Publications

Abstract

Conversation requires a substantial amount of coordination between dialogue participants, from managing turn taking to negotiating mutual understanding. Part of this coordination effort surfaces as the reuse of linguistic behaviour across speakers, a process often referred to as alignment. While the presence of linguistic alignment is well documented in the literature, several questions remain open, including the extent to which patterns of reuse across speakers have an impact on the emergence of labelling conventions for novel referents. In this study, we put forward a methodology for automatically detecting shared lemmatised constructions---expressions with a common lexical core used by both speakers within a dialogue---and apply it to a referential communication corpus where participants aim to identify novel objects for which no established labels exist. Our analyses uncover the usage patterns of shared constructions in interaction and reveal that features such as their frequency and the amount of different constructions used for a referent are associated with the degree of object labelling convergence the participants exhibit after social interaction. More generally, the present study shows that automatically detected shared constructions offer a useful level of analysis to investigate the dynamics of reference negotiation in dialogue.

Abstract

Gestures are integral components of face-to-face communication. They unfold over time, often following predictable movement phases of preparation, stroke, and re-
traction. Yet, the prevalent approach to automatic gesture detection treats the problem as binary classification, classifying a segment as either containing a gesture or not, thus failing to capture its inherently sequential and contextual nature. To address this, we introduce a novel framework that reframes the task as a multi-phase sequence labeling problem rather than binary classification. Our model processes sequences of skeletal movements over time windows, uses Transformer encoders to learn contextual embeddings, and leverages Conditional Random Fields to perform sequence labeling. We evaluate our proposal on a large dataset of diverse co-speech gestures in task-oriented face-to-face dialogues. The results consistently demonstrate that our method significantly outperforms strong baseline models in detecting gesture strokes. Furthermore, applying Transformer encoders to learn contextual embeddings from movement sequences substantially improves gesture unit detection. These results highlight our framework’s capacity to capture the fine-grained dynamics of co-speech gesture phases, paving the way for more nuanced and accurate gesture detection and analysis.

Abstract

In face-to-face dialogues, the form-meaning relationship of co-speech gestures varies depending on contextual factors such as what the gestures refer to and the individual characteristics of speakers. These factors make co-speech gesture representation learning challenging. How can we learn meaningful gestures representations considering gestures’ variability and relationship with speech? This paper tackles this challenge by employing self-supervised contrastive learning techniques to learn gesture representations from skeletal and speech information. We propose an approach that includes both unimodal and multimodal pre-training to ground gesture representations in co-occurring speech. For training, we utilize a face-to-face dialogue dataset rich with representational iconic gestures. We conduct thorough intrinsic evaluations of the learned representations through comparison with human-annotated pairwise gesture similarity. Moreover, we perform a diagnostic probing analysis to assess the possibility of recovering interpretable gesture features from the learned representations. Our results show a significant positive correlation with human-annotated gesture similarity and reveal that the similarity between the learned representations is consistent with well-motivated patterns related to the dynamics of dialogue interaction. Moreover, our findings demonstrate that several features concerning the form of gestures can be recovered from the latent representations. Overall, this study shows that multimodal contrastive learning is a promising approach for learning gesture representations, which opens the door to using such representations in larger-scale gesture analysis studies.

Abstract

Purpose

To perform a scoping review to investigate the psychosocial impact of having an altered facial expression in five neurological diseases.
Methods

A systematic literature search was performed. Studies were on Bell’s palsy, facioscapulohumeral muscular dystrophy (FSHD), Moebius syndrome, myotonic dystrophy type 1, or Parkinson’s disease patients; had a focus on altered facial expression; and had any form of psychosocial outcome measure. Data extraction focused on psychosocial outcomes.
Results

Bell’s palsy, myotonic dystrophy type 1, and Parkinson’s disease patients more often experienced some degree of psychosocial distress than healthy controls. In FSHD, facial weakness negatively influenced communication and was experienced as a burden. The psychosocial distress applied especially to women (Bell’s palsy and Parkinson’s disease), and patients with more severely altered facial expression (Bell’s palsy), but not for Moebius syndrome patients. Furthermore, Parkinson’s disease patients with more pronounced hypomimia were perceived more negatively by observers. Various strategies were reported to compensate for altered facial expression.
Conclusions

This review showed that patients with altered facial expression in four of five included neurological diseases had reduced psychosocial functioning. Future research recommendations include studies on observers’ judgements of patients during social interactions and on the effectiveness of compensation strategies in enhancing psychosocial functioning.
Implications for rehabilitation

Negative effects of altered facial expression on psychosocial functioning are common and more abundant in women and in more severely affected patients with various neurological disorders.

Health care professionals should be alert to psychosocial distress in patients with altered facial expression.

Learning of compensatory strategies could be a beneficial therapy for patients with psychosocial distress due to an altered facial expression.

Abstract

During face-to-face conversation, transitions between speaker turns are incredibly fast. These fast turn exchanges seem to involve next speakers predicting upcoming semantic information, such that next turn planning can begin before a current turn is complete. Given that face-to-face conversation also involves the use of communicative bodily signals, an important question is how bodily signals such as co-speech hand gestures play into these processes of prediction and fast responding. In this corpus study, we found that hand gestures that depict or refer to semantic information started before the corresponding information in speech, which held both for the onset of the gesture as a whole, as well as the onset of the stroke (the most meaningful part of the gesture). This early timing potentially allows listeners to use the gestural information to predict the corresponding semantic information to be conveyed in speech. Moreover, we provided further evidence that questions with gestures got faster responses than questions without gestures. However, we found no evidence for the idea that how much a gesture precedes its lexical affiliate (i.e., its predictive potential) relates to how fast responses were given. The findings presented here highlight the importance of the temporal relation between speech and gesture and help to illuminate the potential mechanisms underpinning multimodal language processing during face-to-face conversation.

Abstract

Most language use occurs in face-to-face conversation, which involves rapid turn-taking. Seeing communicative bodily signals in addition to hearing speech may facilitate such fast responding. We tested whether this holds for co-speech hand gestures by investigating whether these gestures speed up button press responses to questions. Sixty native speakers of Dutch viewed videos in which an actress asked yes/no-questions, either with or without a corresponding iconic hand gesture. Participants answered the questions as quickly and accurately as possible via button press. Gestures did not impact response accuracy, but crucially, gestures sped up responses, suggesting that response planning may be finished earlier when gestures are seen. How much gestures sped up responses was not related to their timing in the question or their timing with respect to the corresponding information in speech. Overall, these results are in line with the idea that multimodality may facilitate fast responding during face-to-face conversation.

Abstract

The typical pattern of alternating turns in conversation seems trivial at first sight. But a closer look quickly reveals the cognitive challenges involved, with much of it resulting from the fast-paced nature of conversation. One core ingredient to turn coordination is the anticipation of upcoming turn ends so as to be able to ready oneself for providing the next contribution. Across two experiments, we investigated two variables inherent to face-to-face conversation, the presence of visual bodily signals and preceding discourse context, in terms of their contribution to turn end anticipation. In a reaction time paradigm, participants anticipated conversational turn ends better when seeing the speaker and their visual bodily signals than when they did not, especially so for longer turns. Likewise, participants were better able to anticipate turn ends when they had access to the preceding discourse context than when they did not, and especially so for longer turns. Critically, the two variables did not interact, showing that visual bodily signals retain their influence even in the context of preceding discourse. In a pre-registered follow-up experiment, we manipulated the visibility of the speaker's head, eyes and upper body (i.e. torso + arms). Participants were better able to anticipate turn ends when the speaker's upper body was visible, suggesting a role for manual gestures in turn end anticipation. Together, these findings show that seeing the speaker during conversation may critically facilitate turn coordination in interaction.

Abstract

Human language is extremely versatile, combining a limited set of signals in an unlimited number of ways. However, it is unknown whether conversational visual signals feed into the composite utterances with which speakers communicate their intentions. We assessed whether different combinations of visual signals lead to different intent interpretations of the same spoken utterance. Participants viewed a virtual avatar uttering spoken questions while producing single visual signals (i.e., head turn, head tilt, eyebrow raise) or combinations of these signals. After each video, participants classified the communicative intention behind the question. We found that composite utterances combining several visual signals conveyed different meaning compared to utterances accompanied by the single visual signals. However, responses to combinations of signals were more similar to the responses to related, rather than unrelated, individual signals, indicating a consistent influence of the individual visual signals on the whole. This study therefore provides first evidence for compositional, non-additive (i.e., Gestalt-like) perception of multimodal language.

Abstract

The natural ecology of language is conversation, with individuals taking turns speaking to communicate in a back-and-forth fashion. Language in this context involves strings of words that a listener must process while simultaneously planning their own next utterance. It would thus be highly advantageous if language users distributed information within an utterance in a way that may facilitate this processing–planning dynamic. While some studies have investigated how information is distributed at the level of single words or clauses, or in written language, little is known about how information is distributed within spoken utterances produced during naturalistic conversation. It also is not known how information distribution patterns of spoken utterances may differ across languages. We used a set of matched corpora (CallHome) containing 898 telephone conversations conducted in six different languages (Arabic, English, German, Japanese, Mandarin, and Spanish), analyzing more than 58,000 utterances, to assess whether there is evidence of distinct patterns of information distributions at the utterance level, and whether these patterns are similar or differed across the languages. We found that English, Spanish, and Mandarin typically show a back-loaded distribution, with higher information (i.e., surprisal) in the last half of utterances compared with the first half, while Arabic, German, and Japanese showed front-loaded distributions, with higher information in the first half compared with the last half. Additional analyses suggest that these patterns may be related to word order and rate of noun and verb usage. We additionally found that back-loaded languages have longer turn transition times (i.e.,time between speaker turns)