Publications

Abstract

Word count estimation (WCE) from audio recordings has a number of applications, including quantifying the amount of speech that language-learning infants hear in their natural environments, as captured by daylong recordings made with devices worn by infants. To be applicable in a wide range of scenarios and also low-resource domains, WCE tools should be extremely robust against varying signal conditions and require minimal access to labeled training data in the target domain. For this purpose, earlier work has used automatic syllabification of speech, followed by a least-squares-mapping of syllables to word counts. This paper compares a number of previously proposed syllabifiers in the WCE task, including a supervised bi-directional long short-term memory (BLSTM) network that is trained on a language for which high quality syllable annotations are available (a “high resource language”), and reports how the alternative methods compare on different languages and signal conditions. We also explore additive noise and varying-channel data augmentation strategies for BLSTM training, and show how they improve performance in both matching and mismatching languages. Intriguingly, we also find that even though the BLSTM works on languages beyond its training data, the unsupervised algorithms can still outperform it in challenging signal conditions on novel languages.

Abstract

For almost two decades, the poor performance observed with the so-called Director task has been interpreted as evidence of limited use of Theory of Mind in communication. Here we propose a probabilistic model of common ground in referential communication that derives three inferences from an utterance: what the speaker is talking about in a visual context, what she knows about the context, and what referential expressions she prefers. We tested our model by comparing its inferences with those made by human participants and found that it closely mirrors their judgments, whereas an alternative model compromising the hearer’s expectations of cooperativeness and efficiency reveals a worse fit to the human data. Rather than assuming that common ground is fixed in a given exchange and may or may not constrain reference resolution, we show how common ground can be inferred as part of the process of reference assignment.

Abstract

In this paper, I will argue that although the study of multimodal interaction offers exciting new prospects for Human Computer Interaction and human-human communication research, language is the primary form of communication, even in multimodal systems. I will support this claim with theoretical and empirical arguments, mainly drawn from human-human communication research, and will discuss the implications for multimodal communication research and Human-Computer Interaction.

Abstract

While there are many studies on information retrieval models using full-text, there are presently no comparison studies of full-text retrieval vs. retrieval only over the titles of documents. On the one hand, the full-text of documents like scientific papers is not always available due to, e.g., copyright policies of academic publishers. On the other hand, conducting a search based on titles alone has strong limitations. Titles are short and therefore may not contain enough information to yield satisfactory search results. In this paper, we compare different retrieval models regarding their search performance on the full-text vs. only titles of documents. We use different datasets, including the three digital library datasets: EconBiz, IREON, and PubMed. The results show that it is possible to build effective title-based retrieval models that provide competitive results comparable to full-text retrieval. The difference between the average evaluation results of the best title-based retrieval models is only 3% less than those of the best full-text-based retrieval models.

Abstract

Most of the study of basic emotion expressions has focused on facial expressions and little work has been done to specifically investigate happiness, the only positive of the basic emotions (Ekman & Friesen, 1971). However, a theoretical suggestion has been made that happiness could be broken down into discrete positive emotions, which each fulfil the criteria of basic emotions, and that these would be expressed vocally (Ekman, 1992). To empirically test this hypothesis, 20 participants categorised 80 paralinguistic sounds using the labels achievement, amusement, contentment, pleasure and relief. The results suggest that achievement, amusement and relief are perceived as distinct categories, which subjects accurately identify. In contrast, the categories of contentment and pleasure were systematically confused with other responses, although performance was still well above chance levels. These findings are initial evidence that the positive emotions engage distinct vocal expressions and may be considered to be distinct emotion categories.

Abstract

Fine-Tracker is a speech-based model of human speech
recognition. While previous work has shown that Fine-Tracker
is successful at modelling aspects of human spoken-word
recognition, its speech recognition performance is not
comparable to that of human performance, possibly due to
suboptimal intermediate articulatory feature (AF)
representations. This study investigates the effect of improved
AF representations, obtained using a state-of-the-art deep
convolutional network, on Fine-Tracker’s simulation and
recognition performance: Although the improved AF quality
resulted in improved speech recognition; it, surprisingly, did
not lead to an improvement in Fine-Tracker’s simulation power.

Abstract

This work focuses on efficient use of the training material by selecting the optimal set of model topologies. We do this by training multiple word models of each word class, based on a subclassification according to a priori knowledge of the training material. We will examine classification criteria with respect to duration of the word, gender of the speaker, position of the word in the utterance, pauses in the vicinity of the word, and combinations of these. Comparative experiments were carried out on a corpus consisting of Dutch spoken connected digit strings and isolated digits, which are recorded in a wide variety of acoustic conditions. The results show, that classification based on gender of the speaker, position of the digit in the string, pauses in the vicinity of the training tokens, and models based on a combination of these criteria perform significantly better than the set with single models per digit.

Abstract

In this paper, we investigate the ability of SpeM, our recognition system based on the combination of an automatic phone recogniser and a wordsearch module, to determine as early as possible during the word recognition process whether a word is likely to be recognised correctly (this we refer to as ‘on-line’ early word recognition). We present two measures that can be used to predict whether a word is correctly recognised: the Bayesian word activation and the amount of available (acoustic) information for a word. SpeM was tested on 1,463 polysyllabic words in 885 continuous speech utterances. The investigated predictors indicated that a word activation that is 1) high (but not too high) and 2) based on more phones is more reliable to predict the correctness of a word than a similarly high value based on a small number of phones or a lower value of the word activation.

Abstract

We present DiscoGeM, a crowdsourced corpus of 6,505 implicit discourse relations from three genres: political speech,
literature, and encyclopedic texts. Each instance was annotated by 10 crowd workers. Various label aggregation methods
were explored to evaluate how to obtain a label that best captures the meaning inferred by the crowd annotators. The results
show that a significant proportion of discourse relations in DiscoGeM are ambiguous and can express multiple relation senses.
Probability distribution labels better capture these interpretations than single labels. Further, the results emphasize that text
genre crucially affects the distribution of discourse relations, suggesting that genre should be included as a factor in automatic
relation classification. We make available the newly created DiscoGeM corpus, as well as the dataset with all annotator-level
labels. Both the corpus and the dataset can facilitate a multitude of applications and research purposes, for example to
function as training data to improve the performance of automatic discourse relation parsers, as well as facilitate research into
non-connective signals of discourse relations.

Abstract

Previous studies have indicated that vocal and facial expressions of the ‘basic’ emotions share aspects of processing. Thus amygdala damage compromises the perception of fear and anger from the face and from the voice. In the current study we tested the hypothesis that there exist positive basic emotions, expressed mainly in the voice (Ekman, 1992). Vocal stimuli were produced to express the specific positive emotions of amusement, achievement, pleasure, contentment and relief.

Abstract

The present study examined two acoustic cues in the production
of lexical stress in Dutch: spectral tilt and overall intensity.
Sluijter and Van Heuven (1996) reported that spectral tilt is a
more reliable cue to stress than intensity. However, that study
included only a small number of talkers (10) and only syllables
with the vowels /aː/ and /ɔ/.
The present study re-examined this issue in a larger and
more variable dataset. We recorded 38 native speakers of Dutch
(20 females) producing 744 tokens of Dutch segmentally
overlapping words (e.g., VOORnaam vs. voorNAAM, “first
name” vs. “respectable”), targeting 10 different vowels, in
variable sentence contexts. For each syllable, we measured
overall intensity and spectral tilt following Sluijter and Van
Heuven (1996).
Results from Linear Discriminant Analyses showed that,
for the vowel /aː/ alone, spectral tilt showed an advantage over
intensity, as evidenced by higher stressed/unstressed syllable
classification accuracy scores for spectral tilt. However, when
all vowels were included in the analysis, the advantage
disappeared.
These findings confirm that spectral tilt plays a larger role
in signaling stress in Dutch /aː/ but show that, for a larger
sample of Dutch vowels, overall intensity and spectral tilt are
equally important.

Abstract

The results of an eye-tracking experiment are presented in which Dutch listeners' eye movements were monitored as they heard sentences and saw four pictured objects. Participants were instructed to click on the object mentioned in the sentence. In the critical sentences, a stop-initial target (e.g., "pot") was preceded by an [s], thus causing ambiguity regarding whether the sentence refers to a stop-initial or a cluster-initial word (e.g., "spot"). Participants made fewer fixations to the target pictures when the stop and the preceding [s] were cross-spliced from the cluster-initial word than when they were spliced from a different token of the sentence containing the stop-initial word. Acoustic analyses showed that the two versions differed in various measures, but only one of these - the duration of the [s] - correlated with the perceptual effect. Thus, in this context, the [s] duration information is an important factor guiding word recognition.

Abstract

We report an individual with music-odor synaesthesia who experiences automatic and vivid odor sensations when she hears music. S’s odor associations were recorded on two days, and compared with those of two control participants. Overall, S produced longer descriptions, and her associations were of multiple odors at once, in comparison to controls who typically reported a single odor. Although odor associations were qualitatively different between S and controls, ratings of the consistency of their descriptions did not differ. This demonstrates that crossmodal associations between music and odor exist in non-synaesthetes too. We also found that S is better at discriminating between odors than control participants, and is more likely to experience emotion, memories and evaluations triggered by odors, demonstrating the broader impact of her synaesthesia.

Abstract

This paper presents a quantitative analysis of the
turn-taking mechanism evidenced in 93 telephone
dialogues that were taken from the 9-million-word
Spoken Dutch Corpus. While the first part of the paper
focuses on the temporal phenomena of turn taking, such
as durations of pauses and overlaps of turns in the
dialogues, the second part explores the discoursefunctional
aspects of utterances in a subset of 8
dialogues that were annotated especially for this
purpose. The results show that speakers adapt their turntaking
behaviour to the interlocutor’s behaviour.
Furthermore, the results indicate that male-male dialogs
show a higher proportion of overlapping turns than
female-female dialogues.

Abstract

Sequences of reaction times (RT) produced by participants in an experiment are not only influenced by the stimuli, but by many other factors as well, including fatigue, attention, experience, IQ, handedness, etc. These confounding factors result in longterm effects (such as a participant’s overall reaction capability) and in short- and medium-time fluctuations in RTs (often referred to as ‘local speed effects’). Because stimuli are usually presented in a random sequence different for each participant, local speed effects affect the underlying ‘true’ RTs of specific trials in different ways across participants. To be able to focus statistical analysis on the effects of the cognitive process under study, it is necessary to reduce the effect of confounding factors as much as possible. In this paper we propose and compare techniques and criteria for doing so, with focus on reducing (‘filtering’) the local speed effects. We show that filtering matters substantially for the significance analyses of predictors in linear mixed effect regression models. The performance of filtering is assessed by the average between-participant correlation between filtered RT sequences and by Akaike’s Information Criterion, an important measure of the goodness-of-fit of linear mixed effect regression models.

Abstract

On the basis of two-speaker spontaneous conversations, it is shown that the distributions of both pauses and speech-overlaps of telephone and faceto-face dialogues have different statistical properties. Pauses in a face-to-face
dialogue last up to 4 times longer than pauses in telephone conversations in functionally comparable conditions. There is a high correlation (0.88 or larger) between the average pause duration for the two speakers across face-to-face
dialogues and telephone dialogues. The data provided form a first quantitative analysis of the complex turn-taking mechanism evidenced in the dialogues available in the 9-million-word Spoken Dutch Corpus.

Abstract

The recent advent of deep learning techniques in speech tech-nology and in particular in automatic speech recognition hasyielded substantial performance improvements. This suggeststhat deep neural networks (DNNs) are able to capture structurein speech data that older methods for acoustic modeling, suchas Gaussian Mixture Models and shallow neural networks failto uncover. In image recognition it is possible to link repre-sentations on the first couple of layers in DNNs to structuralproperties of images, and to representations on early layers inthe visual cortex. This raises the question whether it is possi-ble to accomplish a similar feat with representations on DNNlayers when processing speech input. In this paper we presentthree different experiments in which we attempt to untanglehow DNNs encode speech signals, and to relate these repre-sentations to phonetic knowledge, with the aim to advance con-ventional phonetic concepts and to choose the topology of aDNNs more efficiently. Two experiments investigate represen-tations formed by auto-encoders. A third experiment investi-gates representations on convolutional layers that treat speechspectrograms as if they were images. The results lay the basisfor future experiments with recursive networks.

Abstract

We estimate lexical Concreteness for millions of words across 77 languages. Using a simple regression framework, we combine vector-based models of lexical semantics with experimental norms of Concreteness in English and Dutch. By applying techniques to align vector-based semantics across distinct languages, we compute and release Concreteness estimates at scale in numerous languages for which experimental norms are not currently available. This paper lays out the technique and its efficacy. Although this is a difficult dataset to evaluate immediately, Concreteness estimates computed from English correlate with Dutch experimental norms at $\rho$ = .75 in the vocabulary at large, increasing to $\rho$ = .8 among Nouns. Our predictions also recapitulate attested relationships with word frequency. The approach we describe can be readily applied to numerous lexical measures beyond Concreteness

Abstract

Do all languages convey semantic knowledge in the same way? If language simply mirrors the structure of the world, the answer should be a qualified “yes”. If, however, languages impose structure as much as reflecting it, then even ostensibly the “same” word in different languages may mean quite different things. We provide a first pass at a large-scale quantification of cross-linguistic semantic alignment of approximately 1000 meanings in 55 languages. We find that the translation equivalents in some domains (e.g., Time, Quantity, and Kinship) exhibit high alignment across languages while the structure of other domains (e.g., Politics, Food, Emotions, and Animals) exhibits substantial cross-linguistic variability. Our measure of semantic alignment correlates with known phylogenetic distances between languages: more phylogenetically distant languages have less semantic alignment. We also find semantic alignment to correlate with cultural distances between societies speaking the languages, suggesting a rich co-adaptation of language and culture even in domains of experience that appear most constrained by the natural world

Abstract

In situated communication, reference to an entity in the shared visual context can be established using eitheranexpression that conveys precise (minimally specified) or redundant (over-specified) information. There is, however, along-lasting debate in psycholinguistics concerningwhether the latter hinders referential processing. We present evidence from an eyetrackingexperiment recordingfixations as well asthe Index of Cognitive Activity –a novel measure of cognitive workload –supporting the view that over-specifications facilitate processing. We further present originalevidence that, above and beyond the effect of specificity,referring expressions thatuniformly reduce referential entropyalso benefitprocessing

Abstract

Machine learning models of visual action recognition are typically trained and tested on data from specific situations where actions are associated with certain objects. It is an open question how action-object associations in the training set influence a model's ability to generalize beyond trained situations. We set out to identify properties of training data that lead to action recognition models with greater generalization ability. To do this, we take inspiration from a cognitive mechanism called cross-situational learning, which states that human learners extract the meaning of concepts by observing instances of the same concept across different situations. We perform controlled experiments with various types of action-object associations, and identify key properties of action-object co-occurrence in training data that lead to better classifiers. Given that these properties are missing in the datasets that are typically used to train action classifiers in the computer vision literature, our work provides useful insights on how we should best construct datasets for efficiently training for better generalization.

Abstract

The task of automatic playlist continuation is generating a list of recommended tracks that can be added to an existing playlist. By suggesting appropriate tracks, i. e., songs to add to a playlist, a recommender system can increase the user engagement by making playlist creation easier, as well as extending listening beyond the end of current playlist. The ACM Recommender Systems Challenge 2018 focuses on such task. Spotify released a dataset of playlists, which includes a large number of playlists and associated track listings. Given a set of playlists from which a number of tracks have been withheld, the goal is predicting the missing tracks in those playlists. We participated in the challenge as the team Unconscious Bias and, in this paper, we present our approach. We extend adversarial autoencoders to the problem of automatic playlist continuation. We show how multiple input modalities, such as the playlist titles as well as track titles, artists and albums, can be incorporated in the playlist continuation task.

Abstract

Before infants become mature speakers of their native language, they must acquire a robust word-recognition system which allows them to strike the balance between allowing some variation (mood, voice, accent) and recognizing variability that potentially changes meaning (e.g. cat vs hat). The current meta-analysis quantifies how the latter, termed mispronunciation sensitivity, changes over infants’ first three years, testing competing predictions of mainstream language acquisition theories. Our results show that infants were sensitive to mispronunciations, but accepted them as labels for target objects. Interestingly, and in contrast to predictions of mainstream theories, mispronunciation sensitivity was not modulated by infant age, suggesting that a sufficiently flexible understanding of native language phonology is in place at a young age.

Abstract

This study investigates the influence of both phonotactic and acoustic cues on the segmentation of spoken English. Listeners detected embedded English words in nonsense sequences (word spotting). Words aligned with phonotactic boundaries were easier to detect than words without such alignment. Acoustic cues to boundaries could also have signaled word boundaries, especially when word onsets lacked phonotactic alignment. However, only one of several durational boundary cues showed a marginally significant correlation with response times (RTs). The results suggest that word segmentation in English is influenced primarily by phonotactic constraints and only secondarily by acoustic aspects of the speech signal.

Abstract

Two eye-tracking experiments examined linguistic gender effects in non-native spoken-word recognition. French participants, who knew German well, followed spoken instructions in German to click on pictures on a computer screen (e.g., Wo befindet sich die Perle, “where is the pearl”) while their eye movements were monitored. The name of the target picture was preceded by a gender-marked article in the instructions. When a target and a competitor picture (with phonologically similar names) were of the same gender in both German and French, French participants fixated competitor pictures more than unrelated pictures. However, when target and competitor were of the same gender in German but of different gender in French, early fixations to the competitor picture were reduced. Competitor activation in the non-native language was seemingly constrained by native gender information. German listeners showed no such viewing time difference. The results speak against a form-based account of the linguistic gender effect. They rather support the notion that the effect originates from the grammatical level of language processing.

Abstract

Previous research has shown that listeners make use of their knowledge of phonotactic constraints to segment speech into individual words. The present study investigates the influence of phonotactics when segmenting a non-native language. German and English listeners detected embedded English words in nonsense sequences. German listeners also had knowledge of English, but English listeners had no knowledge of German. Word onsets were either aligned with a syllable boundary or not, according to the phonotactics of the two languages. Words aligned with either German or English phonotactic boundaries were easier for German listeners to detect than words without such alignment. Responses of English listeners were influenced primarily by English phonotactic alignment. The results suggest that both native and non-native phonotactic constraints influence lexical segmentation of a non-native, but familiar, language.

Abstract

In this paper, we present empirical data from a corpus study on the linear order of subjects and objects in German main clauses. The aim was to establish the validity of three well-known ordering constraints: given complements tend to occur before new complements, definite before indefinite, and pronoun before full noun phrase complements. Frequencies of occurrences were derived for subject-first and object-first sentences from the German Negra corpus. While all three constraints held on subject-first sentences, results for object-first sentences varied. Our findings suggest an influence of grammatical functions on the ordering of verb complements.

Abstract

Demonstratives—simple referential devices like this and that—are linguistic universals, but their meaning varies cross-linguistically. In languages like English and Italian, demonstratives are thought to encode the referent’s distance from the producer (e.g., that one means “the one far away from me”),
while in others, like Portuguese and Spanish, they encode relative distance from both producer and receiver (e.g., aquel means “the one far away from both of us”). Here we propose that demonstratives are also sensitive to the receiver’s focus of attention, hence requiring a deeper form of social cognition
than previously thought. We provide initial empirical and computational evidence for this idea, suggesting that producers use
demonstratives to redirect the receiver’s attention towards the intended referent, rather than only to indicate its physical distance.

Abstract

Picture book reading is a common word-learning context from which parents repeatedly name objects to their child and it has been found to facilitate early word learning. To learn the correct word-object mappings in a book-reading context, infants need to be able to link what they see with what they hear. However, given multiple objects on every book page, it is not clear how infants direct their attention to objects named by parents. The aim of the current study is to examine how infants mechanistically discover the correct word-object mappings during book reading in real time. We used head-mounted eye-tracking during parent-infant picture book reading and measured the infant's moment-by-moment visual attention to the named referent. We also examined how gesture cues provided by both the child and the parent may influence infants' attention to the named target. We found that although parents provided many object labels during book reading, infants were not able to attend to the named objects easily. However, their abilities to follow and use gestures to direct the other social partner’s attention increase the chance of looking at the named target during parent naming.