluencies uttered. It is therefore surprising that an interactive version of the paradigm where speakers' utterances are freely generated suggests that listeners' judgements may be misguided: Speakers tend to be more disfluent when telling the truth, although listeners steadfastly continue to distrust disfluent speech.

Sandra Götz

(Justus Liebig University Giessen)

Do (non-linguistic) variables affect learners’ (dis)fluency?
A learner corpus-based perspective.

It has been noted by various scholars that second-language acquisition (SLA) theory needs to take into account learning context as a determining factor in a learners’ spoken fluency (e.g. Norris and Ortega 2001; Freed et al. 2004). Researchers in the field of language testing, as well, have documented a considerable degree of variability in learners’ fluency, depending on non-linguistic context variables, such as the test-taker’s or interlocutor’s gender, the communicative style during the test or the task type (e.g. Porter 1991; Kormos 1999; O’Sullivan 2000; Csépes 2009). However, we face a number of challenges in systematising the multitude of variables that affect learners’ fluency performance, because it is very difficult to operationalize them quantitatively. Consequently, empirical research into interrelations between these non-linguistic learning context variables and language learners’ fluency far has mainly either focused on a small number of learners (in the fields of SLA and language testing) or a small set of non-linguistic learning context variables (in learner corpus research). Thus, if we want to abstract from the fluency development of individual learners to the general picture of the output of a greater number of typical learners regardless of the learners’ L1, we need to take into account comparable data from many learners that have in common a defined set of typical and representative (non-linguistic) variables. This is one of the central rationales in compiling learner corpora as repositories of learners’ natural language use (cf. Ellis 1994; Granger 2002).

In this talk, I will discuss how learner corpus research into fluency can benefit immensely from taking into consideration learning context variables gathered from the learner profiles of the Louvain International Database of Spoken English Interlanguage (LINDSEI; cf. Gilquin et al. 2010). LINDSEI includes data of learners with 11 different L1s and provides a variety of meta-data on each individual learner in the corpus (e.g. sociobiographic data, languages the learners have been exposed to, the number of years of English instruction at school and at university, time spent abroad, or other language(s) spoken by the learner), as well as on the interviewers (e.g. gender, languages spoken, familiarity with the interviewee). By way of presenting the findings of some case studies conducted on LINDSEI, I will discuss the effect of these non-linguistic variables on learners’ fluency, including the fluency development after a stay abroad or the effect of sociolinguistic variables on the learner’s fluency performance regardless of their L1. Finally, the implications of the findings of these case studies will be outlined, particularly focusing on the implications on learner corpus-based fluency research and the fluency development of advanced learners of English.

References

• Csépes, I. 2009. Measuring oral proficiency through paired-task performance. Frankfurt am Main: Peter Lang.
• Ellis, R. 1994. The study of second language acquisition. Oxford: Oxford University Press.
• Freed, B., Segalowitz, N. and Dewy, D. P. 2004. “Context of learning and second language fluency in French: comparing regular classroom, study abroad, and intensive domestic immersion programs”. Studies in Second Language Acquisition 26(2), 275-301.
• Gilquin, G., De Cock, S. and Granger, S. 2010. Louvain International Database of Spoken English Interlanguage. Handbook and CD-ROM. Louvain-la-Neuve: Presses universitaires de Louvain.
• Granger, S. 2002. “A bird’s eye view of learner corpus research”. In S. Granger, J. Hung and S. Petch-Tyson (eds) Computer Learner Corpora, Second Language Acquisition and Foreign Language Teaching. Amsterdam: John Benjamins, 3-33.
• Kormos, J. 1999. Simulating conversations in oral-proficiency assessment: A conversation analysis of role plays and non-scripted interviews in language exams. Language Testing 16(2):163-188.
• Norris, J.M. and Ortega, L. 2001. “Does type of instruction make a difference? Substantive findings from a meta-analytic review”, Language Learning 51 (s1), 157-213.
• O’Sullivan, B. 2000. Exploring gender and oral proficiency interview performance. System 28(3), 373-386.
• Porter, D. 1991. Affective factors in language testing. In J.C. Alderson & B. North (eds), Language testing in the 1990s. London: Macmillan, 32-40.

Modeling disfluencies across domains

Disfluencies are on-line editing strategies with several (para)linguistic functions. Everyday we are annalists of our own speech and of others, monitoring distinct linguistic and paralinguistic factors in our communications, using disfluencies to make speech a more error-free system, a more edited message, and a more structured system with coherent and cohesive mechanisms.

This presentation focuses on the analysis of disfluencies, aiming at a characterization of the regular patterns in their production in European Portuguese, and at contributing towards the fully automatic processing of structural metadata events. This analysis was strongly supported on prosodic feature processing, and involved corpora of very different characteristics. For that purpose a framework was built for metadata annotation including prosodic features, a crucial step for Portuguese, since prior to this work our in-house speech recognizer had no integration of such features. This framework allowed us to access several layers of linguistic information (e.g., acoustic-prosodic, POS, pragmatic) in a very flexible way and proved to be a suitable tool for the analysis of metadata events.

The robustness of acoustic-prosodic features across domains was investigated using university lectures and dialogues. Different models trained with one corpus were tested on the other, revealing that models can be quite robust across corpora for this task, despite their distinct nature. The model trained on dialogues proved to be the more robust one, possibly due to the fact that dialogues contain more contrastive tempo characteristics, while sharing with university lectures most of the pitch and energy patterns on disfluent sequences. Therefore, a model created with such data generalizes better.

In our current research, we try to extend this study to other domains, including human-computer interactions both with virtual and embodied agents.

Signs of (dis)fluency throughout development: The language use of Deaf children who are native users of a signed language considering adult examples of (dis)fluency

Native users of signed languages are notably perceptive about a stranger’s signed language skills. Lifelong signers can typically determine whether someone they have never met before is a native signer, like them, or if the stranger was exposed to the signed language sometime after birth—especially if their exposure did not occur until late in childhood or beyond. Not unlike the perceptual abilities of native users of spoken languages, signers may be utilizing multiple cues—some of which are subtle and others more overt—that signal whether a person had early exposure to the visual-gestural language. Such cues might be considered among the metrics for an individual’s fluency in a signed language. However, what can be said about judgments of a native speaker’s fluency in a signed language, especially if that speaker is a child?

Empirical studies of so-called fluency markers in a signed language are not numerous, although several studies of fluencemes in French Belgian Sign Language (LSFB) have highlighted markers of fluency in adult native signers, including rate of signing (Notarrigo & Meurant 2015) and use of repetition (Notarrigo, Meurant, & Simon 2016), among others. Such studies have pushed the boundaries of where signed language researchers look for effects of early language exposure on language processing and use.
Other work on fluency in signed language has questioned the role of multilingualism. For example, a fluent signer of one sign language who also knows another sign language might exhibit examples of a signed “accent”. This has been shown for adult signers of American Sign Language (ASL) and Mexican Sign Language (LSM) by Quinto-Pozos (2002, 2008) and signers of Al-Sayyid Bedouin Sign Language (ABSL) and Israeli Sign Language (ISL) by Sandler (2014). These authors share a focus on aspects of phonetics and phonology in the language use of late-learners, highlighting features such as handshape, movement, and so-called ‘hand prominence’ (fingertip, radial, or ulnar prominence in articulation).

What about fluency within a deaf user’s first language, especially during development? Judging a native signer’s fluency at various stages in their childhood is particularly challenging task because children are notably variable in their linguistic and psychosocial development. What are the cues that lead a fluent signer to judge a child’s signed language use as (less-than) fluent? Are there differences between disfluency that is caused by late exposure to language versus disfluency that co-occurs with a developmental disorder or deficit (e.g., of language and/or cognition)?

I will discuss various aspects of fluency in signed language use. Studies of adults that focus on aspects of phonetics and phonology will be reviewed in order to establish what has been discussed in terms of second-language (L2) use. In addition, the presentation will highlight aspects of childhood development that may signal unexpected disfluency, according to reports from educators and developmental specialists at schools for the deaf and case study data that we have collected in our lab.Signed language use by deaf children who are native signers provides a fertile ground for investigating aspects of fluency in the visual-gestural modality throughout development.

References
Notarrigo, I & Meurant, L. (2015, July). Markers of (dis)fluency across signers’ profiles in French Belgian Sign Language (LSFB) A comparative analysis between Native, Near-Native and Late Signers. Poster Presentation. 2nd International Conference on Sign Language Acquisition (ICSLA). Amsterdam, The Netherlands.
Notarrigo, I., Meurant, L., & Simon, A.C. (2016, January). Repetition of signs according to language background in French Belgian Sign Language (LSFB): A comparative analysis between Native, Near-Native and Late Signers. Poster Presentation. Theoretical Issues in Sign Language Research (TISLR) 12. Melbourne, Australia.
Quinto-Pozos, D. (2002). Contact between Mexican Sign Language and American Sign Language in Two Texas Border Town. Unpublished doctoral dissertation. University of Texas at Austin.
Quinto-Pozos, D. (2008). Sign language contact & interference: ASL & LSM. Language in Society 37, 2. 161-189.
Sandler, W. (2014). The emergence of the phonetic and phonological features in sign language. Nordlyd 41.2: 183-212,