Publications

Abstract

Humans learn language by interaction with their environment and listening to other humans. It should also be possible for computational models to learn language directly from speech but so far most approaches require text. We improve on existing neural network approaches to create visually grounded embeddings for spoken utterances. Using a combination of a multi-layer GRU, importance sampling, cyclic learning rates, ensembling and vectorial self-attention our results show a remarkable increase in image-caption retrieval performance over previous work. Furthermore, we investigate which layers in the model learn to recognise words in the input. We find that deeper network layers are better at encoding word presence, although the final layer has slightly lower performance. This shows that our visually grounded sentence encoder learns to recognise words from the input even though it is not explicitly trained for word recognition.

Abstract

The anterior lordosis of the cervical spine is thought
to contribute to pitch (fo) production by influencing
cricoid rotation as a function of larynx height. This
study examines the matter of inter-individual
variation in cervical spine shape and whether this has
an influence on how fo is produced along increasing
or decreasing scales, using the ArtiVarK dataset,
which contains real-time MRI pitch production data.
We find that the cervical spine actively participates in
fo production, but the amount of displacement
depends on individual shape. In general, anterior
spine motion (tending toward cervical lordosis)
occurs for low fo, while posterior movement (tending
towards cervical kyphosis) occurs for high fo.

Abstract

Effekte linguisticher Typologie auf die Sprach-entwicklung. In G. Plehn (Ed.), Jahrbuch der Max-Planck Gesellsch

Abstract

In order to study the cognitive processes underlying speech comprehension, neuro-physiological measures (e.g., EEG and MEG), or behavioural measures (e.g., reaction times and response accuracy) can be applied. Compared to behavioural measures, EEG signals can provide a more fine-grained and complementary view of the processes that take place during the unfolding of an auditory stimulus.

EEG signals are often analysed after having chosen specific time windows, which are usually based on the temporal structure of ERP components expected to be sensitive to the experimental manipulation. However, as the timing of ERP components may vary between experiments, trials, and participants, such a-priori defined analysis time windows may significantly hamper the exploratory power of the analysis of components of interest. In this paper, we explore a wide-window analysis method applied to EEG signals collected in an auditory repetition priming experiment.

This approach is based on a bank of temporal filters arranged along the time axis in combination with linear mixed effects modelling. Crucially, it permits a temporal decomposition of effects in a single comprehensive statistical model which captures the entire EEG trace.

Abstract

The human language processor is conceived as a system that consists of several interrelated subsystems. Each subsystem performs a specific task in the complex process of language comprehension and production. A subsystem receives a particular input, performs certain specific operations on this input and yields a particular output. The subsystems can be characterized in terms of the transformations that relate the input representations to the output representations. An important issue in describing the language processing system is to identify the subsystems and to specify the relations between the subsystems. These relations can be conceived in two different ways. In one conception the subsystems are autonomous. They are related to each other only by the input-output channels. The operations in one subsystem are not affected by another system. The subsystems are modular, that is they are independent. In the other conception, the different subsystems influence each other. A subsystem affects the processes in another subsystem. In this conception there is an interaction between the subsystems.

Abstract

Two word-spotting experiments are reported that examine whether the Possible-Word Constraint (PWC) [1] is a language-specific or language-universal strategy for the segmentation of continuous speech. The PWC disfavours parses which leave an impossible residue between the end of a candidate word and a known boundary. The experiments examined cases where the residue was either a CV syllable with a lax vowel, or a CVC syllable with a schwa. Although neither syllable context is a possible word in English, word-spotting in both contexts was easier than with a context consisting of a single consonant. The PWC appears to be language-universal rather than language-specific.

Abstract

Simulations explored the inability of the TRACE model of spoken-word recognition to model the effects on human listening of subcategorical mismatch in word forms. The source of TRACE's failure lay not in interactive connectivity, not in the presence of inter-word competition, and not in the use of phonemic representations, but in the need for continuously optimised interpretation of the input. When an analogue of TRACE was allowed to cycle to asymptote on every slice of input, an acceptable simulation of the subcategorical mismatch data was achieved. Even then, however, the simulation was not as close as that produced by the Merge model, which has inter-word competition, phonemic representations and continuous optimisation (but no interactive connectivity).

Abstract

Our bodily experiences play an important role in the way that we think and speak. Abstract language is, however, difficult to reconcile with this body-centred view, unless we appreciate the role metaphors play. To explore the role of the senses across semantic domains, we focus on perception metaphors, and examine their realisation across diverse languages, methods, and approaches. To what extent do mappings in perception metaphor adhere to predictions based on our biological propensities; and to what extent is there space for cross-linguistic and cross-cultural variation? We find that while some metaphors have widespread commonality, there is more diversity attested than should be comfortable for universalist accounts.

Abstract

A database is presented of relative familiarity ratings for 24 sets of Japanese words, each set comprising words overlapping in the initial portions. These ratings are useful for the generation of material sets for research in the recognition of spoken words.

Abstract

-* indicates shared first authorship -
The present study investigated the effects of selective attention on the processing of morphosyntactic errors in unattended parts of speech. Two groups of German native (L1) speakers participated in the present study. Participants listened to sentences in which irregular verbs were manipulated in three different conditions (correct, incorrect but attested ablaut pattern, incorrect and crosslinguistically unattested ablaut pattern). In order to track fast dynamic neural reactions to the stimuli, electroencephalography was used. After each sentence, participants in Experiment 1 performed a semantic judgement task, which deliberately distracted the participants from the syntactic manipulations and directed their attention to the semantic content of the sentence. In Experiment 2, participants carried out a syntactic judgement task, which put their attention on the critical stimuli. The use of two different attentional tasks allowed for investigating the impact of selective attention on speech processing and whether morphosyntactic processing steps are performed automatically. In Experiment 2, the incorrect attested condition elicited a larger N400 component compared to the correct condition, whereas in Experiment 1 no differences between conditions were found. These results suggest that the processing of morphosyntactic violations in irregular verbs is not entirely automatic but seems to be strongly affected by selective attention.

Abstract

Language production involves the retrieval of information from memory, the planning of an articulatory program, and executive control and self-monitoring. These processes can be related to the domains of long-term memory, motor control, and executive control. Here, we argue that studying neuronal oscillations provides an important opportunity to understand how general neuronal computational principles support language production, also helping elucidate relationships between language and other domains of cognition. For each relevant domain, we provide a brief review of the findings in the literature with respect to neuronal oscillations. Then, we show how similar patterns are found in the domain of language production, both through review of previous literature and novel findings. We conclude that neurophysiological mechanisms, as reflected in modulations of neuronal oscillations, may act as a fundamental basis for bringing together and enriching the fields of language and cognition.

Abstract

Spontaneously occurring speech is often seamlessly accompanied by hand gestures. Detailed
observations of video data suggest that speech and gesture are tightly synchronized in time,
consistent with a dynamic interplay between body and mind. However, spontaneous gesturespeech
synchrony has rarely been objectively quantified beyond analyses of video data, which
do not allow for identification of kinematic properties of gestures. Consequently, the point in
gesture which is held to couple with speech, the so-called moment of “maximum effort”, has
been variably equated with the peak velocity, peak acceleration, peak deceleration, or the onset
of the gesture. In the current exploratory report, we provide novel evidence from motiontracking
and acoustic data that peak velocity is closely aligned, and shortly leads, the peak pitch
(F0) of speech

Abstract

Thematic roles such as Agent and Instrument have a long-standing place in theories of event representation. Nonetheless, the structure of these categories has been difficult to determine. We investigated how instrumental events, such as someone slicing bread with a knife, are categorized in English. Speakers described a variety of typical and atypical instrumental events, and we determined the similarity structure of their descriptions using correspondence analysis. We found that events where the instrument is an extension of an intentional agent were most likely to elicit similar language, highlighting the importance of agency in structuring instrumental categories.

Abstract

Although humans are the ultimate “natural language generators”, the area of psycholinguistic modeling has been somewhat underrepresented in recent approaches to Natural Language Generation in computer science. To draw attention to the area and illustrate its potential relevance to Natural Language Generation, I provide an overview of recent work on psycholinguistic modeling of language production together with some key empirical findings, state-of-the-art experimental techniques, and their historical roots. The techniques include analyses of speech-error corpora, chronometric analyses, eyetracking, and neuroimaging.
The overview is built around the issue of cognitive control in natural language generation, concentrating on the production of single words, which is an essential ingredient of the generation of larger utterances. Most of the work exploited the fact that human speakers are good but not perfect at resisting temptation, which has provided some critical clues about the nature of the underlying system.

Abstract

This paper deals with the Shawi nominalizing suffixes -su’~-ru’~-nu’ ‘general nominalizer’, -napi/-te’/-tun‘performer/agent nominalizer’, -pi’‘patient nominalizer’, and -nan ‘instrument nominalizer’. The goal of this article is to provide a description of nominalization in Shawi. Throughout this paper I apply the Generalized Scale Model (GSM) (Malchukov, 2006) to Shawi verbal nominalizations, with the intention of presenting a formal representation that will provide a basis for future areal and typological studies of nominalization. In addition, I dialogue with Shibatani’s model to see how the loss or gain of categories correlates with the lexical or grammatical nature of nominalizations. strong nominalization in Shawi correlates with lexical nominalization, whereas weak nominalizations correlate with grammatical nominalization. A typology which takes into account the productivity of the nominalizers is also discussed.

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

Listeners’ signals of recipiency, such as “Mm-hm” or “uh-huh” in English, are the most elementary or minimal “conversational turns” possible. Minimal, because apart from acknowledging recipiency and inviting the speaker to continue with his/her next turn, they do not add any new information to the discourse of the conversation. The goal of this project is to gather cross cultural information on listeners’ feedback behaviour during conversation. Listeners in a conversation usually provide short signals that indicate to the speaker that they are still “with the speaker”. These signals could be verbal (like for instance “mm hm” in English or “hm hm” in Dutch) or nonverbal (visual), like nodding. Often, these signals are produced in overlap with the speaker’s vocalisation. If listeners do not produce these signals, speakers often invite them explicitly (e.g. “are you still there?” in a telephone conversation). Our goal is to investigate what kind of signals are used by listeners of different languages to signal “recipiency” to the speaker.

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

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

The current version of the psycholinguistic model of human word recognition Shortlist suffers from two unrealistic constraints. First, the input of Shortlist must consist of a single string of phoneme symbols. Second, the current version of the search in Shortlist makes it difficult to deal with insertions and deletions in the input phoneme string. This research attempts to fully automatically derive a phoneme string from the acoustic signal that is as close as possible to the number of phonemes in the lexical representation of the word. We optimised an Automatic Phone Recogniser (APR) using two approaches, viz. varying the value of the mismatch parameter and optimising the APR output strings on the output of Shortlist. The approaches show that it will be very difficult to satisfy the input requirements of the present version of Shortlist with a phoneme string generated by an APR.

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

Due to pronunciation variation, many insertions and deletions of phones occur in spontaneous speech. The psycholinguistic model of human speech recognition Shortlist is not well able to deal with phone insertions and deletions and is therefore not well suited for dealing with real-life input. The research presented in this paper explains how Shortlist can benefit from pronunciation variation modelling in dealing with real-life input. Pronunciation variation was modelled by including variants into the lexicon of Shortlist. A series of experiments was carried out to find the optimal acoustic model set for transcribing the training material that was used as basis for the generation of the variants. The Shortlist experiments clearly showed that Shortlist benefits from pronunciation variation modelling. However, the performance of Shortlist stays far behind the performance of other, more conventional speech recognisers.

Abstract

The time course of metrical encoding, i.e. stress, during speech production is investigated. In a first experiment, participants were presented with pictures whose bisyllabic Dutch names had initial or final stress (KAno 'canoe' vs. kaNON 'cannon'; capital letters indicate stressed syllables). Picture names were matched for frequency and object recognition latencies. When participants were asked to judge whether picture names had stress on the first or second syllable, they showed significantly faster decision times for initially stressed targets than for targets with final stress. Experiment 2 replicated this effect with trisyllabic picture names (faster RTs for penultimate stress than for ultimate stress). In our view, these results reflect the incremental phonological encoding process. Wheeldon and Levelt (1995) found that segmental encoding is a process running from the beginning to the end of words. Here, we present evidence that the metrical pattern of words, i.e. stress, is also encoded incrementally.

Abstract

The representative corpus SYN2000 in the Czech National Corpus (CNK) project containing 100 million word forms taken from different types of texts. I have tried to determine the extent and depth of the linguistic material in the corpus. First, I chose the adjectives indicating the basic colors of the spectrum and other parts of speech (names and adverbs) derived from these adjectives. An analysis of three examples - black, white and red - shows the extent of the linguistic wealth and diversity we are looking at: because of size limitations, no existing dictionary is capable of embracing all analyzed nuances. Currently, we can only hope that the next dictionary of contemporary Czech, built on the basis of the Czech National Corpus, will be electronic. Without the size limitations, we would be able us to include many of the fine nuances of language

Abstract

This paper addresses the role of the adverb in Dutch direct object scrambling constructions. We report four experiments in which we investigate whether the structural position and the scope sensitivity of the adverb affect acceptability judgments of scrambling constructions and native speakers' tendency to scramble definite objects. We conclude that the type of adverb plays a key role in Dutch word ordering preferences.

Abstract

We tested whether statistical averageness (SA) at the level of the individual speaker could predict a speaker’s intelligibility. 28 female and 21 male speakers of Dutch were recorded producing 336 sentences,
each containing two target nouns. Recordings were compared to those of all other same-sex speakers using dynamic time warping (DTW). For each sentence, the DTW distance constituted a metric
of phonetic distance from one speaker to all other speakers. SA comprised the average of these distances. Later, the same participants performed a word recognition task on the target nouns in the same sentences, under three degraded listening conditions. In all three conditions, accuracy increased with SA. This held even when participants listened to their own utterances. These findings suggest that listeners process speech with respect to the statistical
properties of the language spoken in their community, rather than using their own speech as a reference

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

Like every other social sphere, science is influenced by individual characteristics of researchers. However, for investigations on scientific networks, only little data about the social background of researchers, e.g. social origin, gender, affiliation etc., is available.
This paper introduces ”Take it personally - TIP”, a conceptual model and library currently under development, which aims to support the
semantic enrichment of publication databases with semantically related background information which resides elsewhere in the (semantic) web, such as Wikidata.
The supplementary information enriches the original information in the publication databases and thus facilitates the creation of complex scientific knowledge graphs. Such enrichment helps to improve the scientometric analysis of scientific publications as they can also take social backgrounds of researchers into account and to understand social structure in research communities.

Abstract

Feedforward deep convolutional neural networks (DCNNs) are matching and even surpassing human performance on object recognition. This performance suggests that activation of a loose collection of image
features could support the recognition of natural object categories, without dedicated systems to solve specific visual subtasks. Recent findings in humans however, suggest that while feedforward activity may suffice for
sparse scenes with isolated objects, additional visual operations ('routines') that aid the recognition process (e.g. segmentation or grouping) are needed for more complex scenes. Linking human visual processing to
performance of DCNNs with increasing depth, we here explored if, how, and when object information is differentiated from the backgrounds they appear on. To this end, we controlled the information in both objects
and backgrounds, as well as the relationship between them by adding noise, manipulating background congruence and systematically occluding parts of the image. Results indicated less distinction between object- and background features for more shallow networks. For those networks, we observed a benefit of training on segmented objects (as compared to unsegmented objects). Overall, deeper networks trained on natural
(unsegmented) scenes seem to perform implicit 'segmentation' of the objects from their background, possibly by improved selection of relevant features.

Abstract

Human ethologists have shown that humans are both attracted to others and at the same time fear them. They refer to this kind of fear with the technical term ‘social fear’ and claim that “it is alleviated with personal acquaintance but remains a principle characteristic of interpersonal behaviour. As a result, we maintain various degrees of greater distance between ourselves and others depending on the amount of confidence we have in the other” (Eibl-Eibesfeldt 1989: 335). The goal of this task is to conduct exploratory, heuristic research to establish a new subproject that – based on a corpus of video data – will investigate various forms of human spatial behaviour cross-culturally.

Abstract

Die Sprachwissenschaft hat den Begriff und das Konzept ›Rituelle Kommunikation‹ von der vergleichenden Verhaltensforschung übernommen. Humanethologen unterscheiden eine Reihe von sogenannten ›Ausdrucksbewegungen‹, die in der Mimik, der Gestik, der Personaldistanz (Proxemik) und der Körperhaltung (Kinesik) zum Ausdruck kommen. Viele dieser Ausdrucksbewegungen haben sich zu spezifischen Signalen entwickelt. Ethologen definieren Ritualisierung als Veränderung von Verhaltensweisen im Dienst der Signalbildung. Die zu Signalen ritualisierten Verhaltensweisen sind Rituale. Im Prinzip kann jede Verhaltensweise zu einem Signal werden, entweder im Laufe der Evolution oder durch Konventionen, die in einer bestimmten Gemeinschaft gültig sind, die solche Signale kulturell entwickelt hat und die von ihren Mitgliedern tradiert und gelernt werden.

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

ndividual talkers may use various strategies to enrich their speech while speaking in noise (i.e., Lombard speech) to improve their intelligibility. The resulting acoustic-phonetic changes in Lombard speech vary amongst different speakers, but it is unclear what causes these talker differences, and what impact these differences have on intelligibility. This study investigates the potential role of articulatory control in talkers’ Lombard speech enrichment success. Seventy-eight speakers read out sentences in both their habitual style and in a condition where they were instructed to speak clearly while hearing loud speech-shaped noise. A diadochokinetic (DDK) speech task that requires speakers to repetitively produce word or non-word sequences as accurately and as rapidly as possible, was used to quantify their articulatory control. Individuals’ predicted intelligibility in both speaking styles (presented at -5 dB SNR) was measured using an acoustic glimpse-based metric: the High-Energy Glimpse Proportion (HEGP). Speakers’ HEGP scores show a clear effect of speaking condition (better HEGP scores in the Lombard than habitual condition), but no simple effect of articulatory control on HEGP, nor an interaction between speaking condition and articulatory control. This indicates that individuals’ speech enrichment success as measured by the HEGP metric was not predicted by DDK performance.