Publications

Abstract

Humans readily engage in idle chat and heated discussions and negotiate tough joint decisions without ever having to think twice about how to keep the conversation grounded in mutual understanding. However, current attempts at identifying and assessing the conversational devices that make this possible are fragmented across disciplines and investigate single devices within single contexts. We present a comprehensive conceptual framework to investigate conversational devices, their relations, and how they adjust to contextual demands. In two corpus studies, we systematically test the role of three conversational devices: backchannels, repair, and linguistic entrainment. Contrasting affiliative and task-oriented conversations within participants, we find that conversational devices adaptively adjust to the increased need for precision in the latter: We show that low-precision devices such as backchannels are more frequent in affiliative conversations, whereas more costly but higher-precision mechanisms, such as specific repairs, are more frequent in task-oriented conversations. Further, task-oriented conversations involve higher complementarity of contributions in terms of the content and perspective: lower semantic entrainment and less frequent (but richer) lexical and syntactic entrainment. Finally, we show that the observed variations in the use of conversational devices are potentially adaptive: pairs of interlocutors that show stronger linguistic complementarity perform better across the two tasks. By combining motivated comparisons of several conversational contexts and theoretically informed computational analyses of empirical data the present work lays the foundations for a comprehensive conceptual framework for understanding the use of conversational devices in dialogue.

Abstract

Establishing and maintaining mutual understanding in everyday conversations is crucial. To do so, people employ a variety of conversational devices, such as backchannels, repair, and linguistic entrainment. Here, we explore whether the use of conversational devices might be influenced by cross-linguistic differences in the speakers’ native language, comparing two matched languages—Danish and Norwegian—differing primarily in their sound structure, with Danish being more opaque, that is, less acoustically distinguished. Across systematically manipulated conversational contexts, we find that processes supporting mutual understanding in conversations vary with external constraints: across different contexts and, crucially, across languages. In accord with our predictions, linguistic entrainment was overall higher in Danish than in Norwegian, while backchannels and repairs presented a more nuanced pattern. These findings are compatible with the hypothesis that native speakers of Danish may compensate for its opaque sound structure by adopting a top-down strategy of building more conversational redundancy through entrainment, which also might reduce the need for repairs. These results suggest that linguistic differences might be met by systematic changes in language processing and use. This paves the way for further cross-linguistic investigations and critical assessment of the interplay between cultural and linguistic factors on the one hand and conversational dynamics on the other.

Abstract

A fundamental fact about human minds is that they are never truly alone: all minds are steeped in situated interaction. That social interaction matters is recognised by any experimentalist who seeks to exclude its influence by studying individuals in isolation. On this view, interaction complicates cognition. Here we explore the more radical stance that interaction co-constitutes cognition: that we benefit from looking beyond single minds towards cognition as a process involving interacting minds. All around the cognitive sciences, there are approaches that put interaction centre stage. Their diverse and pluralistic origins may obscure the fact that collectively, they harbour insights and methods that can respecify foundational assumptions and fuel novel interdisciplinary work. What might the cognitive sciences gain from stronger interactional foundations? This represents, we believe, one of the key questions for the future. Writing as a multidisciplinary collective assembled from across the classic cognitive science hexagon and beyond, we highlight the opportunity for a figure-ground reversal that puts interaction at the heart of cognition. The interactive stance is a way of seeing that deserves to be a key part of the conceptual toolkit of cognitive scientists.

Abstract

Many of the world’s languages feature an open lexical class of ideophones, words whose marked forms and sensory meanings invite iconic associations. Ideophones (also known as mimetics or expressives) are well-known from languages in Asia, Africa and the Americas, where they often form a class on the same order of magnitude as other major word classes and take up a considerable functional load as modifying expressions or predicates. Across languages, commonalities in the morphosyntactic behaviour of ideophones can be related to their nature and origin as vocal depictions. At the same time there is ample room for linguistic diversity, raising the need for fine-grained grammatical description of ideophone systems. As vocal depictions, ideophones often form a distinct lexical stratum seemingly conjured out of thin air; but as conventionalized words, they inevitably grow roots in local linguistic systems, showing relations to adverbs, adjectives, verbs and other linguistic resources devoted to modification and predication

Abstract

No class of words has better claims to universality than interjections. At the same time, no category has more variable content than this one, traditionally the catch-all basket for linguistic items that bear a complicated relation to sentential syntax. Interjections are a mirror reflecting methodological and theoretical assumptions more than a coherent linguistic category that affords unitary treatment. This chapter focuses on linguistic items that typically function as free-standing utterances, and on some of the conceptual, methodological, and theoretical questions generated by such items. A key move is to study these items in the setting of conversational sequences, rather than from the “flatland” of sequential syntax. This makes visible how some of the most frequent interjections streamline everyday language use and scaffold complex language. Approaching interjections in terms of their sequential positions and interactional functions has the potential to reveal and explain patterns of universality and diversity in interjections.

Abstract

Large language models that exhibit instruction-following behaviour represent one of the biggest recent upheavals in conversational interfaces, a trend in large part fuelled by the release of OpenAI's ChatGPT, a proprietary large language model for text generation fine-tuned through reinforcement learning from human feedback (LLM+RLHF). We review the risks of relying on proprietary software and survey the first crop of open-source projects of comparable architecture and functionality. The main contribution of this paper is to show that openness is differentiated, and to offer scientific documentation of degrees of openness in this fast-moving field. We evaluate projects in terms of openness of code, training data, model weights, RLHF data, licensing, scientific documentation, and access methods. We find that while there is a fast-growing list of projects billing themselves as 'open source', many inherit undocumented data of dubious legality, few share the all-important instruction-tuning (a key site where human labour is involved), and careful scientific documentation is exceedingly rare. Degrees of openness are relevant to fairness and accountability at all points, from data collection and curation to model architecture, and from training and fine-tuning to release and deployment.

Abstract

Speech recognition systems are a key intermediary in voice-driven human-computer interaction. Although speech recognition works well for pristine monologic audio, real-life use cases in open-ended interactive settings still present many challenges. We argue that timing is mission-critical for dialogue systems, and evaluate 5 major commercial ASR systems for their conversational and multilingual support. We find that word error rates for natural conversational data in 6 languages remain abysmal, and that overlap remains a key challenge (study 1). This impacts especially the recognition of conversational words (study 2), and in turn has dire consequences for downstream intent recognition (study 3). Our findings help to evaluate the current state of conversational ASR, contribute towards multidimensional error analysis and evaluation, and identify phenomena that need most attention on the way to build robust interactive speech technologies.

Abstract

Iconicity in language is receiving increased attention from many fields, but our understanding of iconicity is only as good as the measures we use to quantify it. We collected iconicity measures for 304 Japanese words from English-speaking participants, using rating and guessing tasks. The words included ideophones (structurally marked depictive words) along with regular lexical items from similar semantic domains (e.g., fuwafuwa ‘fluffy’, jawarakai ‘soft’). The two measures correlated, speaking to their validity. However, ideophones received consistently higher iconicity ratings than other items, even when guessed at the same accuracies, suggesting the rating task is more sensitive to cues like structural markedness that frame words as iconic. These cues did not always guide participants to the meanings of ideophones in the guessing task, but they did make them more confident in their guesses, even when they were wrong. Consistently poor guessing results reflect the role different experiences play in shaping construals of iconicity. Using multiple measures in tandem allows us to explore the interplay between iconicity and these external factors. To facilitate this, we introduce a reproducible workflow for creating rating and guessing tasks from standardised wordlists, while also making improvements to the robustness, sensitivity and discriminability of previous approaches.

Abstract

What is language and who or what can be said to have it? In this essay we consider this question in the context of interactions with non-humans, specifically: animals and computers. While perhaps an odd pairing at first glance, here we argue that these domains can offer contrasting perspectives through which we can explore and reimagine language. The interactions between humans and animals, as well as between humans and computers, reveal both the essence and the boundaries of language: from examining the role of sequence and contingency in human-animal interaction, to unravelling the challenges of natural interactions with “smart” speakers and language models. By bringing together disparate fields around foundational questions, we push the boundaries of linguistic inquiry and uncover new insights into what language is and how it functions in diverse non-humanexclusive contexts.

Abstract

Prosociality and cooperation are key to what makes us human. But different cultural norms can shape our evolved capacities for interaction, leading to differences in social relations. How people share resources has been found to vary across cultures, particularly when stakes are high and when interactions are anonymous. Here we examine prosocial behavior among familiars (both kin and non-kin) in eight cultures on five continents, using video recordings of spontaneous requests for immediate, low-cost assistance (e.g., to pass a utensil). We find that, at the smallest scale of human interaction, prosocial behavior follows cross-culturally shared principles: requests for assistance are very frequent and mostly successful; and when people decline to give help, they normally give a reason. Although there are differences in the rates at which such requests are ignored, or require verbal acceptance, cultural variation is limited, pointing to a common foundation for everyday cooperation around the world.

Abstract

Iconicity, or the resemblance between form and meaning, is often ascribed to a special status and contrasted with default assumptions of arbitrariness in spoken language. But does iconicity in spoken language have a special status when it comes to learnability? A simple way to gauge learnability is to see how well something is retrieved from memory. We can further contrast this with guessability, to see (1) whether the ease of guessing the meanings of ideophones outperforms the rate at which they are remembered; and (2) how willing participants’ are to reassess what they were taught in a prior task—a novel contribution of this study. We replicate prior guessing and memory tasks using ideophones and adjectives from Japanese, Korean, and Igbo. Our results show that although native Cantonese speakers guessed ideophone meanings above chance level, they memorized both ideophones and adjectives with comparable accuracy. However, response time data show that participants took significantly longer to respond correctly to adjective–meaning pairs—indicating a discrepancy in a cognitive effort that favored the recognition of ideophones. In a follow-up reassessment task, participants who were taught foil translations were more likely to choose the true translations for ideophones rather than adjectives. By comparing the findings from our guessing and memory tasks, we conclude that iconicity is more accessible if a task requires participants to actively seek out sound-meaning associations.

Abstract

Continuers —words like mm, mmhm, uhum and the like— are among the most frequent types of responses in conversation. They play a key role in joint action coordination by showing positive evidence of understanding and scaffolding narrative delivery. Here we investigate the hypothesis that their functional importance along with their conversational ecology places selective pressures on their form and may lead to cross-linguistic similarities through convergent cultural evolution. We compare continuer tokens in linguistically diverse conversational corpora and find languages make available highly similar forms. We then approach the causal mechanism of convergent cultural evolution using exemplar modelling, simulating the process by which a combination of effort minimization and functional specialization may push continuers to a particular region of phonological possibility space. By combining comparative linguistics and computational modelling we shed new light on the question of how language structure is shaped by and for social interaction.

Abstract

Informal social interaction is the primordial home of human language. Linguistically diverse conversational corpora are an important and largely untapped resource for computational linguistics and language technology. Through the efforts of a worldwide language documentation movement, such corpora are increasingly becoming available. We show how interactional data from 63 languages (26 families) harbours insights about turn-taking, timing, sequential structure and social action, with implications for language technology, natural language understanding, and the design of conversational interfaces. Harnessing linguistically diverse conversational corpora will provide the empirical foundations for flexible, localizable, humane language technologies of the future.

Abstract

We present a dataset of behavioural and fMRI observations acquired in the context of humans involved in multimodal referential communication. The dataset contains audio/video and motion-tracking recordings of face-to-face, task-based communicative interactions in Dutch, as well as behavioural and neural correlates of participants’ representations of dialogue referents. Seventy-one pairs of unacquainted participants performed two interleaved interactional tasks in which they described and located 16 novel geometrical objects (i.e., Fribbles) yielding spontaneous interactions of about one hour. We share high-quality video (from three cameras), audio (from head-mounted microphones), and motion-tracking (Kinect) data, as well as speech transcripts of the interactions. Before and after engaging in the face-to-face communicative interactions, participants’ individual representations of the 16 Fribbles were estimated. Behaviourally, participants provided a written description (one to three words) for each Fribble and positioned them along 29 independent conceptual dimensions (e.g., rounded, human, audible). Neurally, fMRI signal evoked by each Fribble was measured during a one-back working-memory task. To enable functional hyperalignment across participants, the dataset also includes fMRI measurements obtained during visual presentation of eight animated movies (35 minutes total). We present analyses for the various types of data demonstrating their quality and consistency with earlier research. Besides high-resolution multimodal interactional data, this dataset includes different correlates of communicative referents, obtained before and after face-to-face dialogue, allowing for novel investigations into the relation between communicative behaviours and the representational space shared by communicators. This unique combination of data can be used for research in neuroscience, psychology, linguistics, and beyond.

Abstract

Human joint action is inherently cooperative, manifested in the collaborative efforts of participants to minimize communicative trouble through interactive repair. Although interactive repair requires sophisticated cognitive abilities,
it can be dissected into basic building blocks shared with non-human animal species. A review of the primate literature shows that interactionally contingent signal sequences are at least common among species of nonhuman great apes, suggesting a gradual evolution of repair. To pioneer a cross-species assessment of repair this paper aims at (i) identifying necessary precursors of human interactive repair; (ii) proposing a coding framework for its comparative study in humans and non-human species; and (iii) using this framework to analyse examples of interactions of humans (adults/children) and non-human great apes. We hope this paper will serve as a primer for cross-species comparisons of communicative breakdowns and how they are repaired.

Abstract

Response tokens (also known as backchannels, continuers, or feedback) are a frequent feature of human interaction, where they serve to display understanding and streamline turn-taking. We propose a bottom-up method to study responsive behaviour across 16 languages (8 language families). We use sequential context and recurrence of turns formats to identify candidate response tokens in a language-agnostic way across diverse conversational corpora. We then use UMAP clustering directly on speech signals to represent structure and variation. We find that (i) written orthographic annotations underrepresent the attested variation, (ii) distinctions between formats can be gradient rather than discrete, (iii) most languages appear to make available a broad distinction between a minimal nasal format `mm' and a fuller `yeah’-like format. Charting this aspect of human interaction contributes to our understanding of interactional infrastructure across languages and can inform the design of speech technologies.

Abstract

We present an analysis pipeline and best practice guidelines for building and curating corpora of everyday conversation in diverse languages. Surveying language documentation corpora and other resources that cover 67 languages and varieties from 28 phyla, we describe the compilation and curation process, specify minimal properties of a unified format for interactional data, and develop methods for quality control that take into account turn-taking and timing. Two case studies show the broad utility of conversational data for (i) charting human interactional infrastructure and (ii) tracing challenges and opportunities for current ASR solutions. Linguistically diverse conversational corpora can provide new insights for the language sciences and stronger empirical foundations for language technology.

Abstract

How does communicative efficiency shape language use? We approach this question by studying it at the level of the dyad, and in terms of multimodal utterances. We investigate whether and how people minimize their joint speech and gesture efforts in face-to-face interactions, using linguistic and kinematic analyses. We zoom in on other-initiated repair—a conversational microcosm where people coordinate their utterances to solve problems with perceiving or understanding. We find that efforts in the spoken and gestural modalities are wielded in parallel across repair turns of different types, and that people repair conversational problems in the most cost-efficient way possible, minimizing the joint multimodal effort for the dyad as a whole. These results are in line with the principle of least collaborative effort in speech and with the reduction of joint costs in non-linguistic joint actions. The results extend our understanding of those coefficiency principles by revealing that they pertain to multimodal utterance design.

Abstract

When people establish shared symbols for novel objects or concepts, they have been shown to rely on the use of multiple communicative modalities as well as on alignment (i.e., cross-participant repetition of communicative behavior). Yet these interactional resources have rarely been studied together, so little is known about if and how people combine multiple modalities in alignment to achieve joint reference. To investigate this, we systematically track the emergence of lexical and gestural alignment in a referential communication task with novel objects. Quantitative analyses reveal that people frequently use a combination of lexical and gestural alignment, and that such multimodal alignment tends to emerge earlier compared to unimodal alignment. Qualitative analyses of the interactional contexts in which alignment emerges reveal how people flexibly deploy lexical and gestural alignment (independently, simultaneously or successively) to adjust to communicative pressures.

Abstract

Ook al hebben we het niet altijd door, onze zintuigen werken altijd samen. Als je iemand ziet praten, bijvoorbeeld, verwerken je hersenen automatisch tegelijkertijd het geluid van de woorden en de bewegingen van de lippen. Omdat onze zintuigen altijd samenwerken zijn onze hersenen erg gevoelig voor dingen die ‘samenhoren’ en goed bij elkaar passen. In dit hoofdstuk beschrijven we een project onderzoekend leren met als thema ‘Samenwerkende zintuigen’.