Publications

Abstract

Prediction in language has traditionally been studied using
simple designs in which neural responses to expected
and unexpected words are compared in a categorical
fashion. However, these designs have been contested
as being ‘prediction encouraging’, potentially exaggerating
the importance of prediction in language understanding.
A few recent studies have begun to address
these worries by using model-based approaches to probe
the effects of linguistic predictability in naturalistic stimuli
(e.g. continuous narrative). However, these studies
so far only looked at very local forms of prediction, using
models that take no more than the prior two words into
account when computing a word’s predictability. Here,
we extend this approach using a state-of-the-art neural
language model that can take roughly 500 times longer
linguistic contexts into account. Predictability estimates
fromthe neural network offer amuch better fit to EEG data
from subjects listening to naturalistic narrative than simpler
models, and reveal strong surprise responses akin to
the P200 and N400. These results show that predictability
effects in language are not a side-effect of simple designs,
and demonstrate the practical use of recent advances
in AI for the cognitive neuroscience of language.

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

Time is a fundamental aspect of human cognition and action. All languages have developed rich means to express various facets of time, such as bare time spans, their position on the time line, or their duration. The articles in this volume give an overview of what we know about the neural and cognitive representations of time that speakers can draw on in language. Starting with an overview of the main devices used to encode time in natural language, such as lexical elements, tense and aspect, the research presented in this volume addresses the relationship between temporal language, culture, and thought, the relationship between verb aspect and mental simulations of events, the development of temporal concepts, time perception, the storage and retrieval of temporal information in autobiographical memory, and neural correlates of tense processing and sequence planning. The psychological and neurobiological findings presented here will provide important insights to inform and extend current studies of time in language and in language acquisition.

Abstract

Thesaurus alignment plays an important role in realising efficient access to heterogeneous Cultural Heritage data. Current ontology alignment techniques, however, provide only limited value for such access as they consider little if any requirements from realistic use cases or application scenarios. In this paper, we focus on two real-world scenarios in a library context: thesaurus merging and book re-indexing. We identify their particular requirements and describe our approach of deploying and evaluating thesaurus alignment techniques in this context. We have applied our approach for the Ontology Alignment Evaluation Initiative, and report on the performance evaluation of participants’ tools wrt. the application scenario at hand. It shows that evaluations of tools requires significant effort, but when done carefully, brings many benefits.

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

n this study we investigated the durational correlates of lexical stress and pitch accent at normal and fast speech rate in Dutch. Previous literature on English shows that durations of lexically unstressed vowels are reduced more than stressed vowels when speakers increase their speech rate. We found that the same holds for Dutch, irrespective of whether the unstressed vowel is schwa or a "full" vowel. In the same line, we expected that vowels in words without a pitch accent would be shortened relatively more than vowels in words with a pitch accent. This was not the case: if anything, the accented vowels were shortened relatively more than the unaccented vowels. We conclude that duration is an important cue for lexical stress, but not for pitch accent.

Abstract

Studies on fast speech have shown that word-level timing of fast speech differs from that of normal rate speech in that unstressed syllables are shortened more than stressed syllables as speech rate increases. An earlier experiment showed that the intelligibility of time-compressed speech could not be improved by making its temporal organisation closer to natural fast speech. To test the hypothesis that segmental intelligibility is more important than prosodic timing in listening to timecompressed speech, the intelligibility of bisyllabic words was tested in three time-compression conditions: either stressed and unstressed syllable were compressed to the same degree, or the stressed syllable was compressed more than the unstressed syllable, or the reverse. As was found before, imitating wordlevel timing of fast speech did not improve intelligibility over linear compression. However, the results did not confirm the hypothesis either: there was no difference in intelligibility between the three compression conditions. We conclude that segmental intelligibility plays an important role, but further research is necessary to decide between the contributions of prosody and segmental intelligibility to the word-level intelligibility of time-compressed speech.

Abstract

Adult-to-child interactions are often characterized by prosodically-exaggerated speech accompanied by visually captivating co-speech gestures. In a series of adult studies, we have shown that these gestures are linked in a sophisticated manner to the prosodic structure of adults' utterances. In the current study, we use the Preferential Looking Paradigm to demonstrate that two-year-olds can use the alignment of these gestures to speech to deduce the meaning of words.

Abstract

Speakers express word meaning through systematic but non-canonical acoustic variation of tone of voice (ToV), i.e., variation of speaking rate, pitch, vocal effort, or loudness. Words are, for example, pronounced at a higher pitch when referring to small than to big referents. In the present study, we examined whether listeners can use ToV to learn the meaning of novel adjectives (e.g., “blicket”). During training, participants heard sentences such as “Can you find the blicket one?” spoken with ToV representing hot-cold, strong-weak, and big-small. Participants’ eye movements to two simultaneously shown objects with properties representing the relevant two endpoints (e.g., an elephant and an ant for big-small) were monitored. Assignment of novel adjectives to endpoints was counterbalanced across participants. During test, participants heard the sentences spoken with a neutral ToV, while seeing old or novel picture pairs varying along the same dimensions (e.g., a truck and a car for big-small). Participants had to click on the adjective’s referent. As evident from eye movements, participants did not infer the intended meaning during first exposure, but learned the meaning with the help of ToV during training. At test listeners applied this knowledge to old and novel items even in the absence of informative ToV.

Abstract

This study investigates effects of prosodic factors (prominence, boundary) on coarticulatory Vnasalization in Australian English (AusE) in CVN and NVC in comparison to those in American English
(AmE). As in AmE, prominence was found to
lengthen N, but to reduce V-nasalization, enhancing N’s nasality and V’s orality, respectively (paradigmatic contrast enhancement). But the prominence effect in CVN was more robust than that in AmE. Again similar to findings in AmE, boundary
induced a reduction of N-duration and V-nasalization phrase-initially (syntagmatic contrast enhancement), and increased the nasality of both C and V phrasefinally.
But AusE showed some differences in terms
of the magnitude of V nasalization and N duration. The results suggest that the linguistic contrast enhancements underlie prosodic-structure modulation of coarticulatory V-nasalization in
comparable ways across dialects, while the fine phonetic detail indicates that the phonetics-prosody interplay is internalized in the individual dialect’s phonetic grammar.

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

The ”download first, then process paradigm” is still the predominant working method amongst the research community. The web-based paradigm, however, offers many advantages from a tool development and data management perspective as they allow a quick adaptation to changing research environments. Moreover, new ways of combining tools and data are increasingly becoming available and will eventually enable a true web-based workflow approach, thus challenging the ”download first, then process” paradigm. The necessary infrastructure for managing, exploring and enriching language resources via the Web will need to be delivered by projects like CLARIN and DARIAH

Abstract

In this paper, we describe a unifying approach to tackle data heterogeneity issues for lexica and related resources. We present LEXUS, our software that implements the Lexical Markup Framework (LMF) to uniformly describe and manage lexica of different structures. LEXUS also makes use of a central Data Category Registry (DCR) to address terminological issues with regard to linguistic concepts as well as the handling of working and object languages. Finally, we report on ViCoS, a LEXUS extension, providing support for the definition of arbitrary semantic relations between lexical entries or parts thereof.

Abstract

To achieve true interoperability for valuable linguistic resources different levels of variation need to be addressed. ISO Technical Committee 37, Terminology and other language and content resources, is developing a Data Category Registry. This registry will provide a reusable set of data categories. A new implementation, dubbed ISOcat, of the registry is currently under construction. This paper shortly describes the new data model for data categories that will be introduced in this implementation. It goes on with a sketch of the standardization process. Completed data categories can be reused by the community. This is done by either making a selection of data categories using the ISOcat web interface, or by other tools which interact with the ISOcat system using one of its various Application Programming Interfaces. Linguistic resources that use data categories from the registry should include persistent references, e.g. in the metadata or schemata of the resource, which point back to their origin. These data category references can then be used to determine if two or more resources share common semantics, thus providing a level of interoperability close to the source data and a promising layer for semantic alignment on higher levels

Abstract

We present a software module, the LAT Bridge, which enables bidirectionalcommunication between the annotation and exploration tools developed at the MaxPlanck Institute for Psycholinguistics as part of our Language ArchivingTechnology (LAT) tool suite. These existing annotation and exploration toolsenable the annotation, enrichment, exploration and archive management oflinguistic resources. The user community has expressed the desire to usedifferent combinations of LAT tools in conjunction with each other. The LATBridge is designed to cater for a number of basic data interaction scenariosbetween the LAT annotation and exploration tools. These interaction scenarios(e.g. bootstrapping a wordlist, searching for annotation examples or lexicalentries) have been identified in collaboration with researchers at ourinstitute.We had to take into account that the LAT tools for annotation and explorationrepresent a heterogeneous application scenario with desktop-installed andweb-based tools. Additionally, the LAT Bridge has to work in situations wherethe Internet is not available or only in an unreliable manner (i.e. with a slowconnection or with frequent interruptions). As a result, the LAT Bridge’sarchitecture supports both online and offline communication between the LATannotation and exploration tools.

Abstract

Many theories hold that semantic variation in the world’s languages can be explained in terms of a universal conceptual space that is partitioned differently by different languages. Recent work has supported this view in the semantic domain of containers (Malt et al., 1999), and assumed it in the domain of spatial relations (Khetarpal et al., 2009), based in both cases on similarity judgments derived from pile-sorting of stimuli. Here, we reanalyze data from these two studies and find a more complex picture than these earlier studies suggested. In both cases we find that sorting is similar across speakers of different languages (in line with the earlier studies), but nonetheless reflects the sorter’s native language (in contrast with the earlier studies). We conclude that there are cross-culturally shared conceptual tendencies that can be revealed by pile-sorting, but that these tendencies may be modulated to some extent by language. We discuss the implications of these findings for accounts of semantic variation.

Abstract

Words can be distinguished by segmental differences or by suprasegmental differences or both. Studies from English suggest that suprasegmentals play little role in human spoken-word recognition; English stress, however, is nearly always unambiguously coded in segmental structure (vowel quality); this relationship is less close in Dutch. The present study directly compared the effects of segmental and suprasegmental mispronunciation on word recognition in Dutch. There was a strong effect of suprasegmental mispronunciation, suggesting that Dutch listeners do exploit suprasegmental information in word recognition. Previous findings indicating the effects of mis-stressing for Dutch differ with stress position were replicated only when segmental change was involved, suggesting that this is an effect of segmental rather than suprasegmental processing.

Abstract

Previous studies on Cantonese Chinese showed that rising
question intonation contours on low-toned words lead to
frequent misperceptions of the tones. Here we explored the
processing consequences of this interaction between tone and
intonation by comparing the processing and identification of
monosyllabic critical words at the end of questions and
statements, using a tone identification task, and ERPs as an
online measure of speech comprehension. Experiment 1
yielded higher error rates for the identification of low tones at
the end of questions and a larger N400-P600 pattern, reflecting
processing difficulty and reanalysis, compared to other
conditions. In Experiment 2, we investigated the effect of
immediate lexical context on the tone by intonation interaction.
Increasing contextual constraints led to a reduction in errors
and the disappearance of the P600 effect. These results
indicate that there is an immediate interaction between tone,
intonation, and context in online speech comprehension. The
difference in performance and activation patterns between the
two experiments highlights the significance of context in
understanding a tone language, like Cantonese-Chinese.

Abstract

A recurrently connected attractor neural network with a Hebbian learning rule is currently our best ANN analogy for a piece cortex. Functionally biological memory operates on a spectrum of time scales with regard to induction and retention, and it is modulated in complex ways by sub-cortical neuromodulatory systems. Moreover, biological memory networks are commonly believed to be highly distributed and engage many co-operating cortical areas. Here we focus on the temporal aspects of induction and retention of memory in a connectionist type attractor memory model of a piece of cortex. A continuous time, forgetful Bayesian-Hebbian learning rule is described and compared to the characteristics of LTP and LTD seen experimentally. More generally, an attractor network implementing this learning rule can operate as a long-term, intermediate-term, or short-term memory. Modulation of the print-now signal of the learning rule replicates some experimental memory phenomena, like e.g. the von Restorff effect.

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

Digital processing of medical images has helped physicians and patients during past years by allowing examination and diagnosis on a very precise level. Nowadays possibly the biggest deal of support it can offer for modern healthcare is the use of high performance computing architectures to treat the huge amounts of data that can be collected by modern acquisition devices. This paper presents a parallel processing implementation of an image segmentation algorithm that operates on a computer cluster equipped with 10 processing units. Thanks to well-organized distribution of the workload we manage to significantly shorten the execution time of the developed algorithm and reach a performance gain very close to linear.

Abstract

Verb learning is important for young children. While most
previous research has focused on linguistic and conceptual
challenges in early verb learning (e.g. Gentner, 1982, 2006),
the present paper examined early verb learning at the
attentional level and quantified the input for early verb learning
by measuring verb-action co-occurrence statistics in parent-
child interaction from the learner’s perspective. To do so, we
used head-mounted eye tracking to record fine-grained
multimodal behaviors during parent-infant joint play, and
analyzed parent speech, parent and infant action, and infant
attention at the moments when parents produced verb labels.
Our results show great variability across different action verbs,
in terms of frequency of verb utterances, frequency of
corresponding actions related to verb meanings, and infants’
attention to verbs and actions, which provide new insights on
why some verbs are harder to learn than others.

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

Young children demonstrate the ability to make inferences about the preferences of other agents based on their choices. However, there exists no overarching account of what children are doing when they learn about preferences or how they use that knowledge. We use a rational model of preference learning, drawing on ideas from economics and computer science, to explain the behavior of children in several recent experiments. Specifically, we show how a simple econometric model can be extended to capture two- to four-year-olds’ use of statistical information in inferring preferences, and their generalization of these preferences.

Abstract

We examined how communicators can switch between speaker and listener role with such accurate timing. During conversations, the majority of role transitions happens with a gap or overlap of only a few hundred milliseconds. This suggests that listeners can predict when the turn of the current speaker is going to end. Our hypothesis is that listeners know when a turn ends because they know how it ends. Anticipating the last words of a turn can help the next speaker in predicting when the turn will end, and also in anticipating the content of the turn, so that an appropriate response can be prepared in advance. We used the stimuli material of an earlier experiment (De Ruiter, Mitterer & Enfield, 2006), in which subjects were listening to turns from natural conversations and had to press a button exactly when the turn they were listening to ended. In the present experiment, we investigated if the subjects can complete those turns when only an initial fragment of the turn is presented to them. We found that the subjects made better predictions about the last words of those turns that had more accurate responses in the earlier button press experiment.

Abstract

Continuous Bag of Words (CBOW) is a powerful text embedding method. Due to its strong capabilities to encode word content, CBOW embeddings perform well on a wide range of downstream tasks while being efficient to compute. However, CBOW is not capable of capturing the word order. The reason is that the computation of CBOW's word embeddings is commutative, i.e., embeddings of XYZ and ZYX are the same. In order to address this shortcoming, we propose a
learning algorithm for the Continuous Matrix Space Model, which we call Continual Multiplication of Words (CMOW). Our algorithm is an adaptation of word2vec, so that it can be trained on large quantities of unlabeled text. We empirically show that CMOW better captures linguistic properties, but it is inferior to CBOW in memorizing word content. Motivated by these findings, we propose a hybrid model that combines the strengths of CBOW and CMOW. Our results show that the hybrid CBOW-CMOW-model retains CBOW's strong ability to memorize word content while at the same time substantially improving its ability to encode other linguistic information by 8%. As a result, the hybrid also performs better on 8 out of 11 supervised downstream tasks with an average improvement of 1.2%.

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

Studies have claimed that blind people’s spatial representations are different from sighted people, and blind people display superior auditory processing. Due to the nature of auditory and haptic information, it has been proposed that blind people have spatial representations that are more sequential than sighted people. Even the temporary loss of sight—such as through blindfolding—can affect spatial representations, but not much research has been done on this topic. We compared blindfolded and sighted people’s linguistic spatial expressions and non-linguistic localization accuracy to test how blindfolding affects the representation of path in auditory motion events. We found that blindfolded people were as good as sighted people when localizing simple sounds, but they outperformed sighted people when localizing auditory motion events. Blindfolded people’s path related speech also included more sequential, and less holistic elements. Our results indicate that even temporary loss of sight influences spatial representations of auditory motion events

Abstract

Lombard speech, speech produced in noise, is acoustically different from speech produced in quiet (plain speech) in several ways, including having a higher and wider F0 range (pitch). Extensive research on native Lombard speech does not consider that non-natives experience a higher cognitive load while producing
speech and that the native language may influence the non-native speech. We investigated pitch range in plain and Lombard speech in native and non-natives.
Dutch and American-English speakers read contrastive question-answer pairs in quiet and in noise in English, while the Dutch also read Dutch sentence pairs. We found that Lombard speech is characterized by a wider pitch range than plain speech, for all speakers (native English, non-native English, and native Dutch).
This shows that non-natives also widen their pitch range in Lombard speech. In sentences with early-focus, we see the same increase in pitch range when going from plain to Lombard speech in native and non-native English, but a smaller increase in native Dutch. In sentences with late-focus, we see the biggest increase for the native English, followed by non-native English and then native Dutch. Together these results indicate an effect of the native language on non-native Lombard speech.

Abstract

Lombard speech, speech produced in noise, is
typically produced with a higher fundamental
frequency (F0, pitch) compared to speech in quiet. This paper examined the potential differences in native and non-native Lombard speech by analyzing median pitch in sentences with early- or late-focus produced in quiet and noise. We found an increase in pitch in late-focus sentences in noise for Dutch speakers in both English and Dutch, and for American-English speakers in English. These results
show that non-native speakers produce Lombard speech, despite their higher cognitive load. For the early-focus sentences, we found a difference between the Dutch and the American-English speakers. Whereas the Dutch showed an increased F0 in noise
in English and Dutch, the American-English speakers did not in English. Together, these results suggest that some acoustic characteristics of Lombard speech, such as pitch, may be language-specific, potentially
resulting in the native language influencing the non-native Lombard speech.

Abstract

Grice in pragmatics and Levelt in psycholinguistics have proposed models of human communication where the starting point of communicative action is an individual intention. This assumption, though, has to face serious objections with regard to the alleged existence of explicit representations of the communicative goals to be pursued. Here evidence is surveyed which shows that in fact speaking may ordinarily be a quite automatic activity prompted by contextual cues and driven by behavioural schemata abstracted away from social regularities. On the one hand, this means that there could exist no intentions in the sense of explicit representations of communicative goals, following from deliberate reasoning and triggering the communicative action. On the other hand, however, there are reasons to allow for a weaker notion of intention than this, according to which communication is an intentional affair, after all. Communicative action is said to be intentional in this weaker sense to the extent that it is subject to a double mechanism of control, with respect both to present-directed and future-directed intentions.

Abstract

Lexical knowledge influences how human listeners make decisions about speech sounds. Positive lexical effects (faster responses to target sounds in words than in nonwords) are robust across several laboratory tasks, while negative effects (slower responses to targets in more word-like nonwords than in less word-like nonwords) have been found in phonetic decision tasks but not phoneme monitoring tasks. The present experiments tested whether negative lexical effects are therefore a task-specific consequence of the forced choice required in phonetic decision. We compared phoneme monitoring and phonetic decision performance using the same Dutch materials in each task. In both experiments there were positive lexical effects, but no negative lexical effects. We observe that in all studies showing negative lexical effects, the materials were made by cross-splicing, which meant that they contained perceptual evidence supporting the lexically-consistent phonemes. Lexical knowledge seems to influence phonemic decision-making only when there is evidence for the lexically-consistent phoneme in the speech signal.

Abstract

We briefly describe the Merge model of phonemic decision-making, and, in the light of general arguments about the possible role of feedback in spoken-word recognition, defend Merge's feedforward structure. Merge not only accounts adequately for the data, without invoking feedback connections, but does so in a parsimonious manner.

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 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

Words are reduced in spontaneous speech. If reductions are constrained by functional (i.e., perception and production) constraints, they should not be arbitrary. This hypothesis was tested by examing the pronunciations of high- to mid-frequency words in a Dutch and a German spontaneous speech corpus. In logistic-regression models the "reduction likelihood" of a phoneme was predicted by fixed-effect predictors such as position within the word, word length, word frequency, and stress, as well as random effects such as phoneme identity and word. The models for Dutch and German show many communalities. This is in line with the assumption that similar functional constraints influence reductions in both languages.

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

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 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

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

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

Japanese listeners detected Japanese words embedded at the end of nonsense sequences (e.g., kaba 'hippopotamus' in gyachikaba). When the final portion of the preceding context together with the initial portion of the word (e.g., here, the sequence chika) was compatible with many lexical competitors, recognition of the embedded word was more difficult than when such a sequence was compatible with few competitors. This clear effect of competition, established here for preceding context in Japanese, joins similar demonstrations, in other languages and for following contexts, to underline that the functional architecture of the human spoken-word recognition system is a universal one.

Abstract

Dutch and Spanish differ in how predictable the stress pattern is as a function of the segmental content: it is correlated with syllable weight in Dutch but not in Spanish. In the present study, two experiments were run to compare the abilities of Dutch and Spanish speakers to separately process segmental and stress information. It was predicted that the Spanish speakers would have more difficulty focusing on the segments and ignoring the stress pattern than the Dutch speakers. The task was a speeded classification task on CVCV syllables, with blocks of trials in which the stress pattern could vary versus blocks in which it was fixed. First, we found interference due to stress variability in both languages, suggesting that the processing of segmental information cannot be performed independently of stress. Second, the effect was larger for Spanish than for Dutch, suggesting that that the degree of interference from stress variation may be partially mitigated by the predictability of stress placement in the language.

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

Cognitive neuroscience approaches the brain as a cognitive system: a system that functionally is conceptualized in terms of information processing. We outline some aspects of this concept and consider a physical system to be an information processing device when a subclass of its physical states can be viewed as representational/cognitive and transitions between these can be conceptualized as a process operating on these states by implementing operations on the corresponding representational structures. We identify a generic and fundamental problem in cognition: sequentially organized structured processing. Structured sequence processing provides the brain, in an essential sense, with its processing logic. In an approach addressing this problem, we illustrate how to integrate levels of analysis within a framework of adaptive dynamical systems. We note that the dynamical system framework lends itself to a description of asynchronous event-driven devices, which is likely to be important in cognition because the brain appears to be an asynchronous processing system. We use the human language faculty and natural language processing as a concrete example through out.

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

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

Abstract

Dutch listeners' looks to printed words were tracked while they listened to instructions to click with their mouse on one of them. When presented with targets from word pairs where the first two syllables were segmentally identical but differed in stress location, listeners used stress information to recognize the target before segmental information disambiguated the words. Furthermore, the amount of lexical competition was influenced by the presence or absence of word-initial stress.

Abstract

Segmental as well as suprasegmental information is used by Dutch listeners to recognize words. The time-course of the effect of suprasegmental stress information on spoken-word recognition was investigated in a previous study, in which we tracked Dutch listeners' looks to arrays of four printed words as they listened to spoken sentences. Each target was displayed along with a competitor that did not differ segmentally in its first two syllables but differed in stress placement (e.g., 'CENtimeter' and 'sentiMENT'). The listeners' eye-movements showed that stress information is used to recognize the target before distinct segmental information is available. Here, we examine the role of durational information in this effect. Two experiments showed that initial-syllable duration, as a cue to lexical stress, is not interpreted dependent on the speaking rate of the preceding carrier sentence. This still held when other stress cues like pitch and amplitude were removed. Rather, the speaking rate of the preceding carrier affected the speed of word recognition globally, even though the rate of the target itself was not altered. Stress information modulated lexical competition, but did so independently of the rate of the preceding carrier, even if duration was the only stress cue present.

Abstract

To understand spoken words listeners have to cope with seemingly meaningless variability in the speech signal. Speakers vary, for example, their tone of voice (ToV) by changing speaking rate, pitch, vocal effort, and loudness. This variation is independent of "linguistic prosody" such as sentence intonation or speech rhythm. The variation due to ToV, however, is not random. Speakers use, for example, higher pitch when referring to small objects than when referring to large objects and importantly, adult listeners are able to use these non-lexical ToV cues to distinguish between the meanings of antonym pairs (e.g., big-small; Nygaard, Herold, & Namy, 2009). In the present study, we asked whether listeners infer the meaning of novel adjectives from ToV and subsequently interpret these adjectives according to the learned meaning even in the absence of ToV. Moreover, if listeners actually acquire these adjectival meanings, then they should generalize these word meanings to novel referents. ToV would thus be a semantic cue to lexical acquisition. This hypothesis was tested in an exposure-test paradigm with adult listeners. In the experiment listeners' eye movements to picture pairs were monitored. The picture pairs represented the endpoints of the adjectival dimensions big-small, hot-cold, and strong-weak (e.g., an elephant and an ant represented big-small). Four picture pairs per category were used. While viewing the pictures participants listened to lexically unconstraining sentences containing novel adjectives, for example, "Can you find the foppick one?" During exposure, the sentences were spoken in infant-directed speech with the intended adjectival meaning expressed by ToV. Word-meaning pairings were counterbalanced across participants. Each word was repeated eight times. Listeners had no explicit task. To guide listeners' attention to the relation between the words and pictures, three sets of filler trials were included that contained real English adjectives (e.g., full-empty). In the subsequent test phase participants heard the novel adjectives in neutral adult-directed ToV. Test sentences were recorded before the speaker was informed about intended word meanings. Participants had to choose which of two pictures on the screen the speaker referred to. Picture pairs that were presented during the exposure phase and four new picture pairs per category that varied along the critical dimensions were tested. During exposure listeners did not spontaneously direct their gaze to the intended referent at the first presentation. But as indicated by listener's fixation behavior, they quickly learned the relationship between ToV and word meaning over only two exposures. Importantly, during test participants consistently identified the intended referent object even in the absence of informative ToV. Learning was found for all three tested categories and did not depend on whether the picture pairs had been presented during exposure. Listeners thus use ToV not only to distinguish between antonym pairs but they are able to extract word meaning from ToV and assign this meaning to novel words. The newly learned word meanings can then be generalized to novel referents even in the absence of ToV cues. These findings suggest that ToV can be used as a semantic cue to lexical acquisition. References Nygaard, L. C., Herold, D. S., & Namy, L. L. (2009) The semantics of prosody: Acoustic and perceptual evidence of prosodic correlates to word meaning. Cognitive Science, 33. 127-146.

Abstract

A aquisição da leitura decorre ao longo de diversas etapas, desde o momento em que a criança inicia o contacto com o alfabeto até ao momento em que se torna um leitor competente, apto a ler correcta e fluentemente. Compreender a evolução desta competência através de uma análise da diferenciação do peso de variáveis preditoras da leitura possibilita teorizar sobre os mecanismos cognitivos envolvidos nas diferentes fases de desenvolvimento da leitura. Realizámos um estudo transversal com 568 alunos do segundo ao quarto ano do primeiro ciclo do Ensino Básico, em que se avaliou o impacto de capacidades de processamento fonológico, nomeação rápida, conhecimento letra-som e vocabulário, bem como de capacidades cognitivas mais gerais (inteligência não-verbal e memória de trabalho), na exactidão e velocidade da leitura. De uma forma geral, os resultados mostraram que, apesar da consciência fonológica permanecer como o preditor mais importante da exactidão e fluência da leitura, o seu peso decresce à medida que a escolaridade aumenta. Observou-se também que, à medida que o contributo da consciência fonológica para a explicação da velocidade de leitura diminuía, aumentava o contributo de outras variáveis mais associadas ao automatismo e reconhecimento lexical, tais como a nomeação rápida e o vocabulário. Em suma, podemos dizer que ao longo da escolaridade se observa uma alteração dinâmica dos processos cognitivos subjacentes à leitura, o que sugere que a criança evolui de uma estratégia de leitura ancorada em processamentos sub-lexicais, e como tal mais dependente de processamentos fonológicos, para uma estratégia baseada no reconhecimento ortográfico das palavras.

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

The recognition of emotions of disgust, anger and fear have been shown to be significantly impaired in Huntington’s disease (eg,Sprengelmeyer et al, 1997, 2006; Gray et al, 1997; Milders et al, 2003,Montagne et al, 2006; Johnson et al, 2007; De Gelder et al, 2008). The relative impairment of these emotions might have implied a recognition impairment specific to negative emotions. Could the asymmetric recognition deficits be due not to the complexity of the emotion but rather reflect the complexity of the task? In the current study, 15 Huntington’s patients and 16 control subjects were presented with negative and positive non-speech emotional vocalisations that were to be identified as anger, fear, sadness, disgust, achievement, pleasure and amusement in a forced-choice paradigm. This experiment more accurately matched the negative emotions with positive emotions in a homogeneous modality. The resulting dually impaired ability of Huntington’s patients to identify negative and positive non-speech emotional vocalisations correctly provides evidence for an overall emotional recognition deficit in the disease. These results indicate that previous findings of a specificity in emotional recognition deficits might instead be due to the limitations of the visual modality. Previous experiments may have found an effect of emotional specificy due to the presence of a single positive emotion, happiness, in the midst of multiple negative emotions. In contrast with the previous literature, the study presented here points to a global deficit in the recognition of emotional sounds.

Abstract

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

Abstract

This paper presents the first application of polynomial modeling as a means for validating phonological pitch accent labels to German data. It is compared to traditional phonetic analysis (measuring minima, maxima, alignment). The traditional method fares better in classification, but results are comparable in statistical accent pair testing. Robustness tests show that pitch correction is necessary in both cases. The approaches are discussed in terms of their practicability, applicability to other domains of research and interpretability of their results.

Abstract

The current study reports specific cases in which a positive transfer of perceptual ability from the music domain to the language domain occurs. We tested whether musical training enhances discrimination and identification performance of L2 speech sounds (timing features, nasal consonants and vowels). Native Dutch and Japanese speakers with different musical training experience, matched for their estimated verbal IQ, participated in the experiments. Results indicated that musical training strongly increases one’s ability to perceive timing information in speech signals. We also found a benefit of musical training on discrimination performance for a subset of the tested vowel contrasts.

Abstract

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

Abstract

Despite differences in language, culture, and ecology, some human characteristics are similar in people all over the world, while other features vary from one group to the next. These similarities and differences can inform arguments about what aspects of the human mind are part of our shared biological heritage and which are predominantly products of culture and language. I will present data from a cross-cultural project investigating the recognition of non-verbal vocalizations of emotions, such as screams and laughs, across two highly different cultural groups. English participants were compared to individuals from remote, culturally isolated Namibian villages. Vocalizations communicating the so-called “basic emotions” (anger, disgust, fear, joy, sadness, and surprise) were bidirectionally recognised. In contrast, a set of additional positive emotions was only recognised within, but not across, cultural boundaries. These results indicate that a number of primarily negative emotions are associated with vocalizations that can be recognised across cultures, while at least some positive emotions are communicated with culture-specific signals. I will discuss these findings in the context of accounts of emotions at differing levels of analysis, with an emphasis on the often-neglected positive emotions.

Abstract

Many studies suggest that emotional signals can be recognized across cultures and modalities. But to what extent are these signals innate and to what extent are they learned? This study investigated whether auditory learning is necessary for the production of recognizable emotional vocalizations by examining the vocalizations produced by people born deaf. Recordings were made of eight congenitally deaf Dutch individuals, who produced non-verbal vocalizations of a range of negative and positive emotions. Perception was examined in a forced-choice task with hearing Dutch listeners (n = 25). Considerable variability was found across emotions, suggesting that auditory learning is more important for the acquisition of certain types of vocalizations than for others. In particular, achievement and surprise sounds were relatively poorly recognized. In contrast, amusement and disgust vocalizations were well recognized, suggesting that for some emotions, recognizable vocalizations can develop without any auditory learning. The implications of these results for models of emotional communication are discussed, and other routes of social learning available to the deaf individuals are considered.

Abstract

Vocalizations like screams and laughs are used to communicate affective states, but what acoustic cues in these signals require vocal learning and which ones are innate? This study investigated the role of auditory learning in the production of non-verbal emotional vocalizations by examining the vocalizations produced by people born deaf. Recordings were made of congenitally deaf Dutch individuals and matched hearing controls, who produced non-verbal vocalizations of a range of negative and positive emotions. Perception was examined in a forced-choice task with hearing Dutch listeners (n = 25), and judgments were analyzed together with acoustic cues, including envelope, pitch, and spectral measures. Considerable variability was found across emotions and acoustic cues, and the two types of information were related for a sub-set of the emotion categories. These results suggest that auditory learning is less important for the acquisition of certain types of vocalizations than for others (particularly amusement and relief), and they also point to a less central role for auditory learning of some acoustic features in affective non-verbal vocalizations. The implications of these results for models of vocal emotional communication are discussed.

Abstract

In the context of the source-filter theory of speech, it is well established that intelligibility is heavily reliant on information carried by the filter, that is, spectral cues (e.g., Faulkner et al., 2001; Shannon et al., 1995). However, the extraction of other types of information in the speech signal, such as emotion and identity, is less well understood. In this study we investigated the extent to which emotion recognition in speech depends on filterdependent cues, using a forced-choice emotion identification task at ten levels of noise-vocoding ranging between one and 32 channels. In addition, participants performed a speech intelligibility task with the same stimuli. Our results indicate that compared to speech intelligibility, emotion recognition relies less on spectral information and more on cues typically signaled by source variations, such as voice pitch, voice quality, and intensity. We suggest that, while the reliance on spectral dynamics is likely a unique aspect of human speech, greater phylogenetic continuity across species may be found in the communication of affect in vocalizations.

Abstract

Research using event-related brain potentials (ERPs) has demonstrated an early differential effect in fronto-central regions when processing emotional, as compared to affectively neutral facial stimuli (e.g., Eimer & Holmes, 2002). In this talk, data demonstrating a similar effect in the auditory domain will be presented. ERPs were recorded in a one-back task where participants had to identify immediate repetitions of emotion category, such as a fearful sound followed by another fearful sound. The stimulus set consisted of non-verbal emotional vocalisations communicating positive and negative sounds, as well as neutral baseline conditions. Similarly to the facial domain, fear sounds as compared to acoustically controlled neutral sounds, elicited a frontally distributed positivity with an onset latency of about 150 ms after stimulus onset. These data suggest the existence of a rapid multi-modal frontocentral mechanism discriminating emotional from non-emotional human signals.

Abstract

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

Abstract

Listeners outperform ASR systems in every speech recognition task. However, what is not clear is where this human advantage originates. This paper investigates the role of acoustic feature representations. We test four (MFCCs, PLPs, Mel Filterbanks, Rate Maps) acoustic representations, with and without ‘pitch’ information, using the same backend. The results are compared with listener results at the level of articulatory feature classification. While no acoustic feature representation reached the levels of human performance, both MFCCs and Rate maps achieved good scores, with Rate maps nearing human performance on the classification of voicing. Comparing the results on the most difficult articulatory features to classify showed similarities between the humans and the SVMs: e.g., ‘dental’ was by far the least well identified by both groups. Overall, adding pitch information seemed to hamper classification performance.

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.