Publications

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

This paper presents an analysis of semantic association norms for German nouns. In contrast to prior studies, we not only collected associations elicited by written representations of target objects but also by their pictorial representations. In a first analysis, we identified systematic differences in the type and distribution of associate responses for the two presentation forms. In a second analysis, we applied a soft cluster analysis to the collected target-response pairs. We subsequently used the clustering to predict noun ambiguity and to discriminate senses in our target nouns.

Abstract

Many studies investigating the temporal alignment of co-speech
gestures to acoustic units in the speech signal find a close
coupling of the gestural landmarks and pitch accents or the
stressed syllable of pitch-accented words. In English, a pitch
accent is anchored in the lexically stressed syllable. Hence, it is
unclear whether it is the lexical phonological dimension of
stress, or the phrase-level prominence that determines the
details of speech-gesture synchronization. This paper explores
the relation between gestural phase transitions and accentual F0
peaks in Stockholm Swedish, which exhibits a lexical pitch
accent distinction. When produced with phrase-level
prominence, there are three different configurations of
lexicality of F0 peaks and the status of the syllable it is aligned
with. Through analyzing the alignment of the different F0 peaks
with gestural onsets in spontaneous dyadic conversations, we
aim to contribute to our understanding of the role of lexical
prosodic phonology in the co-production of speech and gesture.
The results, though limited by a small dataset, still suggest
differences between the three types of peaks concerning which
types of gesture phase onsets they tend to align with, and how
well these landmarks align with each other, although these
differences did not reach significance.

Abstract

This paper describes recent experimental evidence which shows that models of spoken word recognition must incorporate both inhibition between competing lexical candidates and a sensitivity to metrical cues to lexical segmentation. A new version of the Shortlist [1][2] model incorporating the Metrical Segmentation Strategy [3] provides a detailed simulation of the data.

Abstract

Phonetic convergence—i.e., adapting one’s speech
towards that of an interlocutor—has been shown
to occur in human-human conversations as well as
human-machine interactions. Here, we investigate
the hypothesis that human-to-robot convergence is
influenced by the human’s perception of the robot
and by the conversation’s topic. We conducted a
within-subjects experiment in which 33 participants
interacted with two robots differing in their eye gaze
behavior—one looked constantly at the participant;
the other produced gaze aversions, similarly to a
human’s behavior. Additionally, the robot asked
questions with increasing intimacy levels.
We observed that the speakers tended to converge
on F0 to the robots. However, this convergence
to the robots was not modulated by how the
speakers perceived them or by the topic’s intimacy.
Interestingly, speakers produced lower F0 means
when talking about more intimate topics. We
discuss these findings in terms of current theories of
conversational convergence.

Abstract

In this paper we describe a general scheme for internationalization of our talking head, Baldi, to speak other languages. We describe the modular structure of the auditory/visual synthesis software. As an example, we have created a synthetic Arabic talker, which is evaluated using a noisy word recognition task comparing this talker with a natural one.

Abstract

In this paper we try to contribute to the body of knowledge about the acquisition of English epistemic modal verbs (e.g. Mary may/has to be at school). Semantically, these verbs encode possibility or necessity with respect to available evidence. Pragmatically, the use of epistemic modals often gives rise to scalar conversational inferences (Mary may be at school -> Mary doesn’t have to be at school). The acquisition of epistemic modals is challenging for children on both these levels. In this paper, we present findings from two studies which were conducted with 5-year-old children and adults. Our findings, unlike previous work, show that 5-yr-olds have mastered epistemic modal semantics, including the notions of necessity and possibility. However, they are still in the process of acquiring epistemic modal pragmatics.

Abstract

En este artículo se desarrolla un método automático de segmentación de fonemas no supervisado. Este método utiliza el algoritmo de agrupación de máximo margen [1] para realizar segmentación de fonemas sobre habla continua sin necesidad de información a priori para el entrenamiento del sistema.

Abstract

Recent FMRI studies indicate that language related brain regions are engaged in artificial grammar (AG) processing. In the present study we investigate the Reber grammar by means of formal analysis and network simulations. We outline a new method for describing the network dynamics and propose an approach to grammar extraction based on the state-space dynamics of the network. We conclude that statistical frequency-based and rule-based acquisition procedures can be viewed as complementary perspectives on grammar learning, and more generally, that classical cognitive models can be viewed as a special case of a dynamical systems perspective on information processing

Abstract

This study investigates whether the acoustic durations of derivational affixes in Dutch are affected by the frequency of the word they occur in. In a word naming experiment, subjects were presented with a large number of words containing one of the affixes ge-, ver-, ont, or -lijk. Their responses were recorded on DAT tapes, and the durations of the affixes were measured using Automatic Speech Recognition technology. To investigate whether frequency also affected durations when speech rate was high, the presentation rate of the stimuli was varied. The results show that a higher frequency of the word as a whole led to shorter acoustic realizations for all affixes. Furthermore, affixes became shorter as the presentation rate of the stimuli increased. There was no interaction between word frequency and presentation rate, suggesting that the frequency effect also applies in situations in which the speed of articulation is very high.

Abstract

Children’s ability to share attention with another social partner (i.e., joint attention) has been found to support language development. Despite the large amount of research examining the effects of joint attention on language in hearing population, little is known about how deaf children learning sign languages achieve joint attention with their caregivers during natural social interaction and how caregivers provide and scaffold learning opportunities for their children. The present study investigates the properties and timing of joint attention surrounding familiar and novel naming events and their relationship to children’s vocabulary. Naturalistic play sessions of caretaker-child-dyads using American Sign Language were analyzed in regards to naming events of either familiar or novel object labeling events and the surrounding joint attention events. We observed that most naming events took place in the context of a successful joint attention event and that sign familiarity was related to the timing of naming events within the joint attention events. Our results suggest that caregivers are highly sensitive to their child’s visual attention in interactions and modulate joint attention differently in the context of naming events of familiar vs. novel object labels.

Abstract

A series of experiment tested Ekman’s (1992) hypothesis that there are a set of positive basic emotions that are expressed using vocal para-linguistic sounds, e.g. laughter and cheers. The proposed categories investigated were amusement, contentment, pleasure, relief and triumph. Behavioural testing using a forced-choice task indicated that participants were able to reliably recognize vocal expressions of the proposed emotions. A cross-cultural study in the preliterate Himba culture in Namibia confirmed that these categories are also recognized across cultures. A recognition test of acoustically manipulated emotional vocalizations established that the recognition of different emotions utilizes different vocal cues, and that these in turn differ from the cues used when comprehending speech. In a study using fMRI we found that relative to a signal correlated noise baseline, the paralinguistic expressions of emotion activated bilateral superior temporal gyri and sulci, lateral and anterior to primary auditory cortex, which is consistent with the processing of non linguistic vocal cues in the auditory ‘what’ pathway. Notably amusement was associated with greater activation extending into both temporal poles and amygdale and insular cortex. Overall, these results support the claim that ‘happiness’ can be fractionated into amusement, pleasure, relief and triumph.

Abstract

The ultimate goal of our research is to develop a computational model of human speech recognition that is able to capture the effects of fine-grained acoustic variation on speech recognition behaviour. As part of this work we are investigating automatic feature classifiers that are able to create reliable and accurate transcriptions of the articulatory behaviour encoded in the acoustic speech signal. In the experiments reported here, we compared support vector machines (SVMs) with multilayer perceptrons (MLPs). MLPs have been widely (and rather successfully) used for the task of multi-value articulatory feature classification, while (to the best of our knowledge) SVMs have not. This paper compares the performances of the two classifiers and analyses the results in order to better understand the articulatory representations. It was found that the MLPs outperformed the SVMs, but it is concluded that both classifiers exhibit similar behaviour in terms of patterns of errors.

Abstract

This paper addresses the issue of large-vocabulary recognition in a specific word class. We propose a two-pass strategy in which only major cities are explicitly represented in the first stage lexicon. An unknown word model encoded as a phone loop is used to detect OOV city names (referred to as rare city names). After which SpeM, a tool that can extract words and word-initial cohorts from phone graphs on the basis of a large fallback lexicon, provides an N-best list of promising city names on the basis of the phone sequences generated in the first stage. This N-best list is then inserted into the second stage lexicon for a subsequent recognition pass. Experiments were conducted on a set of spontaneous telephone-quality utterances each containing one rare city name. We tested the size of the N-best list and three types of language models (LMs). The experiments showed that SpeM was able to include nearly 85% of the correct city names into an N-best list of 3000 city names when a unigram LM, which also boosted the unigram scores of a city name in a given state, was used.

Abstract

We have recently developed a new model of human speech recognition, based on automatic speech recognition techniques [1]. The present paper has two goals. First, we show that the new model performs well in the recognition of lexically ambiguous input. These demonstrations suggest that the model is able to operate in the same optimal way as human listeners. Second, we discuss how to relate the behaviour of a recogniser, designed to discover the optimum path through a word lattice, to data from human listening experiments. We argue that this requires a metric that combines both path-based and word-based measures of recognition performance. The combined metric varies continuously as the input speech signal unfolds over time.

Abstract

In this paper, we illustrate the close parallels between the research fields of human speech recognition (HSR) and automatic speech recognition (ASR) using a computational model of human word recognition, SpeM, which was built using techniques from ASR. We show that ASR has proven to be useful for improving models of HSR by relieving them of some of their shortcomings. However, in order to build an integrated computational model of all aspects of HSR, a lot of issues remain to be resolved. In this process, ASR algorithms and techniques definitely can play an important role.

Abstract

In this paper, we present a novel computational model of human speech recognition – called SpeM – based on the theory underlying Shortlist. We will show that SpeM, in combination with an automatic phone recogniser (APR), is able to simulate the human speech recognition process from the acoustic signal to the ultimate recognition of words. This joint model takes an acoustic speech file as input and calculates the activation flows of candidate words on the basis of the degree of fit of the candidate words with the input. Experiments showed that SpeM outperforms Shortlist on the recognition of ‘real-life’ input. Furthermore, SpeM performs only slightly worse than an off-the-shelf full-blown automatic speech recogniser in which all words are equally probable, while it provides a transparent computationally elegant paradigm for modelling word activations in human word recognition.

Abstract

This study investigated the encoding of metrical information during speech production in Dutch. In Experiment 1, participants were asked to judge whether bisyllabic picture names had initial or final stress. Results showed significantly faster decision times for initially stressed targets (e.g., LEpel 'spoon') than for targets with final stress (e.g., liBEL 'dragon fly'; capital letters indicate stressed syllables) and revealed that the monitoring latencies are not a function of the picture naming or object recognition latencies to the same pictures. Experiments 2 and 3 replicated the outcome of the first experiment with bi- and trisyllabic picture names. These results demonstrate that metrical information of words is encoded rightward incrementally during phonological encoding in speech production. The results of these experiments are in line with Levelt's model of phonological encoding.

Abstract

This presentation will address aspects of the expression of emotion in non-verbal vocal behaviour, specifically attempting to determine the roles of both positive and negative emotions, their acoustic bases, and the extent to which these are recognized in non-Western cultures.

Abstract

This study investigated the impact of a speaker's gaze direction
on a listener's comprehension of discourse. Previous research
suggests that hand gestures play a role in referent allocation,
enabling listeners to better understand the discourse. The
current study aims to determine whether the speaker's gaze
direction has a similar effect on reference resolution as co-
speech gestures. Thirty native Japanese speakers participated in
the study and were assigned to one of three conditions:
congruent, incongruent, or speech-only. Participants watched
36 videos of an actor narrating a story consisting of three
sentences with two protagonists. The speaker consistently
used hand gestures to allocate one protagonist to the lower right
and the other to the lower left space, while directing her gaze to
either space of the target person (congruent), the other person
(incongruent), or no particular space (speech-only). Participants
were required to verbally answer a question about the target
protagonist involved in an accidental event as quickly as
possible. Results indicate that participants in the congruent
condition exhibited faster reaction times than those in the
incongruent condition, although the difference was not
significant. These findings suggest that the speaker's gaze
direction is not enough to facilitate a listener's comprehension
of discourse.

Abstract

Speech is perceived relative to the speech rate in the context. It is unclear, however, what information listeners use to compute speech rate. The present study examines whether listeners use the number of
syllables per unit time (i.e., syllable rate) as a measure of speech rate, as indexed by subsequent vowel perception. We ran two rate-normalization experiments in which participants heard duration-matched word lists that contained either monosyllabic
vs. bisyllabic words (Experiment 1), or monosyllabic vs. trisyllabic pseudowords (Experiment 2). The participants’ task was to categorize an /ɑ-aː/ continuum that followed the word lists. The monosyllabic condition was perceived as slower (i.e., fewer /aː/ responses) than the bisyllabic and
trisyllabic condition. However, no difference was observed between bisyllabic and trisyllabic contexts. Therefore, while syllable rate is used in perceiving speech rate, other factors, such as fast speech processes, mean F0, and intensity, must also influence rate normalization.

Abstract

Three experiments examined whether infants recognise functors in phrases, and whether their representations of functors are phonetically well specified. Eight- and 13- month-old English infants heard monosyllabic lexical words preceded by real functors (e.g., the, his) versus nonsense functors (e.g., kuh); the latter were minimally modified segmentally (but not prosodically) from real functors. Lexical words were constant across conditions; thus recognition of functors would appear as longer listening time to sequences with real functors. Eightmonth- olds' listening times to sequences with real versus nonsense functors did not significantly differ, suggesting that they did not recognise real functors, or functor representations lacked phonetic specification. However, 13-month-olds listened significantly longer to sequences with real functors. Thus, somewhere between 8 and 13 months of age infants learn familiar functors and represent them with segmental detail. We propose that accumulated frequency of functors in input in general passes a critical threshold during this time.

Abstract

This paper presents research findings on manual gestures
associated with negation in Indonesian, utilizing data sourced
from talk shows available on YouTube. The study reveals that
Indonesian speakers employ six recurrent negation gestures,
which have been observed in various languages worldwide.
This suggests that gestures exhibiting a stable form-meaning
relationship and recurring frequently in relation to negation are
prevalent around the globe, although their distribution may
differ across cultures and languages. Furthermore, the paper
demonstrates that negation gestures are not strictly tied to
verbal negation. Overall, the aim of this paper is to contribute
to a deeper understanding of the conventional usage and cross-
linguistic distribution of recurrent gestures.

Abstract

This study focuses on embodied uses of recognitional
demonstratives. While multimodal conversation analytic
studies have shown how gesture and speech interact in the
elaboration of exophoric references, little attention has been
given to the multimodal configuration of other types of
referential actions. Based on a video-recorded corpus of
professional meetings held in French, this qualitative study
shows that a subtype of deictic references, namely recognitional
references, are frequently associated with iconic gestures, thus
challenging the traditional distinction between exophoric and
endophoric uses of deixis.

Abstract

This paper describes ongoing research aiming at the description of variation in speech as represented by asynchronous articulatory features. We will first illustrate how distances in the articulatory feature space can be used for event detection along speech trajectories in this space. The temporal structure imposed by the cosine distance in articulatory feature space coincides to a large extent with the manual segmentation on phone level. The analysis also indicates that the articulatory feature representation provides better such alignments than the MFCC representation does. Secondly, we will present first results that indicate that articulatory features can be used to probe for acoustic differences in the onsets of Dutch singulars and plurals.

Abstract

This paper builds on previous work that aims at unraveling the structure of the speech signal by means of using probabilistic representations. The context of this work is a multi-pass speech recognition system in which a phone lattice is created and used as a basis for a lexical search in which symbolic mismatches are allowed at certain costs. The focus is on the optimization of the costs of phone insertions, deletions and substitutions that are used in the lexical decoding pass. Two optimization approaches are presented, one related to a multi-pass computational model for human speech recognition, the other based on a decoding in which Bayes’ risks are minimized. In the final section, the advantages of these optimization methods are discussed and compared.

Abstract

This paper investigates the interaction between acoustic scores and symbolic mismatch penalties in multi-pass speech decoding techniques that are based on the creation of a segment graph followed by a lexical search. The interaction between acoustic and symbolic mismatches determines to a large extent the structure of the search space of these multipass approaches. The background of this study is a recently developed computational model of human word recognition, called SpeM. SpeM is able to simulate human word recognition data and is built as a multi-pass speech decoder. Here, we focus on unravelling the structure of the search space that is used in SpeM and similar decoding strategies. Finally, we elaborate on the close relation between distances in this search space, and distance measures in search spaces that are based on a combination of acoustic and phonetic features.

Abstract

This paper presents a new interdisciplinary methodology for the
analysis of future conceptualisations in big messy media data.
More specifically, it focuses on the depictions of post-Covid
futures by RT during the pandemic, i.e. on data which are of
interest not just from the perspective of academic research but
also of policy engagement. The methodology has been
developed to support the scaling up of fine-grained data-driven
analysis of discourse utterances larger than individual lexical
units which are centred around ‘will’ + the infinitive. It relies
on the true integration of manual analytical and computational
methods and tools in researching three modalities – textual,
prosodic1, and gestural. The paper describes the process of
building a computational infrastructure for the collection and
processing of video data, which aims to empower the manual
analysis. It also shows how manual analysis can motivate the
development of computational tools. The paper presents
individual computational tools to demonstrate how the
combination of human and machine approaches to analysis can
reveal new manifestations of cohesion between gesture and
prosody. To illustrate the latter, the paper shows how the
boundaries of prosodic units can work to help determine the
boundaries of gestural units for future conceptualisations.

Abstract

Over the course of a conversation, interlocutors sound more and more like each other in a process called convergence. However, the automaticity and grain size of convergence are not well established. This study therefore examined whether female native Dutch speakers converge to large yet sub-phonemic shifts in the F2 of the vowel /e/. Participants first performed a short reading task to establish baseline F2s for the vowel /e/, then shadowed 120 target words (alongside 360 fillers) which contained one instance of a manipulated vowel /e/ where the F2 had been shifted down to that of the vowel /ø/. Consistent exposure to large (sub-phonemic) downward shifts in F2 did not result in convergence. The results raise issues for theories which view convergence as a product of automatic integration between perception and production.

Abstract

This paper reports two experiments with vowels and consonants as phoneme detection targets in real words. In the first experiment, two relatively distinct vowels were compared with two confusible stop consonants. Response times to the vowels were longer than to the consonants. Response times correlated negatively with target phoneme length. In the second, two relatively distinct vowels were compared with their corresponding semivowels. This time, the vowels were detected faster than the semivowels. We conclude that response time differences between vowels and stop consonants in this task may reflect differences between phoneme categories in the variability of tokens, both in the acoustic realisation of targets and in the' representation of targets by subjects.

Abstract

Linguistic context vectors are adapted for measuring the linguistic contexts that accompany gestures and comparable co-linguistic behaviours. Focusing on gestural semiotic types, it is demonstrated that gestural linguistic context vectors carry information associated with gesture. It is suggested that these may be used to approximate gesture meaning in a similar manner to the approximation of word meaning by context vectors.

Abstract

This study investigated the perception of American English phonemes by native listeners. Listeners identified either the consonant or the vowel in all possible English CV and VC syllables. The syllables were embedded in multispeaker babble at three signal-to-noise ratios (0 dB, 8 dB, and 16 dB). Effects of syllable position, signal-to-noise ratio, and articulatory features on vowel and consonant identification are discussed. The results constitute the largest source of data that is currently available on phoneme confusion patterns of American English phonemes by native listeners.

Abstract

This study investigated the perception of American English phonemes by native listeners. Listeners identified either the consonant or the vowel in all possible English CV and VC syllables. The syllables were embedded in multispeaker babble at three signalto-noise ratios (0 dB, 8 dB, and 16 dB). Effects of syllable position, signal-to-noise ratio, and articulatory features on vowel and consonant identification are discussed. The results constitute the largest source of data that is currently available on phoneme confusion patterns of American English phonemes by native listeners.

Abstract

It has been suggested that individual variation in vowel compactness of the native language (L1) and the distance between L1 vowels and vowels in the second language (L2) predict successful L2 vowel acquisition. Moreover, general articulatory skills have been proposed to account for variation in vowel compactness. In the present work, we conceptually replicate a previous study to test these hypotheses with a large sample size, a new language pair and a
new vowel pair. We find evidence that individual variation in L1 vowel compactness has opposing effects for two different vowels. We do not find evidence that individual variation in L1 compactness
is explained by general articulatory skills. We conclude that the results found previously might be specific to sub-groups of L2 learners and/or specific sub-sets of vowel pairs.

Abstract

Utilization of computer tools in linguistic research has gained importance with the maturation of media frameworks for the handling of digital audio and video. The increased use of these tools in gesture, sign language and multimodal interaction studies has led to stronger requirements on the flexibility, the efficiency and in particular the time accuracy of annotation tools. This paper describes the efforts made to make ELAN a tool that meets these requirements, with special attention to the developments in the area of time accuracy. In subsequent sections an overview will be given of other enhancements in the latest versions of ELAN, that make it a useful tool in multimodality research.

Abstract

A number of serious reasons will convince an increasing amount of researchers to store their relevant material in centers which we will call "language resource archives". They combine the duty of taking care of long-term preservation as well as the task to give access to their material to different user groups. Access here is meant in the sense that an active interaction with the data will be made possible to support the integration of new data, new versions or commentaries of all sort. Modern Language Resource Archives will have to adhere to a number of basic principles to fulfill all requirements and they will have to be involved in federations to create joint language resource domains making it even more simple for the researchers to access the data. This paper makes an attempt to formulate the essential pillars language resource archives have to adhere to.