Publications

Abstract

In this work we perform multimodal detection and classification
of head movements from face to face video conversation data.
We have experimented with different models and feature sets
and provided some insight on the effect of independent features,
but also how their interaction can enhance a head movement
classifier. Used features include nose, neck and mid hip position
coordinates and their derivatives together with acoustic features,
namely, intensity and pitch of the speaker on focus. Results
show that when input features are sufficiently processed by in-
teracting with each other, a linear classifier can reach a similar
performance to a more complex non-linear neural model with
several hidden layers. Our best models achieve state-of-the-art
performance in the detection task, measured by macro-averaged
F1 score.

Abstract

The present paper assesses discourse-pragmatic factors as a potential explanation for the subject-object assymetry in early child language. It identifies a set of factors which characterize typical situations of informativeness (Greenfield & Smith, 1976), and uses these factors to identify informative arguments in data from four children aged 2;0 through 3;6 learning Inuktitut as a first language. In addition, it assesses the extent of the links between features of informativeness on one hand and lexical vs. null and subject vs. object arguments on the other. Results suggest that a pragmatics account of the subject-object asymmetry can be upheld to a greater extent than previous research indicates, and that several of the factors characterizing informativeness are good indicators of those arguments which tend to be omitted in early child language.

Abstract

Reprinted from Bowerman, M. (1973). Structural relationships in children's utterances: Semantic or syntactic? In T. Moore (Ed.), Cognitive development and the acquisition of language (pp. 197 213). New York: Academic Press

Abstract

The surge of research activity focussing on children's acquisition of verbs (e.g., Tomasello and Merriman 1996) addresses some fundamental questions: Just how variable across languages, and across individual children, is the process of verb learning? How specific are arguments to particular verbs in early child language? How does the grammatical category 'Verb' develop? The position of Universal Grammar, that a verb category is early, contrasts with that of Tomasello (1992), Pine and Lieven and their colleagues (1996, in press), and many others, that children develop a verb category slowly, gradually building up subcategorizations of verbs around pragmatic, syntactic, and semantic properties of the language they are exposed to. On this latter view, one would expect the language which the child is learning, the cultural milieu and the nature of the interactions in which the child is engaged, to influence the process of acquiring verb argument structures. This paper explores these issues by examining the development of argument representation in the Mayan language Tzeltal, in both its lexical and verbal cross-referencing forms, and analyzing the semantic and pragmatic factors influencing the form argument representation takes. Certain facts about Tzeltal (the ergative/ absolutive marking, the semantic specificity of transitive and positional verbs) are proposed to affect the representation of arguments. The first 500 multimorpheme combinations of 3 children (aged between 1;8 and 2;4) are examined. It is argued that there is no evidence of semantically light 'pathbreaking' verbs (Ninio 1996) leading the way into word combinations. There is early productivity of cross-referencing affixes marking A, S, and O arguments (although there are systematic omissions). The paper assesses the respective contributions of three kinds of factors to these results - structural (regular morphology), semantic (verb specificity) and pragmatic (the nature of Tzeltal conversational interaction).

Abstract

In this paper I take up the idea that human thinking is systematically biased in the direction of interactive thinking (E. Goody's anticipatory interactive planning), that is, that humans are peculiarly good at, and inordinately prone to, attributing intentions and goals to one other (as well as to non-humans), and that they routinely orient to presumptions about each other's intentions in what they say and do. I explore the implications of that idea for an understanding of politeness in interaction, taking as a starting point the Brown and Levinson (1987) model of politeness, which assumes interactive thinking, a notion implicit in the formulation of politeness as strategic orientation to face. Drawing on an analysis of the phenomenon of conventionalized ‘irony’ in Tzeltal, I emphasize that politeness does not inhere in linguistic form per se but is a matter of conveying a polite intention, and argue that Tzeltal irony provides a prime example of one way in which humans' highly-developed intellectual machinery for inferring alter's intentions is put to the service of social relationships.

Abstract

This article is a reprint of chapter 1, the introduction to Brown and Levinson, 1987, Politeness: Some universals in language usage (Cambridge University Press).

Abstract

This study is about the principles for constructing polite speech. We describe and account for some remarkable parallelisms in the linguistic construction of utterances with which people express themselves in different languges and cultures. A motive for these parallels is isolated - politeness, broadly defined to include both polite friendliness and polite formality - and a universal model is constructed outlining the abstract principles underlying polite usages. This is based on the detailed study of three unrelated languages and cultures: the Tamil of south India, the Tzeltal spoken by Mayan Indians in Chiapas, Mexico, and the English of the USA and England, supplemented by examples from other cultures. Of general interest is the point that underneaath the apparent diversity of polite behaviour in different societies lie some general pan-human principles of social interaction, and the model of politenss provides a tool for analysing the quality of social relations in any society.

Abstract

This chapter provides an introduction to the handbook. It succintly overviews the key questions in the field of L3/Ln acquisition and summarizes the scope of all the chapters included. The chapter ends by raising some outstanding questions that the field needs to address.

Abstract

In order to understand early word learning we need to better understand and quantify properties of the input that young children receive. We extended the human simulation paradigm (HSP) using egocentric videos taken from infant head-mounted cameras. The videos were further annotated with gaze information indicating in-the-moment visual attention from the infant. Our new HSP prompted participants for two types of responses, thus differentiating referential from semantic ambiguity in the learning input. Consistent with findings on visual attention in word learning, we find a strongly bimodal distribution over HSP accuracy. Even in this open-ended task, most videos only lead to a small handful of common responses. What's more, referential ambiguity was the key bottleneck to performance: participants can nearly always recover the exact word that was said if they identify the correct referent. Finally, analysis shows that adult learners relied on particular, multimodal behavioral cues to infer those target referents.

Abstract

In this study, we analyze one French signing and one French speaking family’s interaction during dinner. The families composed of two parents and two children aged 3 to 11 were filmed with three cameras to capture all family members’ behaviors. The three videos per dinner were synchronized and coded on ELAN. We annotated all participants’ acting, and languaging.
Our quantitative analyses show how family members collaboratively manage multiple streams of activity through the embodied performances of dining and interacting. We uncover different profiles according to participants’ modality of expression and status (focusing on the mother and the younger child). The hearing participants’ co-activity management illustrates their monitoring of dining and conversing and how they progressively master the affordances of the visual and vocal channels to maintain the simultaneity of the two activities. The deaf mother skillfully manages to alternate smoothly between dining and interacting. The deaf younger child manifests how she is in the process of developing her skills to manage multi-activity. Our qualitative analyses focus on the ecology of visual-gestural and audio-vocal languaging in the context of co-activity according to language and participant. We open new perspectives on the management of gaze and body parts in multimodal languaging.

Abstract

During dialog, interlocutors take turns at speaking with little gap or overlap between their contributions. But language production in monolog is comparatively slow. Theories of dialog tend to agree that interlocutors manage these timing demands by planning a response early, before the current speaker reaches the end of their turn. In the first half of this chapter, I review experimental research supporting these theories. But this research also suggests that planning a response early, while simultaneously comprehending, is difficult. Does response planning need to be this difficult during dialog? In other words, is early-planning always necessary? In the second half of this chapter, I discuss research that suggests the answer to this question is no. In particular, corpora of natural conversation demonstrate that speakers do not directly respond to the immediately preceding utterance of their partner—instead, they continue an utterance they produced earlier. This parallel talk likely occurs because speakers are highly incremental and plan only part of their utterance before speaking, leading to pauses, hesitations, and disfluencies. As a result, speakers do not need to engage in extensive advance planning. Thus, laboratory studies do not provide a full picture of language production in dialog, and further research using naturalistic tasks is needed.

Abstract

Most psycholinguistic studies on morphological processing have examined the role of morphological structure in the visual modality. This chapter discusses morphological processing in the auditory modality, which is an area of research that has only recently received more attention. It first discusses why results in the visual modality cannot straightforwardly be applied to the processing of spoken words, stressing the importance of acknowledging potential modality effects. It then gives a brief overview of the existing research on the role of morphology in the auditory modality, for which an increasing number of studies report that listeners show sensitivity to morphological structure. Finally, the chapter highlights insights gained by looking at morphological processing not only in reading, but also in listening, and it discusses directions for future research

Abstract

Assimilation of place of articulation across a nasal and a following stop consonant is obligatory in Japanese, but not in Dutch. In four experiments the processing of assimilated forms by speakers of Japanese and Dutch was compared, using a task in which listeners blended pseudo-word pairs such as ranga-serupa. An assimilated blend of this pair would be rampa, an unassimilated blend rangpa. Japanese listeners produced significantly more assimilated than unassimilated forms, both with pseudo-Japanese and pseudo-Dutch materials, while Dutch listeners produced significantly more unassimilated than assimilated forms in each materials set. This suggests that Japanese listeners, whose native-language phonology involves obligatory assimilation constraints, represent the assimilated nasals in nasal-stop sequences as unmarked for place of articulation, while Dutch listeners, who are accustomed to hearing unassimilated forms, represent the same nasal segments as marked for place of articulation.

Abstract

Languages contain tens of thousands of words, but these are constructed from a tiny handful of phonetic elements. Consequently, words resemble one another, or can be embedded within one another, a coup stick snot with standing. me process of spoken-word recognition by human listeners involves activation of multiple word candidates consistent with the input, and direct competition between activated candidate words. Further, human listeners are sensitive, at an early, prelexical, stage of speeeh processing, to constraints on what could potentially be a word of the language.

Abstract

The phoneme detection task is widely used in spoken word recognition research. Alphabetically literate participants, however, are more used to explicit representations of letters than of phonemes. The present study explored whether phoneme detection is sensitive to how target phonemes are, or may be, orthographically realised. Listeners detected the target sounds [b,m,t,f,s,k] in word-initial position in sequences of isolated English words. Response times were faster to the targets [b,m,t], which have consistent word-initial spelling, than to the targets [f,s,k], which are inconsistently spelled, but only when listeners’ attention was drawn to spelling by the presence in the experiment of many irregularly spelled fillers. Within the inconsistent targets [f,s,k], there was no significant difference between responses to targets in words with majority and minority spellings. We conclude that performance in the phoneme detection task is not necessarily sensitive to orthographic effects, but that salient orthographic manipulation can induce such sensitivity.

Abstract

Two lexical decision experiments in Cantonese are described in which the recognition of spoken target words as a function of phonological similarity to a preceding prime is investigated. Phonological similaritv in first syllables produced inhibition, while similarity in second syllables led to facilitation. Differences between syllables in tonal and segmental structure had generally similar effects.

Abstract

This chapter highlights that most language behavior consists of speaking and listening. The chapter also reveals differences and similarities between speaking and listening. The laboratory study of word production raises formidable problems; ensuring that a particular word is produced may subvert the spontaneous production process. Word production is investigated via slips and tip-of-the-tongue (TOT), primarily via instances of processing failure and via the technique of via the picture-naming task. The methodology of word production is explained in the chapter. The chapter also explains the phenomenon of interaction between various stages of word production and the process of speech recognition. In this context, it explores the difference between sound and meaning and examines whether or not the comparisons are appropriate between the processes of recognition and production of spoken words. It also describes the similarities and differences in the structure of the recognition and production systems. Finally, the chapter highlights the common issues in recognition and production research, which include the nuances of frequency of occurrence, morphological structure, and phonological structure.

Abstract

Different linguistic structures allow us to highlight distinct aspects of a situation. The aim of this survey is to investigate similarities and differences in the expression of situations or events as “stative” (maintaining a state), “inchoative” (adopting a state) and “agentive” (causing something to be in a state). The questionnaire focuses on the encoding of stative, inchoative and agentive possibilities for the translation equivalents of a set of English verbs.

Abstract

Expressions of human activities and states are a rich area for cross-linguistic comparison. Some languages of the world treat human posture verbs (e.g., sit, lie, kneel) as a special class of predicates, with distinct formal properties. This survey examines lexical, semantic and grammatical patterns for posture verbs, with special reference to contrasts between “stative” (maintaining a posture), “inchoative” (adopting a posture), and “agentive” (causing something to adopt a posture) constructions. The enquiry is thematically linked to the more general questionnaire 'Intransitive Predicate Form Class Survey'.

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

How do neural oscillations support human audiovisual language and communication? Considering the rhythmic nature of audiovisual language, in which stimuli from different sensory modalities unfold over time, neural oscillations represent an ideal candidate to investigate how audiovisual language is processed in the brain. Modulations of oscillatory phase and power are thought to support audiovisual language and communication in multiple ways. Neural oscillations synchronize by tracking external rhythmic stimuli or by re-setting their phase to presentation of relevant stimuli, resulting in perceptual benefits. In particular, synchronized neural oscillations have been shown to subserve the processing and the integration of auditory speech, visual speech, and hand gestures. Furthermore, synchronized oscillatory modulations have been studied and reported between brains during social interaction, suggesting that their contribution to audiovisual communication goes beyond the processing of single stimuli and applies to natural, face-to-face communication.

There are still some outstanding questions that need to be answered to reach a better understanding of the neural processes supporting audiovisual language and communication. In particular, it is not entirely clear yet how the multitude of signals encountered during audiovisual communication are combined into a coherent percept and how this is affected during real-world dyadic interactions. In order to address these outstanding questions, it is fundamental to consider language as a multimodal phenomenon, involving the processing of multiple stimuli unfolding at different rhythms over time, and to study language in its natural context: social interaction. Other outstanding questions could be addressed by implementing novel techniques (such as rapid invisible frequency tagging, dual-electroencephalography, or multi-brain stimulation) and analysis methods (e.g., using temporal response functions) to better understand the relationship between oscillatory dynamics and efficient audiovisual communication.

Abstract

This paper summarizes the results of a descriptive syntactic category analysis of child English no which reveals that young children use and represent no as a determiner and negatives like no pen as NPs, contra standard analyses.

Abstract

Acquiring a second language (L2) requires newly learned information to be integrated with existing knowledge. It has been proposed that several memory systems work together to enable this process of rapidly encoding new information and then slowly incorporating it with existing knowledge, such that it is consolidated and integrated into the language network without catastrophic interference. This chapter focuses on consolidation of L2 vocabulary. First, the complementary learning systems model is outlined, along with the model’s predictions regarding lexical consolidation. Next, word learning studies in first language (L1) that investigate the factors playing a role in consolidation, and the neural mechanisms underlying this, are reviewed. Using the L1 memory consolidation literature as background, the chapter then presents what is currently known about memory consolidation in L2 word learning. Finally, considering what is already known about L1 but not about L2, future research investigating memory consolidation in L2 neurocognition is proposed.

Abstract

The study presented here focuses on two pragmatic gestures:
the hand flip (Ferré, 2011), a gesture of the Palm Up Open
Hand/PUOH family (Müller, 2004) and the closed hand which
can be considered as the opposite kind of movement to the open-
ing of the hands present in the PUOH gesture. Whereas one of
the functions of the hand flip has been described as presenting
a new point in speech (Cienki, 2021), the closed hand gesture
has not yet been described in the literature to the best of our
knowledge. It can however be conceived of as having the oppo-
site function of announcing the end of a point in discourse. The
object of the present study is therefore to determine, with the
study of prosodic features, if the two gestures are found in the
same type of speech units and what their respective scope is.
Drawing from a corpus of three TED Talks in French the
prosodic characteristics of the speech that accompanies the two
gestures will be examined. The hypothesis developed in the
present paper is that their scope should be reflected in the
prosody of accompanying speech, especially pitch key, tone,
and relative pitch range. The prediction is that hand flips and
closing hand gestures are expected to be located at the periph-
ery of Intonation Phrases (IPs), Inter-Pausal Units (IPUs) or
more conversational Turn Constructional Units (TCUs), and are
likely to be co-occurrent with pauses in speech. But because of
the natural slope of intonation in speech, the speech that accom-
pany early gestures in Intonation Phrases should reveal different
features from the speech at the end of intonational units. Tones
should be different as well, considering the prosodic structure
of spoken French.

Abstract

Despite the early emergence of pointing, children are generally not documented to produce iconic gestures until later in development. Although research has described this developmental trajectory and the types of iconic gestures that emerge first, there has been limited focus on iconic gestures within interactional contexts. This study identified the first 10 iconic gestures produced by five monolingual English-speaking children in a naturalistic longitudinal video corpus and analysed the interactional contexts. We found children produced their first iconic gesture between 12 and 20 months and that gestural types varied. Although 34% of gestures could have been imitated or derived from adult or child actions in the preceding context, the majority were produced independently of any observed model. In these cases, adults often led the interaction in a direction where iconic gesture was an appropriate response. Overall, we find infants can represent a referent symbolically and possess a greater capacity for innovation than previously assumed. In order to develop our understanding of how children learn to produce iconic gestures, it is important to consider the immediate interactional context. Conducting naturalistic corpus analyses could be a more ecologically valid approach to understanding how children learn to produce iconic gestures in real life contexts.

Abstract

This paper compares the focus marking systems and the focus size that can be expressed by the different focus markings in four Mande and three Atlantic languages and varieties, namely: Bambara, Dyula, Kakabe, Soninke (Mande), Wolof, Jóola Foñy and Jóola Karon (Atlantic). All of these languages are known to mark focus morphosyntactically, rather than prosodically, as the more well-studied Germanic languages do. However, the Mande languages under discussion use only morphology, in the form of a particle that follows the focus, while the Atlantic ones use a more complex morphosyntactic system in which focus is marked by morphology in the verbal complex and movement of the focused term. It is shown that while there are some syntactic restrictions to how many different focus sizes can be marked in a distinct way, there is also a certain degree of arbitrariness as to which focus sizes are marked in the same way as each other.

Abstract

Sign language is a rich form of communication, uniquely conveying meaning through a combination of gestures, facial expressions, and body movements. Existing research in sign language generation has predominantly focused on text-to-sign pose generation, while speech-to-sign pose generation remains relatively underexplored. Speech-to-sign language generation models can facilitate effective communication between the deaf and hearing communities. In this paper, we propose an architecture that utilises prosodic information from speech audio and semantic context from text to generate sign pose sequences. In our approach, we adopt a multi-tasking strategy that involves an additional task of predicting Facial Action Units (FAUs). FAUs capture the intricate facial muscle movements that play a crucial role in conveying specific facial expressions during sign language generation. We train our models on an existing Indian Sign language dataset that contains sign language videos with audio and text translations. To evaluate our models, we report Dynamic Time Warping (DTW) and Probability of Correct Keypoints (PCK) scores. We find that combining prosody and text as input, along with incorporating facial action unit prediction as an additional task, outperforms previous models in both DTW and PCK scores. We also discuss the challenges and limitations of speech-to-sign pose generation models to encourage future research in this domain. We release our models, results and code to foster reproducibility and encourage future research1.

Abstract

Day-to-day interpretations of “space” are enmeshed in specific cultural and linguistic practices. For example, many cultures have an association between vertical height and social standing; more powerful people may be placed literally higher than others at social gatherings, and be spoken of as having higher status. This questionnaire is a guide for exploring relationships between space, language, and social structure. The goal is to better understand how space is organised in the focus community, and to investigate the extent to which space is used as a model for reproducing social forms.

Abstract

This task investigates the expression of “enter” and “exit” events, and is a supplement to the Motion Elicitation task (https://doi.org/10.17617/2.3003391). Consultants are asked to describe a series of animated clips where a man moves into or out of a house. The clips focus on contrasts to do with perspective (e.g., whether the man appears to move away or towards the viewer) and transitional movement (e.g., whether the man walks or “teleports” into his new location).

Abstract

The previous literature has suggested that the hand movement in co-speech gestures and signs consists of a series of phases with qualitatively different dynamic characteristics. In this paper, we propose a syntagmatic rule system for movement phases that applies to both co-speech gestures and signs. Descriptive criteria for the rule system were developed for the analysis video-recorded continuous production of signs and gesture. It involves segmenting a stream of body movement into phases and identifying different phase types. Two human coders used the criteria to analyze signs and cospeech gestures that are produced in natural discourse. It was found that the criteria yielded good inter-coder reliability. These criteria can be used for the technology of automatic recognition of signs and co-speech gestures in order to segment continuous production and identify the potentially meaningbearing phase.

Abstract

Do our hands 'speak the same language' across cultures? Gesture is the silent partner of spoken languages in face-to-face interaction, but we still have a lot to learn about gesture practices in different speech communities. The primary purpose of this task is to collect data in naturalistic settings that can be used to investigate the linguistic and cultural relativity of gesture performance, especially spatially indicative gestures. It involves video-recording pairs of speakers in both free conversation and more structured communication tasks (e.g., describing film plots).

Please note: the stimuli mentioned in this entry are available elsewhere: 'The Pear Story', a short film made at the University of California at Berkeley; "Frog, where are you?" from the original Mayer (1969) book, as published in the Appendix of Berman & Slobin (1994).

Abstract

Interpreting the meaning of co-speech gesture often involves
identifying a gesture’s ‘lexical affiliate’, the word or phrase to
which it most closely relates (Schegloff 1984). Though there is
work within gesture studies that resists this simplex mapping of
meaning from speech to gesture (e.g. de Ruiter 2000; Kendon
2014; Parrill 2008), including an evolving body of literature on
recurrent gesture and gesture families (e.g. Fricke et al. 2014; Müller 2017), it is still the lexical affiliate model that is most ap-
parent in formal linguistic models of multimodal meaning(e.g.
Alahverdzhieva et al. 2017; Lascarides and Stone 2009; Puste-
jovsky and Krishnaswamy 2021; Schlenker 2020). In this work,
I argue that the lexical affiliate should be carefully reconsidered
in the further development of such models.
In place of the lexical affiliate, I suggest a further shift
toward a frame-based, action schematic approach to gestural
meaning in line with that proposed in, for example, Parrill and
Sweetser (2004) and Müller (2017). To demonstrate the utility
of this approach I present three types of compositional gesture
sequences which I call spatial contrast, spatial embedding, and
cooperative abstract deixis. All three rely on gestural context,
rather than gesture-speech alignment, to convey interactive (i.e.
pragmatic) meaning. The centrality of gestural context to ges-
ture meaning in these examples demonstrates the necessity of
developing a model of gestural meaning independent of its in-
tegration with speech.

Abstract

It has been hypothesised that human reasoning has a non-linguistic foundation, but is nevertheless influenced by the formal means available in a language. For example, Western logic is transparently related to European sentential connectives (e.g., and, if … then, or, not), some of which cannot be unambiguously expressed in other languages. The questionnaire explores reasoning tools and practices through investigating translation equivalents of English sentential connectives and collecting examples of “reasoned arguments”.

Abstract

Wearing the hat of a cognitive anthropologist rather than an historian, I will try to amplify the ideas of Renn’s cited above. I argue that a particular subclass of material objects, namely “cognitive artifacts,” involves a close coupling of mind and artifact that acts like a brain prosthesis. Simple cognitive artifacts are external objects that act as aids to internal
computation, and not all cultures have extended inventories of these. Cognitive artifacts in this sense (e.g., calculating or measuring devices) have clearly played a central role in the history of science. But the notion can be widened to take in less material externalizations of cognition, like writing and language itself. A critical question here is how and why this close coupling of internal computation and external device actually works, a rather neglected question to which I’ll suggest some answers.

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.