Publications

Abstract

We analyze the problem of response suggestion in a closed domain along a real-world scenario of a digital library. We present a text-processing pipeline to generate question-answer pairs from chat transcripts. On this limited amount of training data, we compare retrieval-based, conditioned-generation, and dedicated representation learning approaches for response suggestion. Our results show that retrieval-based methods that strive to find similar, known contexts are preferable over parametric approaches from the conditioned-generation family, when the training data is limited. We, however, identify a specific representation learning approach that is competitive to the retrieval-based approaches despite the training data limitation.

Abstract

We present multi-modal adversarial autoencoders for recommendation and evaluate them on two different tasks: citation recommendation and subject label recommendation. We analyze the effects of adversarial regularization, sparsity, and different input modalities. By conducting 408 experiments, we show that adversarial regularization consistently improves the performance of autoencoders for recommendation. We demonstrate, however, that the two tasks differ in the semantics of item co-occurrence in the sense that item co-occurrence resembles relatedness in case of citations, yet implies diversity in case of subject labels. Our results reveal that supplying the partial item set as input is only helpful, when item co-occurrence resembles relatedness. When facing a new recommendation task it is therefore crucial to consider the semantics of item co-occurrence for the choice of an appropriate model.

Abstract

Reading with Rapid Serial Visual Presentation (RSVP) has shown promise for optimizing screen space and increasing reading speed without compromising comprehension. Given the wide use of small-screen devices, the present study compared RSVP and traditional reading on three types of reading comprehension, reading speed, and subjective measures on a smart watch. Results confirm previous studies that show faster reading speed with RSVP without detracting from comprehension. Subjective data indicate that Traditional is strongly preferred to RSVP as a primary reading method. Given the optimal use of screen space, increased speed and comparable comprehension, future studies should focus on making RSVP a more comfortable format.

Abstract

The aim of this paper is to explore whether indexing terms for an audiovisual program can be derived from contextual texts automatically. For this we apply natural-language processing techniques to contextual texts of two Dutch TV-programs. We use a Dutch domain thesaurus to derive possible metadata. This possible metadata is ranked by an algorithm which uses the relations of the thesaurus. We evaluate the results by comparing them to human made descriptions.

Abstract

Print Friendly XML Gebre, Binyam Gebrekidan, Max Planck Institute for Psycholinguistics, The Netherlands, binyamgebrekidan.gebre [at] mpi.nl Wittenburg, Peter, Max Planck Institute for Psycholinguistics, The Netherlands, peter.wittenburg [at] mpi.nl Introduction Many gesture and sign language researchers manually annotate video recordings to systematically categorize, analyze and explain their observations. The number and kinds of annotations are so diverse and unpredictable that any attempt at developing non-adaptive automatic annotation systems is usually less effective. The trend in the literature has been to develop models that work for average users and for average scenarios. This approach has three main disadvantages. First, it is impossible to know beforehand all the patterns that could be of interest to all researchers. Second, it is practically impossible to find enough training examples for all patterns. Third, it is currently impossible to learn a model that is robustly applicable across all video quality-recording variations.

Abstract

Automatic annotation of gesture strokes is important for many gesture and sign language researchers. The unpredictable diversity of human gestures and video recording conditions require that we adopt a more adaptive case-by-case annotation model. In this paper, we present a work-in progress annotation model that allows a user to a) track hands/face b) extract features c) distinguish strokes from non-strokes. The hands/face tracking is done with color matching algorithms and is initialized by the user. The initialization process is supported with immediate visual feedback. Sliders are also provided to support a user-friendly adjustment of skin color ranges. After successful initialization, features related to positions, orientations and speeds of tracked hands/face are extracted using unique identifiable features (corners) from a window of frames and are used for training a learning algorithm. Our preliminary results for stroke detection under non-ideal video conditions are promising and show the potential applicability of our methodology.

Abstract

Not much is known about event apprehension, the earliest stage of information processing in elicited language production studies, using pictorial stimuli. A reason for our lack of knowledge on this process is that apprehension happens very rapidly (<350 ms after stimulus onset, Griffin & Bock 2000), making it difficult to measure the process directly. To broaden our understanding of apprehension, we analyzed landing positions and onset latencies of first fixations on visual stimuli (pictures of real-world events) given short stimulus presentation times, presupposing that the first fixation directly results from information processing during apprehension

Abstract

Recognizing the speech acts in our interlocutors’ utterances is a crucial prerequisite for conversation. However, it is not a trivial task given that the form and content of utterances is frequently underspecified for this level of meaning. In the present study we investigate participants’ competence in categorizing speech acts in such action-underspecific sentences and explore the time-course of speech act inferencing using a self-paced reading paradigm. The results demonstrate that participants are able to categorize the speech acts with very high accuracy, based on limited context and without any prosodic information. Furthermore, the results show that the exact same sentence is processed differently depending on the speech act it performs, with reading times starting to differ already at the first word. These results indicate that participants are very good at “getting” the speech acts, opening up a new arena for experimental research on action recognition in conversation.

Abstract

Acquiring the sounds of a language involves learning to recognize distributional patterns present in the input. We show that among adult learners, this distributional learning of auditory categories (which are conceived of here as probability density functions in a multidimensional space) is constrained by the salience of the dimensions that form the axes of this perceptual space. Only with a particular ratio of variation in the perceptual dimensions was category learning driven by the distributional properties of the input.

Abstract

“The most profound technologies are those that disappear. They weave themselves into the fabric of everyday life until they are indistinguishable from it.” (Weiser 1991, S. 94). – Die Behauptung stammt aus einem vielzitierten Text von Mark Weiser, ehemals Chief Technology Officer am berühmten Xerox Palo Alto Research Center (PARC), wo nicht nur einige bedeutende computertechnische Innovationen ihren Ursprung hatten, sondern auch grundlegende anthropologische Einsichten zum Umgang mit technischen Artefakten gewonnen wurden.1 In einem populärwissenschaftlichen Artikel mit dem Titel „The Computer for the 21st Century” entwarf Weiser 1991 die Vision einer Zukunft, in der wir nicht mehr mit einem einzelnen PC an unserem Arbeitsplatz umgehen – vielmehr seien wir in jedem Raum umgeben von hunderten elektronischer Vorrichtungen, die untrennbar in Alltagsgegenstände eingebettet und daher in unserer materiellen Umwelt gleichsam „verschwunden“ sind. Dabei ging es Weiser nicht allein um das ubiquitäre Phänomen, das in der Medientheorie als „Transparenz der Medien“ bekannt ist2 oder in allgemeineren Theorien der Alltagserfahrung als eine selbstverständliche Verwobenheit des Menschen mit den Dingen, die uns in ihrem Sinn vertraut und praktisch „zuhanden“ sind.3 Darüber hinaus zielte Weisers Vision darauf, unsere bereits existierende Umwelt durch computerlesbare Daten zu erweitern und in die Operationen eines solchen allgegenwärtigen Netzwerks alltägliche Praktiken gleichsam lückenlos zu integrieren: In der Welt, die Weiser entwirft, öffnen sich Türen für denjenigen, der ein bestimmtes elektronisches Abzeichen trägt, begrüßen Räume Personen, die sie betreten, mit Namen, passen sich Computerterminals an die Präferenzen individueller Nutzer an usw. (Weiser 1991, S. 99).

Abstract

The standard for the analysis of distorted voices is perceptual rating of read-out texts or spontaneous speech. Automatic voice evaluation, however, is usually done on stable sections of sustained vowels. In this paper, text-based and established vowel-based analysis are compared with respect to their ability to measure hoarseness and its subclasses. 73 hoarse patients (48.3±16.8 years) uttered the vowel /e/ and read the German version of the text “The North Wind and the Sun”. Five speech therapists and physicians rated roughness, breathiness, and hoarseness according to the German RBH evaluation scheme. The best human-machine correlations were obtained for measures based on the Cepstral Peak Prominence (CPP; up to |r | = 0.73). Support Vector Regression (SVR) on CPP-based measures and prosodic features improved the results further to r ≈0.8 and confirmed that automatic voice evaluation should be performed on a text recording.

Abstract

This study investigated whether a shadowing task can provide insights in the nature of reduction processes that are typical of casual speech. We focused on the shortening and presence versus absence of schwa and /t/ in Dutch past participles. Results showed that the absence of these segments was affected by the same variables as their shortening, suggesting that absence mostly resulted from extreme gradient shortening. This contrasts with results based on recordings of spontaneous conversations. We hypothesize that this difference is due to non-casual fast speech elicited by a shadowing task.

Abstract

Many current sentence generators lack the ability to compute elliptical versions of coordinated clauses in accordance with the rules for Gapping, Forward and Backward Conjunction Reduction, and SGF (Subject Gap in clauses with Finite/ Fronted verb). We describe a module (implemented in JAVA, with German and Dutch as target languages) that takes non-elliptical coordinated clauses as input and returns all reduced versions licensed by coordinative ellipsis. It is loosely based on a new psycholinguistic theory of coordinative ellipsis proposed by Kempen. In this theory, coordinative ellipsis is not supposed to result from the application of declarative grammar rules for clause formation but from a procedural component that interacts with the sentence generator and may block the overt expression of certain constituents.

Abstract

We describe a generation-oriented workbench for the Performance Grammar (PG) formalism, highlighting the treatment of certain word order and movement constraints in Dutch and German. PG enables a simple and uniform treatment of a heterogeneous collection of linear order phenomena in the domain of verb constructions (variably known as Cross-serial Dependencies, Verb Raising, Clause Union, Extraposition, Third Construction, Particle Hopping, etc.). The central data structures enabling this feature are clausal “topologies”: one-dimensional arrays associated with clauses, whose cells (“slots”) provide landing sites for the constituents of the clause. Movement operations are enabled by unification of lateral slots of topologies at adjacent levels of the clause hierarchy. The PGW generator assists the grammar developer in testing whether the implemented syntactic knowledge allows all and only the well-formed permutations of constituents.

Abstract

Faced with planning problems related to lexical access, speakers take advantage of a major function of disfluencies: buying time. It is reasonable, then, to expect that the structure of disfluencies sheds light on the mechanisms underlying lexical access. Using data from the Switchboard Corpus, we investigated the effect of semantic competition during lexical access on repetition disfluencies. We hypothesized that the more time the speaker needs to access the following unit, the longer the repetition. We examined the repetitions preceding verbs and nouns and tested predictors influencing the accessibility of these items. Results suggest that speed of lexical access negatively correlates with the length of repetition and that the main determinants of lexical access speed differ for verbs and nouns. Longer disfluencies before verbs appear to be due to significant paradigmatic competition from semantically similar verbs. For nouns, they occur when the noun is relatively unpredictable given the preceding context.

Abstract

Several studies examined cue competition in human learning by testing learners on a combination of conflicting cues rooting for different outcomes, with each cue perfectly predicting its outcome. A common result has been that learners faced with cue conflict choose the outcome associated with the rare cue (the Inverse Base Rate Effect, IBRE). Here, we investigate cue competition including IBRE with sentences containing cues to meanings in a visual world. We do not observe IBRE. Instead we find that position in the sentence strongly influences cue salience. Faced with conflict between an initial cue and a non-initial cue, learners choose the outcome associated with the initial cue, whether frequent or rare. However, a frequent configuration of non-initial cues that are not sufficiently salient on their own can overcome a competing salient initial cue rooting for a different meaning. This provides a possible explanation for certain recurring patterns in language change.

Abstract

We automatically predict properties of
wines on the basis of smell and flavor de-
scriptions from experts’ wine reviews. We
show wine experts are capable of describ-
ing their smell and flavor experiences in
wine reviews in a sufficiently consistent
manner, such that we can use their descrip-
tions to predict properties of a wine based
solely on language. The experimental re-
sults show promising F-scores when using
lexical and semantic information to predict
the color, grape variety, country of origin,
and price of a wine. This demonstrates,
contrary to popular opinion, that wine ex-
perts’ reviews really are informative.

Abstract

Longitudinally collected VOT data from an early English-Italian bilingual who became increasingly English-dominant was analyzed. Stops in English were always produced with significantly longer VOT than in Italian. However, the speaker did not show any significant change in the VOT production in either language over time, despite the clear dominance of English in his every day language use later in his life. The results indicate that – unlike L2 learners – early bilinguals may remain unaffected by language use with respect to phonetic realization.

Abstract

During spoken-word recognition, listeners experience phonological competition between multiple word candidates, which increases, relative to optimal listening conditions, when speech is masked by noise. Moreover, listeners activate semantic word knowledge during the word’s unfolding. Here, we replicated the effect of background noise on phonological competition and investigated to which extent noise affects the activation of semantic information in phonological competitors. Participants’ eye movements were recorded when they listened to sentences containing a target word and looked at three types of displays. The displays either contained a picture of the target word, or a picture of a phonological onset competitor, or a picture of a word semantically related to the onset competitor, each along with three unrelated distractors. The analyses revealed that, in noise, fixations to the target and to the phonological onset competitor were delayed and smaller in magnitude compared to the clean listening condition, most likely reflecting enhanced phonological competition. No evidence for the activation of semantic information in the phonological competitors was observed in noise and, surprisingly, also not in the clear. We discuss the implications of the lack of an effect and differences between the present and earlier studies.

Abstract

Co-speech gestures are an integral part of human face-to-face communication, but little is known about how pragmatic factors influence our comprehension of those gestures. The present study investigates how different types of recipients process iconic gestures in a triadic communicative situation. Participants (N = 32) took on the role of one of two recipients in a triad and were presented with 160 video clips of an actor speaking, or speaking and gesturing. Crucially, the actor’s eye gaze was manipulated in that she alternated her gaze between the two recipients. Participants thus perceived some messages in the role of addressed recipient and some in the role of unaddressed recipient. In these roles, participants were asked to make judgements concerning the speaker’s messages. Their reaction times showed that unaddressed recipients did comprehend speaker’s gestures differently to addressees. The findings are discussed with respect to automatic and controlled processes involved in gesture comprehension.

Abstract

What makes some words harder to learn than others in a second language? Although some robust factors have been identified based on small scale experimental studies, many relevant factors are difficult to study in such experiments due to the amount of data necessary to test them. Here, we investigate what factors affect the ease of learning of a word in a second language using a large data set of users learning English as a second language through the Duolingo mobile app. In a regression analysis, we test and confirm the well-studied effect of cognate status on word learning accuracy. Furthermore, we find significant effects for both cross-linguistic semantic alignment and English semantic density, two novel predictors derived from large scale distributional models of lexical semantics. Finally, we provide data on several other psycholinguistically plausible word level predictors. We conclude with a discussion of the limits, benefits and future research potential of using big data for investigating second language learning.

Abstract

Previous experiments and models show that the pressure to communicate can lead to the emergence of symbols in specific tasks. The experiment presented here suggests that the ability to use deictic gestures can reduce the pressure for symbols to emerge in co-operative tasks. In the 'gesture-only' condition, pairs built a structure together in 'Minecraft', and could only communicate using a small range of gestures. In the 'gesture-plus' condition, pairs could also use sound to develop a symbol system if they wished. All pairs were taught a pointing convention. None of the pairs we tested developed a symbol system, and performance was no different across the two conditions. We therefore suggest that deictic gestures, and non-referential means of organising activity sequences, are often sufficient for communication. This suggests that the emergence of linguistic symbols in early hominids may have been late and patchy with symbols only emerging in contexts where they could significantly improve task success or efficiency. Given the communicative power of pointing however, these contexts may be fewer than usually supposed. An approach for identifying these situations is outlined.

Abstract

A common assumption in the cognitive sciences is that artificial and natural language learning rely on shared mechanisms. However, attempts to bridge the two have yielded ambiguous results. We suggest that an empirical disconnect between the computations employed during learning and the methods employed at test may explain these mixed results. Further, we propose statistically-based chunking as a potential computational link between artificial and natural language learning. We compare the acquisition of non-adjacent dependencies to that of natural language structure using two types of tasks: reflection-based 2AFC measures, and processing-based recall measures, the latter being more computationally analogous to the processes used during language acquisition. Our results demonstrate that task-type significantly influences the correlations observed between artificial and natural language acquisition, with reflection-based and processing-based measures correlating within – but not across – task-type. These findings have fundamental implications for artificial-to-natural language comparisons, both methodologically and theoretically.

Abstract

The human hard palate shows between-subject variation
that is known to influence articulatory strategies.
In order to link such variation to human speech, we
are conducting a cross-sectional MRI study on multiple
populations. A model based on Bezier curves
using only three parameters was fitted to hard palate
MRI tracings using evolutionary computation. The
fits produced consistently yield high accuracies. For
future research, this new method may be used to classify
our MRI data on ethnic origins using e.g., cluster
analyses. Furthermore, we may integrate our model
into three-dimensional representations of the vocal
tract in order to investigate its effect on acoustics and
cultural transmission.

Abstract

In Iterated Learning (IL) experiments, a participant’s learned output serves as the next participant’s learning input (Kirby et al., 2014). IL can be used to model cultural transmission and has indicated that weak biases can be amplified through repeated cultural transmission (Kirby et al., 2007). So, for example, structural language properties can emerge over time because languages come to reflect the cognitive constraints in the individuals that learn and produce the language. Similarly, we propose that languages may also reflect certain anatomical biases. Do sound systems adapt to the affordances of the articulation space induced by the vocal tract?
The human vocal tract has inherent nonlinearities which might derive from acoustics and aerodynamics (cf. quantal theory, see Stevens, 1989) or biomechanics (cf. Gick & Moisik, 2015). For instance, moving the tongue anteriorly along the hard palate to produce a fricative does not result in large changes in acoustics in most cases, but for a small range there is an abrupt change from a perceived palato-alveolar [ʃ] to alveolar [s] sound (Perkell, 2012). Nonlinearities such as these might bias all human speakers to converge on a very limited set of phonetic categories, and might even be a basis for combinatoriality or phonemic ‘universals’.
While IL typically uses discrete symbols, Verhoef et al. (2014) have used slide whistles to produce a continuous signal. We conducted an IL experiment with human subjects who communicated using a digital slide whistle for which the degree of nonlinearity is controlled. A single parameter (α) changes the mapping from slide whistle position (the ‘articulator’) to the acoustics. With α=0, the position of the slide whistle maps Bark-linearly to the acoustics. As α approaches 1, the mapping gets more double-sigmoidal, creating three plateaus where large ranges of positions map to similar frequencies. In more abstract terms, α represents the strength of a nonlinear (anatomical) bias in the vocal tract.
Six chains (138 participants) of dyads were tested, each chain with a different, fixed α. Participants had to communicate four meanings by producing a continuous signal using the slide-whistle in a ‘director-matcher’ game, alternating roles (cf. Garrod et al., 2007).
Results show that for high αs, subjects quickly converged on the plateaus. This quick convergence is indicative of a strong bias, repelling subjects away from unstable regions already within-subject. Furthermore, high αs lead to the emergence of signals that oscillate between two (out of three) plateaus. Because the sigmoidal spaces are spatially constrained, participants increasingly used the sequential/temporal dimension. As a result of this, the average duration of signals with high α was ~100ms longer than with low α. These oscillations could be an expression of a basis for phonemic combinatoriality.
We have shown that it is possible to manipulate the magnitude of an articulator-induced non-linear bias in a slide whistle IL framework. The results suggest that anatomical biases might indeed constrain the design space of language. In particular, the signaling systems in our study quickly converged (within-subject) on the use of stable regions. While these conclusions were drawn from experiments using slide whistles with a relatively strong bias, weaker biases could possibly be amplified over time by repeated cultural transmission, and likely lead to similar outcomes.

Abstract

Many factors have been proposed to explain why groups of people use different speech sounds in their language. These range from cultural, cognitive, environmental (e.g., Everett, et al., 2015) to anatomical (e.g., vocal tract (VT) morphology). How could such anatomical properties have led to the similarities and differences in speech sound distributions between human languages?

It is known that hard palate profile variation can induce different articulatory strategies in speakers (e.g., Brunner et al., 2009). That is, different hard palate profiles might induce a kind of bias on speech sound production, easing some types of sounds while impeding others. With a population of speakers (with a proportion of individuals) that share certain anatomical properties, even subtle VT biases might become expressed at a population-level (through e.g., bias amplification, Kirby et al., 2007). However, before we look into population-level effects, we should first look at within-individual anatomical factors. For that, we have developed a computer-simulated analogue for a human speaker: an agent. Our agent is designed to replicate speech sounds using a production and cognition module in a computationally tractable manner.

Previous agent models have often used more abstract (e.g., symbolic) signals. (e.g., Kirby et al., 2007). We have equipped our agent with a three-dimensional model of the VT (the production module, based on Birkholz, 2005) to which we made numerous adjustments. Specifically, we used a 4th-order Bezier curve that is able to capture hard palate variation on the mid-sagittal plane (XXX, 2015). Using an evolutionary algorithm, we were able to fit the model to human hard palate MRI tracings, yielding high accuracy fits and using as little as two parameters. Finally, we show that the samples map well-dispersed to the parameter-space, demonstrating that the model cannot generate unrealistic profiles. We can thus use this procedure to import palate measurements into our agent’s production module to investigate the effects on acoustics. We can also exaggerate/introduce novel biases.

Our agent is able to control the VT model using the cognition module.

Previous research has focused on detailed neurocomputation (e.g., Kröger et al., 2014) that highlights e.g., neurobiological principles or speech recognition performance. However, the brain is not the focus of our current study. Furthermore, present-day computing throughput likely does not allow for large-scale deployment of these architectures, as required by the population model we are developing. Thus, the question whether a very simple cognition module is able to replicate sounds in a computationally tractable manner, and even generalize over novel stimuli, is one worthy of attention in its own right.

Our agent’s cognition module is based on running an evolutionary algorithm on a large population of feed-forward neural networks (NNs). As such, (anatomical) bias strength can be thought of as an attractor basin area within the parameter-space the agent has to explore. The NN we used consists of a triple-layered (fully-connected), directed graph. The input layer (three neurons) receives the formants frequencies of a target-sound. The output layer (12 neurons) projects to the articulators in the production module. A hidden layer (seven neurons) enables the network to deal with nonlinear dependencies. The Euclidean distance (first three formants) between target and replication is used as fitness measure. Results show that sound replication is indeed possible, with Euclidean distance quickly approaching a close-to-zero asymptote.

Statistical analysis should reveal if the agent can also: a) Generalize: Can it replicate sounds not exposed to during learning? b) Replicate consistently: Do different, isolated agents always converge on the same sounds? c) Deal with consolidation: Can it still learn new sounds after an extended learning phase (‘infancy’) has been terminated? Finally, a comparison with more complex models will be used to demonstrate robustness.

Abstract

It has been suggested that some individuals may benefit more from social robots than do others. Using second
language (L2) as an example, the present study examined how individual differences in attitudes toward robots and personality
traits may be related to learning outcomes. Preliminary results with 24 Turkish-speaking adults suggest that negative attitudes
toward robots, more specifically thoughts and anxiety about the negative social impact that robots may have on the society,
predicted how well adults learned L2 words from a social robot. The possible implications of the findings as well as future directions are also discussed

Abstract

In Korean, focus is expressed in accentual phrasing. To ascertain whether words focused in this manner enjoy a processing advantage analogous to that conferred by focus as expressed in, e.g, English and Dutch, we devised sentences with target words in one of four conditions: prosodic focus, syntactic focus, prosodic + syntactic focus, and no focus as a control. 32 native speakers of Korean listened to blocks of 10 sentences, then were presented visually with words and asked whether or not they had heard them. Overall, words with focus were recognised significantly faster and more accurately than unfocused words. In addition, words with syntactic focus or syntactic + prosodic focus were recognised faster than words with prosodic focus alone. As for other languages, Korean focus confers processing advantage on the words carrying it. While prosodic focus does provide an advantage, however, syntactic focus appears to provide the greater beneficial effect for recognition memory

Abstract

Human speakers often produce sentences incrementally. They can start speaking having in mind only a fragmentary idea of what they want to say, and while saying this they refine the contents underlying subsequent parts of the utterance. This capability imposes a number of constraints on the design of a syntactic processor. This paper explores these constraints and evaluates some recent computational sentence generators from the perspective of incremental production.

Abstract

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

Abstract

This study investigates the effects of adult age and speaker abilities on articulatory precision for sibilant productions. Normal-hearing young adults with
better sibilant discrimination have been shown to produce greater spectral sibilant contrasts. As reduced auditory feedback may gradually impact on feedforward
commands, we investigate whether articulatory precision as indexed by spectral mean for [s] and [S] decreases with age, and more particularly with agerelated
hearing loss. Younger, middle-aged and older adults read aloud words starting with the sibilants [s] or [S]. Possible effects of cognitive, perceptual, linguistic and sociolinguistic background variables
on the sibilants’ acoustics were also investigated. Sibilant contrasts were less pronounced for male than female speakers. Most importantly, for the fricative
[s], the spectral mean was modulated by individual high-frequency hearing loss, but not age. These results underscore that even mild hearing loss already affects articulatory precision.

Abstract

In German, word-initial lax fricatives may be produced with substantially reduced glottal vibration after voiceless obstruents. This assimilation occurs more frequently and to a larger extent across prosodic word boundaries than across phrase boundaries. Assimilatory devoicing makes the fricatives more similar to their tense counterparts and could thus hinder word recognition. The present study investigates how listeners cope with assimilatory devoicing. Results of a cross-modal priming experiment indicate that listeners compensate for assimilation in appropriate contexts. Prosodic structure moderates compensation for assimilation: Compensation occurs especially after phrase boundaries, where devoiced fricatives are sufficiently long to be confused with their tense counterparts.

Abstract

Prosodic structure has long been known to constrain phonological processes [1]. More recently, it has also been recognized as a source of fine-grained phonetic variation of speech sounds. In particular, segments in domain-initial position undergo prosodic strengthening [2, 3], which also implies more resistance to coarticulation in higher prosodic domains [5]. The present study investigates the combined effects of prosodic strengthening and assimilatory devoicing on word-initial fricatives in German, the functional implication of both processes for cues to the fortis-lenis contrast, and the influence of prosodic structure on listeners’ compensation for assimilation. Results indicate that 1. Prosodic structure modulates duration and the degree of assimilatory devoicing, 2. Phonological contrasts are maintained by speakers, but differ in phonetic detail across prosodic domains, and 3. Compensation for assimilation in perception is moderated by prosodic structure and lexical constraints.

Abstract

Citations play a crucial role in the scientific discourse, in information retrieval, and in bibliometrics. Many initiatives are currently promoting the idea of having free and open citation data. Creation of citation data, however, is not part of the cataloging workflow in libraries nowadays.
In this paper, we present our project Linked Open Citation Database, in which we design distributed processes and a system infrastructure based on linked data technology. The goal is to show that efficiently cataloging citations in libraries using a semi-automatic approach is possible. We specifically describe the current state of the workflow and its implementation. We show that we could significantly improve the automatic reference extraction that is crucial for the subsequent data curation. We further give insights on the curation and linking process and provide evaluation results that not only direct the further development of the project, but also allow us to discuss its overall feasibility.

Abstract

It is widely held that smells and flavors are impossible to put into words. In this paper we test this claim by seeking predictive patterns in wine reviews, which ostensibly aim to provide guides to perceptual content. Wine reviews have previously been critiqued as random and meaningless. We collected an English corpus of wine reviews with their structured metadata, and applied machine learning techniques to automatically predict the wine's color, grape variety, and country of origin. To train the three supervised classifiers, three different information sources were incorporated: lexical bag-of-words features, domain-specific terminology features, and semantic word embedding features. In addition, using regression analysis we investigated basic review properties, i.e., review length, average word length, and their relationship to the scalar values of price and review score. Our results show that wine experts do share a common vocabulary to describe wines and they use this in a consistent way, which makes it possible to automatically predict wine characteristics based on the review text alone. This means that odors and flavors may be more expressible in language than typically acknowledged.

Abstract

In different fields of the humanities annotations of multimodal resources are a necessary component of the research workflow. Examples include linguistics, psychology, anthropology, etc. However, creation of those annotations is a very laborious task, which can take 50 to 100 times the length of the annotated media, or more. This can be significantly improved by applying innovative audio and video processing algorithms, which analyze the recordings and provide automated annotations. This is the aim of the AVATecH project, which is a collaboration of the Max Planck Institute for Psycholinguistics (MPI) and the Fraunhofer institutes HHI and IAIS. In this paper we present a set of results of automated annotation together with an evaluation of their quality.

Abstract

Introduction Human spoken communication is multimodal, i.e. it encompasses both speech and gesture. Acoustic properties of voice, body movements, facial expression, etc. are an inherent and meaningful part of spoken interaction; they can provide attitudinal, grammatical and semantic information. In the recent years interest in audio-visual corpora has been rising rapidly as they enable investigation of different communicative modalities and provide more holistic view on communication (Kipp et al. 2009). Moreover, for some languages such corpora are the only available resource, as is the case for endangered languages for which no written resources exist.

Abstract

In this paper we describe audio and video processing algorithms that are developed in the scope of AVATecH project. The purpose of these algorithms is to shorten the time taken by manual annotation of audio and video recordings by extracting features from media files and creating semi-automated annotations. We show that the use of such supporting algorithms can shorten the annotation time to 30-50% of the time necessary to perform a fully manual annotation of the same kind.

Abstract

Recent work has investigated the emergence of structure in speech using experiments which use artificial continuous signals. Some experiments have had no limit on the duration which signals can have (e.g. Verhoef et al., 2014), and others have had time limitations (e.g. Verhoef et al., 2015). However, the effect of time constraints on the structure in signals has never been experimentally investigated.

Abstract

Structure of language is not only caused by cognitive processes, but also by physical aspects of the signalling modality. We test the assumptions surrounding the role which the physical aspects of the signal space will have on the emergence of structure in speech. Here, we use a signal creation task to test whether a signal space and a meaning space having similar dimensionalities will generate an iconic system with signal-meaning mapping and whether, when the topologies differ, the emergence of non-iconic structure is facilitated. In our experiments, signals are created using infrared sensors which use hand position to create audio signals. We find that people take advantage of signal-meaning mappings where possible. Further, we use trajectory probabilities and measures of variance to show that when there is a dimensionality mismatch, more structural strategies are used.

Abstract

Different linguistic modalities (speech or sign) offer different levels at which signals can iconically represent the world. One hypothesis argues that this iconicity has an effect on how linguistic structure emerges. However, exactly how and why these effects might come about is in need of empirical investigation. In this contribution, we present a signal creation experiment in which both the signalling space and the meaning space are manipulated so that different levels and types of iconicity are available between the signals and meanings. Signals are produced using an infrared sensor that detects the hand position of participants to generate auditory feedback. We find evidence that iconicity may be maladaptive for the discrimination of created signals. Further, we implemented Hidden Markov Models to characterise the structure within signals, which was also used to inform a metric for iconicity.

Abstract

Studies of sound symbolism have shown that people can associate sound and meaning in consistent ways when presented with maximally contrastive stimulus pairs of nonwords such as bouba/kiki (rounded/sharp) or mil/mal (small/big). Recent work has shown the effect extends to antonymic words from natural languages and has proposed a role for shared cross-modal correspondences in biasing form-to-meaning associations. An important open question is how the associations work, and particularly what the role is of sound-symbolic matches versus mismatches. We report on a learning task designed to distinguish between three existing theories by using a spectrum of sound-symbolically matching, mismatching, and neutral (neither matching nor mismatching) stimuli. Synthesized stimuli allow us to control for prosody, and the inclusion of a neutral condition allows a direct test of competing accounts. We find evidence for a sound-symbolic match boost, but not for a mismatch difficulty compared to the neutral condition.

Abstract

We present the Narrative Brain Dataset, an fMRI dataset that was collected during spoken presentation of short excerpts of three
stories in Dutch. Together with the brain imaging data, the dataset contains the written versions of the stimulation texts. The texts are
accompanied with stochastic (perplexity and entropy) and semantic computational linguistic measures. The richness and unconstrained
nature of the data allows the study of language processing in the brain in a more naturalistic setting than is common for fMRI studies.
We hope that by making NBD available we serve the double purpose of providing useful neural data to researchers interested in natural
language processing in the brain and to further stimulate data sharing in the field of neuroscience of language.

Abstract

Linguistic traits are often seen as reflecting cognitive biases and constraints (e.g. Christiansen & Chater, 2008). However, language must also adapt to properties of the channel through which communication between individuals occurs. Perhaps the most basic aspect of any communication channel is noise. Communicative signals can be blocked, degraded or distorted by other sources in the environment. This poses a fundamental problem for communication. On average, channel disruption accompanies problems in conversation every 3 minutes (27% of cases of other-initiated repair, Dingemanse et al., 2015). Linguistic signals must adapt to this harsh environment. While modern language structures are robust to noise (e.g. Piantadosi et al., 2011), we investigate how noise might have shaped the early emergence of structure in language. The obvious adaptation to noise is redundancy. Signals which are maximally different from competitors are harder to render ambiguous by noise. Redundancy can be increased by adding differentiating segments to each signal (increasing the diversity of segments). However, this makes each signal more complex and harder to learn. Under this strategy, holistic languages may emerge. Another strategy is reduplication - repeating parts of the signal so that noise is less likely to disrupt all of the crucial information. This strategy does not increase the difficulty of learning the language - there is only one extra rule which applies to all signals. Therefore, under pressures for learnability, expressivity and redundancy, reduplicated signals are expected to emerge. However, reduplication is not a pervasive feature of words (though it does occur in limited domains like plurals or iconic meanings). We suggest that this is due to the pressure for redundancy being lifted by conversational infrastructure for repair. Receivers can request that senders repeat signals only after a problem occurs. That is, robustness is achieved by repeating the signal across conversational turns (when needed) instead of within single utterances. As a proof of concept, we ran two iterated learning chains with pairs of individuals in generations learning and using an artificial language (e.g. Kirby et al., 2015). The meaning space was a structured collection of unfamiliar images (3 shapes x 2 textures x 2 outline types). The initial language for each chain was the same written, unstructured, fully expressive language. Signals produced in each generation formed the training language for the next generation. Within each generation, pairs played an interactive communication game. The director was given a target meaning to describe, and typed a word for the matcher, who guessed the target meaning from a set. With a 50% probability, a contiguous section of 3-5 characters in the typed word was replaced by ‘noise’ characters (#). In one chain, the matcher could initiate repair by requesting that the director type and send another signal. Parallel generations across chains were matched for the number of signals sent (if repair was initiated for a meaning, then it was presented twice in the parallel generation where repair was not possible) and noise (a signal for a given meaning which was affected by noise in one generation was affected by the same amount of noise in the parallel generation). For the final set of signals produced in each generation we measured the signal redundancy (the zip compressibility of the signals), the character diversity (entropy of the characters of the signals) and systematic structure (z-score of the correlation between signal edit distance and meaning hamming distance). In the condition without repair, redundancy increased with each generation (r=0.97, p=0.01), and the character diversity decreased (r=-0.99,p=0.001) which is consistent with reduplication, as shown below (part of the initial and the final language): Linear regressions revealed that generations with repair had higher overall systematic structure (main effect of condition, t = 2.5, p < 0.05), increasing character diversity (interaction between condition and generation, t = 3.9, p = 0.01) and redundancy increased at a slower rate (interaction between condition and generation, t = -2.5, p < 0.05). That is, the ability to repair counteracts the pressure from noise, and facilitates the emergence of compositional structure. Therefore, just as systems to repair damage to DNA replication are vital for the evolution of biological species (O’Brien, 2006), conversational repair may regulate replication of linguistic forms in the cultural evolution of language. Future studies should further investigate how evolving linguistic structure is shaped by interaction pressures, drawing on experimental methods and naturalistic studies of emerging languages, both spoken (e.g Botha, 2006; Roberge, 2008) and signed (e.g Senghas, Kita, & Ozyurek, 2004; Sandler et al., 2005).

Abstract

For (semi-)automated subject indexing systems in digital libraries, it is often more practical to use metadata such as the title of a publication instead of the full-text or the abstract. Therefore, it is desirable to have good text mining and text classification algorithms that operate well already on the title of a publication. So far, the classification performance on titles is not competitive with the performance on the full-texts if the same number of training samples is used for training. However, it is much easier to obtain title data in large quantities and to use it for training than full-text data. In this paper, we investigate the question how models obtained from training on increasing amounts of title training data compare to models from training on a constant number of full-texts. We evaluate this question on a large-scale dataset from the medical domain (PubMed) and from economics (EconBiz). In these datasets, the titles and annotations of millions of publications are available, and they outnumber the available full-texts by a factor of 20 and 15, respectively. To exploit these large amounts of data to their full potential, we develop three strong deep learning classifiers and evaluate their performance on the two datasets. The results are promising. On the EconBiz dataset, all three classifiers outperform their full-text counterparts by a large margin. The best title-based classifier outperforms the best full-text method by 9.4%. On the PubMed dataset, the best title-based method almost reaches the performance of the best full-text classifier, with a difference of only 2.9%.

Abstract

Scholars disagree about the extent to which language can tell us
about conceptualisation of the world. Some believe that language
is a direct window onto concepts: Having a word ‘‘bird’’, ‘‘table’’ or
‘‘sour’’ presupposes the corresponding underlying concept, BIRD,
TABLE, SOUR. Others disagree. Words are thought to be uninformative,
or worse, misleading about our underlying conceptual representations;
after all, our mental worlds are full of ideas that we
struggle to express in language. How could this be so, argue sceptics,
if language were a direct window on our inner life? In this presentation,
I consider what language can tell us about the
conceptualisation of taste. By considering linguistic data from
twenty unrelated cultures – varying in subsistence mode (huntergatherer
to industrial), ecological zone (rainforest jungle to desert),
dwelling type (rural and urban), and so forth – I argue any single language is, indeed, impoverished about what it can reveal about
taste. But recurrent lexicalisation patterns across languages can
provide valuable insights about human taste experience. Moreover,
language patterning is part of the data that a good theory of taste
perception has to be answerable for. Taste researchers, therefore,
cannot ignore the crosslinguistic facts.

Abstract

This Research Topic explores the question: what is the relationship between representations of time and space in cultures around the world? This question touches on the broader issue of how humans come to represent and reason about abstract entities – things we cannot see or touch. Time is a particularly opportune domain to investigate this topic. Across cultures, people use spatial representations for time, for example in graphs, time-lines, clocks, sundials, hourglasses, and calendars. In language, time is also heavily related to space, with spatial terms often used to describe the order and duration of events. In English, for example, we might move a meeting forward, push a deadline back, attend a long concert or go on a short break. People also make consistent spatial gestures when talking about time, and appear to spontaneously invoke spatial representations when processing temporal language. A large body of evidence suggests a close correspondence between temporal and spatial language and thought. However, the ways that people spatialize time can differ dramatically across languages and cultures. This research topic identifies and explores some of the sources of this variation, including patterns in spatial thinking, patterns in metaphor, gesture and other cultural systems. This Research Topic explores how speakers of different languages talk about time and space and how they think about these domains, outside of language. The Research Topic invites papers exploring the following issues: 1. Do the linguistic representations of space and time share the same lexical and morphosyntactic resources? 2. To what extent does the conceptualization of time follow the conceptualization of space?

Abstract

The results of a word-spotting experiment are presented in which Dutch listeners tried to spot different types of bisyllabic Dutch words embedded in different types of nonsense contexts. Embedded verbs were not reliably harder to spot than embedded nouns; this suggests that nouns and verbs are recognised via the same basic processes. Iambic words were no harder to spot than trochaic words, suggesting that trochaic words are not in principle easier to recognise than iambic words. Words were harder to spot in consonantal contexts (i.e., contexts which themselves could not be words) than in longer contexts which contained at least one vowel (i.e., contexts which, though not words, were possible words of Dutch). A control experiment showed that this difference was not due to acoustic differences between the words in each context. The results support the claim that spoken-word recognition is sensitive to the viability of sound sequences as possible words.

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

The ultimate goal of our research is to improve an existing speech-based computational model of human speech recognition on the task of simulating the role of fine-grained phonetic information in human speech processing. As part of this work we are investigating articulatory feature classifiers that are able to create reliable and accurate transcriptions of the articulatory behaviour encoded in the acoustic speech signal. Articulatory feature (AF) modelling of speech has received a considerable amount of attention in automatic speech recognition research. Different approaches have been used to build AF classifiers, most notably multi-layer perceptrons. Recently, deep neural networks have been applied to the task of AF classification. This paper aims to improve AF classification by investigating two different approaches: 1) investigating the usefulness of a deep Convolutional neural network (CNN) for AF classification; 2) integrating the Mel filtering operation into the CNN architecture. The results showed a remarkable improvement in classification accuracy of the CNNs over state-of-the-art AF classification results for Dutch, most notably in the minority classes. Integrating the Mel filtering operation into the CNN architecture did not further improve classification performance.

Abstract

This study demonstrates how interaction – specifically negotiation and repair – facilitates the emergence, evolution, and conventionalization of a silent gesture communication system. In a modified iterated learning paradigm, partners communicated noun-verb meanings using only silent gesture. The need to disambiguate similar noun-verb pairs drove these "new" language users to develop a morphology that allowed for quicker processing, easier transmission, and improved accuracy. The specific morphological system that emerged came about through a process of negotiation within the dyad, namely by means of repair. By applying a discourse analytic approach to the use of repair in an experimental methodology for language evolution, we are able to determine not only if interaction facilitates the emergence and learnability of a new communication system, but also how interaction affects such a system

Abstract

Our knowledge of speech perception is largely based on experiments conducted with carefully recorded clear speech presented under good listening conditions to undistracted listeners - a near-ideal situation, in other words. But the reality poses a set of different challenges. First of all, listeners may need to divide their attention between speech comprehension and another task (e.g., driving). Outside the laboratory, the speech signal is often slurred by less than careful pronunciation and the listener has to deal with background noise. Moreover, in a globalized world, listeners need to understand speech in more than their native language. Relatedly, the speakers we listen to often have a different language background so we have to deal with a foreign or regional accent we are not familiar with. Finally, outside the laboratory, speech perception is not an end in itself, but rather a mean to contribute to a conversation. Listeners do not only need to understand the speech they are hearing, they also need to use this information to plan and time their own responses. For this special topic, we invite papers that address any of these ecological aspects of speech perception.

Abstract

Frequent and predictable words are generally pronounced with less effort and are therefore acoustically more reduced than less frequent or unpredictable words. Local predictability can be operationalised by Transitional Probability (TP), which indicates how likely a word is to occur given its immediate context. We investigated whether and how probabilistic reduction effects on word durations change with adult age when reading aloud content words embedded in sentences. The results showed equally large frequency effects on verb and noun durations for both younger (Mage = 20 years) and older (Mage = 68 years) adults. Backward TP also affected word duration for younger and older adults alike. ForwardTP, however, had no significant effect on word duration in either age group. Our results resemble earlier findings of more robust BackwardTP effects compared to ForwardTP effects. Furthermore, unlike often reported decline in predictive processing with aging, probabilistic reduction effects remain stable across adulthood.

Abstract

It has been observed by several researchers that the Khoisan palate tends to lack a prominent alveolar ridge. A preliminary biomechanical model of click production was created to examine if these sounds might be subject to an anatomical bias associated with alveolar ridge size. Results suggest the bias is plausible, taking the form of decreased articulatory effort and improved volume change characteristics, however, further modelling and experimental research is required to solidify the claim.

Abstract

This paper investigated the learnability and generalizability of French schwa alternation by Dutch low-proficiency second language learners. We trained 40 participants on 24 new schwa words by exposing them equally often to the reduced and full forms of these words. We then assessed participants' accuracy and reaction times to these newly learnt words as well as 24 previously encountered schwa words with an auditory lexical decision task. Our results show learning of the new words in both forms. This suggests that lack of exposure is probably the main cause of learners' difficulties with reduced forms. Nevertheless, the full forms were slightly better recognized than the reduced ones, possibly due to phonetic and phonological properties of the reduced forms. We also observed no generalization to previously encountered words, suggesting that our participants stored both of the learnt word forms and did not create a rule that applies to all schwa words.

Abstract

Analyzing EEG signals recorded while participants are listening to continuous speech with the purpose of testing linguistic hypotheses is complicated by the fact that the signals simultaneously reflect exogenous acoustic excitation and endogenous linguistic processing. This makes it difficult to trace subtle differences that occur in mid-sentence position. We apply an analysis based on multivariate temporal response functions to uncover subtle mid-sentence effects. This approach is based on a per-stimulus estimate of the response of the neural system to speech input. Analyzing EEG signals predicted on the basis of the response functions might then bring to light conditionspecific differences in the filtered signals. We validate this approach by means of an analysis of EEG signals recorded with isolated word stimuli. Then, we apply the validated method to the analysis of the responses to the same words in the middle of meaningful sentences.

Abstract

In speech, words are often reduced rather than fully pronounced (e.g., (/ˈsʌmri/ for /ˈsʌməri/, summary). Non-native listeners may have problems in processing these reduced forms, because they have encountered them less often. This paper addresses the question whether this also holds for highly proficient non-natives and for words with similar forms and meanings in the non-natives' mother tongue (i.e., cognates). In an English auditory lexical decision task, natives and highly proficient Dutch non-natives of English listened to cognates and non-cognates that were presented in full or without their post-stress schwa. The data show that highly proficient learners are affected by reduction as much as native speakers. Nevertheless, the two listener groups appear to process reduced forms differently, because non-natives produce more errors on reduced cognates than on non-cognates. While listening to reduced forms, non-natives appear to be hindered by the co-activated lexical representations of cognate forms in their native language.

Abstract

Studies have shown that listeners segmenting unfamiliar languages transfer native-language (L1) segmentation cues. These studies, however, conflated L1 and recent linguistic exposure. The present study investigates the relative influences of L1 and recent linguistic exposure on the use of prosodic cues for segmenting an artificial language (AL). Participants were L1-French listeners, high-proficiency L2-French L1-English listeners, and L1-English listeners without functional knowledge of French. The prosodic cue assessed was F0 rise, which is word-final in French, but in English tends to be word-initial. 30 participants heard a 20-minute AL speech stream with word-final boundaries marked by F0 rise, and decided in a subsequent listening task which of two words (without word-final F0 rise) had been heard in the speech stream. The analyses revealed a marginally significant effect of L1 (all listeners) and, importantly, a significant effect of recent linguistic exposure (L1-French and L2-French listeners): accuracy increased with decreasing time in the US since the listeners’ last significant (3+ months) stay in a French-speaking environment. Interestingly, no effect of L2 proficiency was found (L2-French listeners).

Abstract

Statistical learning plays a key role in language processing, e.g., for speech segmentation. Older adults have been reported to show less statistical learning on the basis of visual input than younger adults. Given age-related changes in perception and cognition, we investigated whether statistical learning is also impaired in the auditory modality in older compared to younger adults and whether individual learning ability is associated with measures of perceptual (i.e., hearing sensitivity) and cognitive functioning in both age groups. Thirty younger and thirty older adults performed an auditory artificial-grammar-learning task to assess their statistical learning ability. In younger adults, perceptual effort came at the cost of processing resources required for learning. Inhibitory control (as indexed by Stroop colornaming performance) did not predict auditory learning. Overall, younger and older adults showed the same amount of auditory learning, indicating that statistical learning ability is preserved over the adult life span.

Abstract

This paper studies the influence of adverse listening conditions on exemplar effects in priming experiments that do not instruct participants to use their episodic memories. We conducted two lexical decision experiments, in which a prime and a target represented the same word type and could be spoken by the same or a different speaker. In Experiment 1, participants listened to clear speech, and showed no exemplar effects: they recognised repetitions by the same speaker as quickly as different speaker repetitions. In Experiment 2, the stimuli contained noise, and exemplar effects did arise. Importantly, Experiment 1 elicited longer average RTs than Experiment 2, a result that contradicts the time-course hypothesis, according to which exemplars only play a role when processing is slow. Instead, our findings support the hypothesis that exemplar effects arise under adverse listening conditions, when participants are stimulated to use their episodic memories in addition to their mental lexicons.