Publications

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

This paper describes a computational model of speech comprehension that takes the acoustic signal as input and predicts reaction times as observed in an auditory lexical decision task. By doing so, we explore a new generation of end-to-end computational models that are able to simulate the behaviour of human subjects participating in a psycholinguistic experiment. So far, nearly all computational models of speech comprehension do not start from the speech signal itself, but from abstract representations of the speech signal, while the few existing models that do start from the acoustic signal cannot directly model reaction times as obtained in comprehension experiments. The main functional components in our model are the perception stage, which is compatible with the psycholinguistic model Shortlist B and is implemented with techniques from automatic speech recognition, and the decision stage, which is based on the linear ballistic accumulation decision model. We successfully tested our model against data from 20 participants performing a largescale auditory lexical decision experiment. Analyses show that the model is a good predictor for the average judgment and reaction time for each word.

Abstract

In everyday life, people see, describe and remember motion events. We tested whether the type of motion event information (path or manner) encoded in speech and gesture predicts which information is remembered and if this varies across speakers of typologically different languages. We focus on intransitive motion events (e.g., a woman running to a tree) that are described differently in speech and co-speech gesture across languages, based on how these languages typologically encode manner and path information (Kita & Özyürek, 2003; Talmy, 1985). Speakers of Dutch (n = 19) and Turkish (n = 22) watched and described motion events. With a surprise (i.e. unexpected) recognition memory task, memory for manner and path components of these events was measured. Neither Dutch nor Turkish speakers’ memory for manner went above chance levels. However, we found a positive relation between path speech and path change detection: participants who described the path during encoding were more accurate at detecting changes to the path of an event during the memory task. In addition, the relation between path speech and path memory changed with native language: for Dutch speakers encoding path in speech was related to improved path memory, but for Turkish speakers no such relation existed. For both languages, co-speech gesture did not predict memory speakers. We discuss the implications of these findings for our understanding of the relations between speech, gesture, type of encoding in language and memory.

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

We are often in a bilingual situation (e.g., overhearing a conversation in the train). We investigated whether first (L1) and second language (L2) phonologies are automatically activated. A masked priming paradigm was used, with Russian words as targets and either Russian or English words as primes. Event-related potentials (ERPs) were recorded while Russian (L1) – English (L2) bilinguals read aloud L1 target words (e.g. РЕЙС /reis/ ‘fl ight’) primed with either L1 (e.g. РАНА /rana/ ‘wound’) or L2 words (e.g. PACK). Target words were read faster when they were preceded by phonologically related L1 primes but not by orthographically related L2 primes. ERPs showed orthographic priming in the 125-200 ms time window. Thus, both L1 and L2 phonologies are simultaneously activated during L1 reading. The results provide support for non-selective models of bilingual reading, which assume automatic activation of the non-target language phonology even when it is not required by the task.

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

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

Sign Language of the Netherlands (NGT) is a visual-gestural language in which linguistic information is conveyed through manual as well as non-manual channels; not only the hands, but also body position, head position and facial expression are important for the language structure. Facial expressions serve grammatical functions in the marking of topics, yes/no questions, and wh-questions (Coerts, 1992). Furthermore, facial expression is used nonlinguistically in the expression of affect (Ekman, 1979). Consequently, at the phonetic level obligatory marking of grammar using facial expression may conflict with the expression of affect. In this study, I investigated the interplay of linguistic and affective functions of brow movements in NGT. Three hypotheses were tested in this thesis. The first is that the affective markers of eyebrows would dominate over the linguistic markers. The second hypothesis predicts that the grammatical markers dominate over the affective brow movements. A third possibility is that a Phonetic Sum would occur in which both functions are combined simultaneously. I elicited sentences combining grammatical and affective functions of eyebrows using a randomised design. Five sentence types were included: declarative sentences, topic sentences, yes-no questions, wh-questions with the wh-sign sentence-final and wh-questions with the wh-sign sentence-initial. These sentences were combined with neutral, surprised, angry, and distressed affect. The brow movements were analysed using the Facial Action Coding System (Ekman, Friesen, & Hager, 2002a). In these sentences, the eyebrows serve a linguistic function, an affective function, or both. One of the possibilities in the latter cases was that a Phonetic Sum would occur that combines both functions simultaneously. Surprisingly, it was found that a Phonetic Sum occurs in which the phonetic weight of Action Unit 4 appears to play an important role. The results show that affect displays may alter question signals in NGT.

Abstract

This study concerns individual differences in speech imitation ability and the role that lexical representations play in imitation. We examined 1) whether imitation of sounds in an unfamiliar language (L0) is related to imitation of sounds in an unfamiliar
non-native accent in the speaker’s native language (L1) and 2) whether it is easier or harder to imitate speech when you know the words to be imitated. Fifty-nine native Dutch speakers imitated words with target vowels in Basque (/a/ and /e/) and Greekaccented
Dutch (/i/ and /u/). Spectral and durational
analyses of the target vowels revealed no relationship between the success of L0 and L1 imitation and no difference in performance between tasks (i.e., L1
imitation was neither aided nor blocked by lexical knowledge about the correct pronunciation). The results suggest instead that the relationship of the vowels to native phonological categories plays a bigger role in imitation

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.

Abstract

It is unknown whether modality affects the efficiency with which humans learn novel word forms and their meanings, with previous studies reporting both written and auditory advantages. The current study implements controls whose absence in previous work likely offers explanation for such contradictory findings. In two novel word learning experiments, participants were trained and tested on pseudoword - novel object pairs, with controls on: modality of test, modality of meaning, duration of exposure and transparency of word form. In both experiments word forms were presented in either their written or spoken form, each paired with a pictorial meaning (novel object). Following a 20-minute filler task, participants were tested on their ability to identify the picture-word form pairs on which they were trained. A between subjects design generated four participant groups per experiment 1) written training, written test; 2) written training, spoken test; 3) spoken training, written test; 4) spoken training, spoken test. In Experiment 1 the written stimulus was presented for a time period equal to the duration of the spoken form. Results showed that when the duration of exposure was equal, participants displayed a written training benefit. Given words can be read faster than the time taken for the spoken form to unfold, in Experiment 2 the written form was presented for 300 ms, sufficient time to read the word yet 65% shorter than the duration of the spoken form. No modality effect was observed under these conditions, when exposure to the word form was equivalent. These results demonstrate, at least for proficient readers, that when exposure to the word form is controlled across modalities the efficiency with which word form-meaning associations are learnt does not differ. Our results therefore suggest that, although we typically begin as aural-only word learners, we ultimately converge on developing learning mechanisms that learn equally efficiently from both written and spoken materials.

Abstract

Psycholinguistic research has typically portrayed speech production as a relatively automatic process. This is because when errors are made, they occur as seldom as one in every thousand words we utter. However, it has long been recognised that we need some form of control over what we are currently saying and what we plan to say. This capacity to both monitor our inner speech and self-correct our speech output has often been assumed to be a property of the language comprehension system. More recently, it has been demonstrated that speech production benefits from interfacing with more general cognitive processes such as selective attention, short-term memory (STM) and online response monitoring to resolve potential conflict and successfully produce the output of a verbal plan. The conditions and levels of representation according to which these more general planning, monitoring and control processes are engaged during speech production remain poorly understood. Moreover, there remains a paucity of information about their neural substrates, despite some of the first evidence of more general monitoring having come from electrophysiological studies of error related negativities (ERNs). While aphasic speech errors continue to be a rich source of information, there has been comparatively little research focus on instances of speech repair. The purpose of this Frontiers Research Topic is to provide a forum for researchers to contribute investigations employing behavioural, neuropsychological, electrophysiological, neuroimaging and virtual lesioning techniques. In addition, while the focus of the research topic is on novel findings, we welcome submission of computational simulations, review articles and methods papers.