Publications

Abstract

Listeners outperform automatic speech recognition systems at every level, including the very basic level of consonant identification. What is not clear is where the human advantage originates. Does the fault lie in the acoustic representations of speech or in the recognizer architecture, or in a lack of compatibility between the two? Many insights can be gained by carrying out a detailed human-machine comparison. The purpose of the Interspeech 2008 Consonant Challenge is to promote focused comparisons on a task involving intervocalic consonant identification in noise, with all participants using the same training and test data. This paper describes the Challenge, listener results and baseline ASR performance.

Abstract

The multimedia annotation tool ELAN was enhanced within the Corpus NGT project by a number of new and improved functions. Most of these functions were not specific to working with sign language video data, and can readily be used for other annotation purposes as well. Their direct utility for working with large amounts of annotation files during the development and use of the Corpus NGT project is what unites the various functions, which are described in this paper. In addition, we aim to characterise future developments that will be needed in order to work efficiently with larger amounts of annotation files, for which a closer integration with the use and display of metadata is foreseen.

Abstract

The Corpus NGT is an ambitious effort to record and archive video data from Sign Language of the Netherlands (Nederlandse Gebarentaal: NGT), guaranteeing online access to all interested parties and long-term availability. Data are collected from 100 native signers of NGT of different ages and from various regions in the country. Parts of these data are annotated and/or translated; the annotations and translations are part of the corpus. The Corpus NGT is accommodated in the Browsable Corpus based at the Max Planck Institute for Psycholinguistics. In this paper we share our experiences in data collection, video processing, annotation/translation and licensing involved in building the corpus.

Abstract

Listeners rely on native-language rhythm in segmenting speech; in different languages, stress-, syllable- or mora-based rhythm is exploited. This language-specificity affects listening to non- native speech, if native procedures are applied even though inefficient for the non-native language. However, speakers of two languages with similar rhythmic interpretation should segment their own and the other language similarly. This was observed to date only for related languages (English-Dutch; French-Spanish). We now report experiments in which Japanese listeners heard Telugu, a Dravidian language unrelated to Japanese, and Telugu listeners heard Japanese. In both cases detection of target sequences in speech was harder when target boundaries mismatched mora boundaries, exactly the pattern that Japanese listeners earlier exhibited with Japanese and other languages. These results suggest that Telugu and Japanese listeners use similar procedures in segmenting speech, and support the idea that languages fall into rhythmic classes, with aspects of phonological structure affecting listeners' speech segmentation.

Abstract

Listeners heard an ambiguous /f-s/ in nonword contexts where only one of /f/ or /s/ was legal (e.g., frul/*srul or *fnud/snud). In later categorisation of a phonetic continuum from /f/ to /s/, their category boundaries had shifted; hearing -rul led to expanded /f/ categories, -nud expanded /s/. Thus phonotactic sequence information alone induces perceptual retuning of phoneme category boundaries; lexical access is not required.

Abstract

Speech segmentation procedures may differ in speakers of different languages. Earlier work based on French speakers listening to French words suggested that the syllable functions as a segmentation unit in speech processing. However, while French has relatively regular and clearly bounded syllables, other languages, such as English, do not. No trace of syllabifying segmentation was found in English listeners listening to English words, French words, or nonsense words. French listeners, however, showed evidence of syllabification even when they were listening to English words. We conclude that alternative segmentation routines are available to the human language processor. In some cases speech segmentation may involve the operation of more than one procedure.

Abstract

The Possible Word Constraint is a proposed mechanism whereby listeners avoid recognising words spuriously embedded in other words. It applies to words leaving a vowelless residue between their edge and the nearest known word or syllable boundary. The present study tests the usefulness of this constraint via lexical statistics of both English and Dutch. The analyses demonstrate that the constraint removes a clear majority of embedded words in speech, and thus can contribute significantly to the efficiency of human speech recognition

Abstract

Assimilation of place of articulation across a nasal and a following stop consonant is obligatory in Japanese, but not in Dutch. In four experiments the processing of assimilated forms by speakers of Japanese and Dutch was compared, using a task in which listeners blended pseudo-word pairs such as ranga-serupa. An assimilated blend of this pair would be rampa, an unassimilated blend rangpa. Japanese listeners produced significantly more assimilated than unassimilated forms, both with pseudo-Japanese and pseudo-Dutch materials, while Dutch listeners produced significantly more unassimilated than assimilated forms in each materials set. This suggests that Japanese listeners, whose native-language phonology involves obligatory assimilation constraints, represent the assimilated nasals in nasal-stop sequences as unmarked for place of articulation, while Dutch listeners, who are accustomed to hearing unassimilated forms, represent the same nasal segments as marked for place of articulation.

Abstract

The perceptual processing of speech is a constant interplay of multiple competing albeit convergent processes: acoustic input vs. higher-level representations, universal mechanisms vs. language-specific, veridical traces of speech experience vs. construction and activation of abstract representations. The present summary concerns the third of these issues. The ability to generalise across experience and to deal with resulting abstractions is the hallmark of human cognition, visible even in early infancy. In speech processing, abstract representations play a necessary role in both production and perception. New sorts of evidence are now informing our understanding of the breadth of this role.

Abstract

In English and Dutch, listeners entrain to prosodic contours to predict where focus will fall in an utterance. However, is this strategy universally available, even in languages with different phonological systems? In a phoneme detection experiment, we examined whether prosodic entrainment is also found in Mandarin Chinese, a tone language, where in principle the use of pitch for lexical identity may take precedence over the use of pitch cues to salience. Consistent with the results from Germanic languages, response times were facilitated when preceding intonation predicted accent on the target-bearing word. Acoustic analyses revealed greater F0 range in the preceding intonation of the predicted-accent sentences. These findings have implications for how universal and language-specific mechanisms interact in the processing of salience.

Abstract

Languages contain tens of thousands of words, but these are constructed from a tiny handful of phonetic elements. Consequently, words resemble one another, or can be embedded within one another, a coup stick snot with standing. me process of spoken-word recognition by human listeners involves activation of multiple word candidates consistent with the input, and direct competition between activated candidate words. Further, human listeners are sensitive, at an early, prelexical, stage of speeeh processing, to constraints on what could potentially be a word of the language.

Abstract

English and French listeners performed two tasks - click location and speeded click detection - with both English and French sentences, closely matched for syntactic and phonological structure. Clicks were located more accurately in open- than in closed-class words in both English and French; they were detected more rapidly in open- than in closed-class words in English, but not in French. The two listener groups produced the same pattern of responses, suggesting that higher-level linguistic processing was not involved in these tasks.

Abstract

The phoneme detection task is widely used in spoken word recognition research. Alphabetically literate participants, however, are more used to explicit representations of letters than of phonemes. The present study explored whether phoneme detection is sensitive to how target phonemes are, or may be, orthographically realised. Listeners detected the target sounds [b,m,t,f,s,k] in word-initial position in sequences of isolated English words. Response times were faster to the targets [b,m,t], which have consistent word-initial spelling, than to the targets [f,s,k], which are inconsistently spelled, but only when listeners’ attention was drawn to spelling by the presence in the experiment of many irregularly spelled fillers. Within the inconsistent targets [f,s,k], there was no significant difference between responses to targets in words with majority and minority spellings. We conclude that performance in the phoneme detection task is not necessarily sensitive to orthographic effects, but that salient orthographic manipulation can induce such sensitivity.

Abstract

Psycholinguists are saddled with a paradox. Their aim is to construct a model of human language processing, which will hold equally well for the processing of any language, but this aim cannot be achieved just by doing experiments in any language. They have to compare processing of many languages, and actively search for effects which are specific to a single language, even though a model which is itself specific to a single language is really the last thing they want.

Abstract

The performance of automatic speech recognition systems is usually assessed in terms of error rate. Human speech recognition produces few errors, but relative difficulty of processing can be assessed via response time techniques. We report the construction of a measure analogous to response time in a machine recognition system. This measure may be compared directly with human response times. We conducted a trial comparison of this type at the phoneme level, including both tense and lax vowels and a variety of consonant classes. The results suggested similarities between human and machine processing in the case of consonants, but differences in the case of vowels.

Abstract

The causal correlations between human genetic variants and linguistic (typological) features could represent the mechanism required for gradual, accretionary models of language evolution. The causal link is mediated by the process of cultural transmission of language across generations in a population of genetically biased individuals. The particular case of Tone, ASPM and Microcephalin is discussed as an illustration. It is proposed that this type of genetically-influenced linguistic bias, coupled with a fundamental role for genetic and linguistic diversities, provides a better explanation for the evolution of language and linguistic universals.

Abstract

Retrieval of autobiographical memories is facilitated by activation of perceptuo-motor aspects of the experience, for example a congruent body position at the time of the experiencing and the time of retelling (Dijkstra, Kaschak, & Zwaan, 2007). The present study examined whether similar retrieval facilitation occurs when the direction of motor action is congruent with the valence of emotional memories. Consistent with evidence that people mentally represent emotions spatially (Casasanto, in press), participants moved marbles between vertically stacked boxes at a higher rate when the direction of movement was congruent with the valence of the memory they retrieved (e.g., upward for positive memories, downward for negative memories) than when direction and valence were incongruent (t(22)=4.24, p<.001). In addition, valence-congruent movements facilitated access to these memories, resulting in shorter retrieval times (t(22)=2.43, p<.05). Results demonstrate bidirectional influences between the emotional content of autobiographical memories and irrelevant motor actions.

Abstract

In this paper we discuss the influence of semantically unexpected information on the prosodic realization of contrast.
For this purpose, we examine the interplay between unexpectedness and various discourse factors that have been claimed to enhance the accentuation of contrastive
information: contrast direction, syntactic status, and discourse distance. We conducted a production experiment in Dutch in which speakers described scenes consisting of moving fruits with unnatural colors. We found that a general cognitive factor such as the unexpectedness of a property has a strong impact on the intonational marking of contrast, over and above the influence of the immediate discourse context.

Abstract

In this paper, we discuss the interplay of factors that influence the intonational marking of contrast in Dutch. In particular, we examine how prominence is expressed at the prosodic level when semantically abnormal information conflicts with contrastive information. For this purpose, we conducted a production experiment in Dutch in which speakers described scenes containing fruits with unnatural colors. We found that semantically abnormal information invokes cognitive prominence which corresponds to intonational prominence. Moreover, the results show that abnormality may overrule the accentual marking of information structural categories such as contrastive focus. If semantically abnormal information becomes integrated into the larger discourse context, its prosodic prominence decreases in favor of the signaling of information structural categories such as contrastive focus.

Abstract

Empirical studies addressing the age factor in second language acquisition have mainly been concerned with a comparison of end state data (from learners before and after the closure of a putative Critical Period for language acquisition) to the native speaker norm. Based on longitudinal corpus data, this paper investigates the affect of age on end state, rate and the process of acquisition and addresses the question of whether different grammatical domains are equally affected. To this end, the paper presents summarized findings from the acquisition of word order and inflectional morphology in L2 German by Russian learners of different ages and discusses theoretical implications that can be drawn from this evidence.

Abstract

Brain-to-brain interfaces, in which brains are physically connected without the intervention of language, promise new ways of collaboration and communication between humans. I examine the narrow view of language implicit in current conceptions of brain-to-brain interfaces and put forward a constructive alternative, stressing the role of language in organising joint agency. Two features of language stand out as crucial: its selectivity, which provides people with much-needed filters between public words and private worlds; and its negotiability, which provides people with systematic opportunities for calibrating understanding and expressing consent and dissent. Without these checks and balances, brain-to-brain interfaces run the risk of reducing people to the level of amoeba in a slime mold; with them, they may mature to become useful extensions of human agency

Abstract

This entry provides some orientation and task suggestions on how to explore the perceptual world of your field site and the interaction between the cultural world and the sensory lexicon in your community. The material consists of procedural texts; soundscapes; other documentary and observational tasks. The goal of this task is to explore the perceptual world of your field site and the interaction between the cultural world and the sensory lexicon in your community.

Abstract

Linguistic discovery is viewpoint-dependent, just like our ideas about what is marginal and what is central in language. In this essay I consider two supposed marginalia —ideophones and interjections— which provide some useful pointers for widening our field of view. Ideophones challenge us to take a fresh look at language and consider how it is that our communication system combines multiple modes of representation. Interjections challenge us to extend linguistic inquiry beyond sentence level, and remind us that language is social-interactive at core. Marginalia, then, are not the obscure, exotic phenomena that can be safely ignored: they represent opportunities for innovation and invite us to keep pushing the edges of linguistic inquiry.

Abstract

Responding to similarity, difference, and relative magnitude (SDM) is ubiquitous in the animal kingdom. However, humans seem unique in the ability to represent relative magnitude (‘more’/‘less’) and similarity (‘same’/‘different’) as abstract relations that take arguments (e.g., greater-than (x,y)). While many models use structured relational representations of magnitude and similarity, little progress has been made on how these representations arise. Models that developuse these representations assume access to computations of similarity and magnitude a priori, either encoded as features or as output of evaluation operators. We detail a mechanism for producing invariant responses to “same”, “different”, “more”, and “less” which can be exploited to compute similarity and magnitude as an evaluation operator. Using DORA (Doumas, Hummel, & Sandhofer, 2008), these invariant responses can serve be used to learn structured relational representations of relative magnitude and similarity from pixel images of simple shapes

Abstract

This paper summarizes the results of a descriptive syntactic category analysis of child English no which reveals that young children use and represent no as a determiner and negatives like no pen as NPs, contra standard analyses.

Abstract

This paper presents Advanced Glossing, a proposal for a general glossing format designed for language documentation, and a specific setup for the Shoebox-program that implements Advanced Glossing to a large extent. Advanced Glossing (AG) goes beyond the traditional Interlinear Morphemic Translation, keeping syntactic and morphological information apart from each other in separate glossing tables. AG provides specific lines for different kinds of annotation – phonetic, phonological, orthographical, prosodic, categorial, structural, relational, and semantic, and it allows for gradual and successive, incomplete, and partial filling in case that some information may be irrelevant, unknown or uncertain. The implementation of AG in Shoebox sets up several databases. Each documented text is represented as a file of syntactic glossings. The morphological glossings are kept in a separate database. As an additional feature interaction with lexical databases is possible. The implementation makes use of the interlinearizing automatism provided by Shoebox, thus obtaining the table format for the alignment of lines in cells, and for semi-automatic filling-in of information in glossing tables which has been extracted from databases

Abstract

Der vorliegende Beitrag zum Kieler Symposium1 stellt die Resultate eines Teilbereichs meiner Arbeit zum Guarani vor, nämlich einen Vorschlag zur Analyse der Personenpräfixe dieser Sprache und der mit ihnen verbundenen grammatischen Kategorien. Die im Titel angedeutete lexikographische Fragestellung bedarf einer näheren Erläuterung, die ich im Zusammenhang mit einer kurzen Darstellung der Motivation für meine Untersuchungen geben will

Abstract

Digitizing and annotating texts and field recordings Given that several initiatives worldwide currently explore the new field of documentation of endangered languages, the E-MELD project proposes to survey and unite procedures, techniques and results in order to achieve its main goal, ''the formulation and promulgation of best practice in linguistic markup of texts and lexicons''. In this context, this year's workshop deals with the processing of recorded texts. I assume the most valuable contribution I could make to the workshop is to show the procedures and methods used in the Awetí Language Documentation Project. The procedures applied in the Awetí Project are not necessarily representative of all the projects in the DOBES program, and they may very well fall short in several respects of being best practice, but I hope they might provide a good and concrete starting point for comparison, criticism and further discussion. The procedures to be exposed include: * taping with digital devices, * digitizing (preliminarily in the field, later definitely by the TIDEL-team at the Max Planck Institute in Nijmegen), * segmenting and transcribing, using the transcriber computer program, * translating (on paper, or while transcribing), * adding more specific annotation, using the Shoebox program, * converting the annotation to the ELAN-format developed by the TIDEL-team, and doing annotation with ELAN. Focus will be on the different types of annotation. Especially, I will present, justify and discuss Advanced Glossing, a text annotation format developed by H.-H. Lieb and myself designed for language documentation. It will be shown how Advanced Glossing can be applied using the Shoebox program. The Shoebox setup used in the Awetí Project will be shown in greater detail, including lexical databases and semi-automatic interaction between different database types (jumping, interlinearization). ( Freie Universität Berlin and Museu Paraense Emílio Goeldi, with funding from the Volkswagen Foundation.)

Abstract

With the present essay I pursue a threefold aim as will be explained in the following paragraphs. Since I cannot expect my readers to be familiar with the language studied, Guaran´ı, more information about this language will be given in the next subsection.

Abstract

This paper describes the verbal Tense-Aspect-Mood system of Awetí (Tupian, Central Brazil) in a Word-and-Paradigm approach. One classification of Awetí verb forms contains clear aspect categories. A second set of independent classifications renders at least four moods and contains a third major TAM classification, factuality, that has one mainly temporal category Future, while others are partially or wholly modal. Structural categories reflect the formal composition of the forms. Some forms are synthetic, ‘marked’ only by means of affixes, but many are analytic, containing auxiliary particles. With selected sample forms we demonstrate in detail the interplay of structural and functional categories in Awetí verb paradigms.

Abstract

Traditionally, there have been three standard assumptions made about the Parallelism Effect on VP-ellipsis, namely that the effect is categorical, that it applies asymmetrically and that it is uniquely due to syntactic factors. Based on the results of a series of experiments involving online and offline tasks, it will be argued that the Parallelism Effect is instead noncategorical and interactional. The factors investigated include construction type, conceptual and morpho-syntactic recoverability, finiteness and anaphor type (to test VP-anaphora). The results show that parallelism is gradient rather than categorical, effects both VP-ellipsis and anaphora, and is influenced by both structural and non-structural factors.

Abstract

We report the results of a large-scale (N=1571) experiment to investigate whether spoken words can emerge from the process of repeated imitation. Participants played a version of the children’s game “Telephone”. The first generation was asked to imitate recognizable environmental sounds (e.g., glass breaking, water splashing); subsequent generations imitated the imitators for a total of 8 generations. We then examined whether the vocal imitations became more stable and word-like, retained a resemblance to the original sound, and became more suitable as learned category labels. The results showed (1) the imitations became progressively more word-like, (2) even after 8 generations, they could be matched above chance to the environmental sound that motivated them, and (3) imitations from later generations were more effective as learned category labels. These results show how repeated imitation can create progressively more word-like forms while retaining a semblance of iconicity.

Abstract

This field guide is for eliciting information about grammatical resources used in describing perceptual events and perception-based properties and states. A list of leading questions outlines an underlying semantic space for events/states of perception, against which language-specific constructions may be defined. It should be used as an entry point into a flexible exploration of the structures and constraints which are specific to the language you are working on. The goal is to provide a cross-linguistically comparable description of the constructions of a language used in describing perceptual events and states. The core focus is to discover any sensory asymmetries, i.e., ways in which different sensory modalities are treated differently with respect to these constructions.

Abstract

This task is designed to elicit virtual 3D ‘models’ created in gesture space using iconic and other representational gestures. This task has been piloted with Lao speakers, where two speakers were asked to explain the meaning of terms referring to different kinds of fish trap mechanisms. The task elicited complex performances involving a range of iconic gestures, and with especially interesting use of (a) the ‘model/diagram’ in gesture space as a virtual object, (b) the non-dominant hand as a prosodic/semiotic anchor, (c) a range of different techniques (indexical and iconic) for evoking meaning with the hand, and (d) the use of nearby objects and parts of the body as semiotic ‘props’.

Abstract

People of all cultures have some degree of concern with categorizing types of communicative social action. All languages have words with meanings like speak, say, talk, complain, curse, promise, accuse, nod, wink, point and chant. But the exact distinctions they make will differ in both quantity and quality. How is communicative social action categorised across languages and cultures? The goal of this task is to establish a basis for cross-linguistic comparison of native metalanguages for social action.

Abstract

Research on video- and audio-recordings of spontaneous naturally-occurring conversation in English has shown that conversation is a rule-guided, practice-oriented domain that can be investigated for its underlying mechanics or structure. Systematic study could yield something like a grammar for conversation. The goal of this task is to acquire a corpus of video-data, for investigating the underlying structure(s) of interaction cross-linguistically and cross-culturally

Abstract

We want to know how people think about their field of kin, on the supposition that it is quasi-spatial. To get some insights here, we need to video a discussion about kinship reckoning, the kinship system, marriage rules and so on, with a view to looking at both the linguistic expressions involved, and the gestures people use to indicate kinship groups and relations. Unlike the task in the 2001 manual, this task is a direct interview method.

Abstract

This task is a complement to the questionnaire ‘Multimodal interaction in your field site: a preliminary investigation’. The objective of the task is to obtain high quality video data on structured and symmetrical dyadic multimodal interaction. The features of interaction we are interested in include turn organization in speech and nonverbal behavior, eye-gaze behavior, use of composite signals (i.e. communicative units of speech-combined-with-gesture), and linguistic and other resources for ‘navigating’ interaction (e.g. words like okay, now, well, and um).

Abstract

This Field Manual entry has been superceded by the 2009 version: https://doi.org/10.17617/2.883564

Abstract

This chapter highlights the research in isolating genetic factors underlying specific language impairment (SLI), or developmental dysphasia, which exploits newly developed genotyping technology, novel statistical methodology, and DNA sequence data generated by the Human Genome Project. The author begins with an overview of results from family, twin, and adoption studies supporting genetic involvement and then goes on to outline progress in a number of genetic mapping efforts that have been recently completed or are currently under way. It has been possible for genetic researchers to pinpoint the specific mutation responsible for some speech and language disorders, providing an example of how the availability of human genomic sequence data can greatly accelerate the pace of disease gene discovery. Finally, the author discusses future prospects on how molecular genetics may offer new insight into the etiology underlying speech and language disorders, leading to improvements in diagnosis and treatment.

Abstract

Several sources of information are used in choosing the intended referent of an ambiguous pronoun. The two sources considered in this paper are foregrounding and context. The first refers to the accessibility of discourse entities. An entity that is foregrounded is more likely to become the pronoun’s referent than an entity that is not. Context information affects pronoun resolution when world knowledge is needed to find the referent. The model presented here simulates how world knowledge invoked by context, together with foregrounding, influences pronoun resolution. It was developed as an extension to the Distributed Situation Space (DSS) model of knowledge-based inferencing in story comprehension (Frank, Koppen, Noordman, & Vonk, 2003), which shall be introduced first.

Abstract

One of the most daunting tasks of a listener is to map a
continuous auditory stream onto known speech sound
categories and lexical items. A major issue with this mapping
problem is the variability in the acoustic realizations of sound
categories, both within and across speakers. Past research has
suggested listeners may use visual information (e.g., lipreading)
to calibrate these speech categories to the current
speaker. Previous studies have focused on audiovisual
recalibration of consonant categories. The present study
explores whether vowel categorization, which is known to show
less sharply defined category boundaries, also benefit from
visual cues.
Participants were exposed to videos of a speaker
pronouncing one out of two vowels, paired with audio that was
ambiguous between the two vowels. After exposure, it was
found that participants had recalibrated their vowel categories.
In addition, individual variability in audiovisual recalibration is
discussed. It is suggested that listeners’ category sharpness may
be related to the weight they assign to visual information in
audiovisual speech perception. Specifically, listeners with less
sharp categories assign more weight to visual information
during audiovisual speech recognition.

Abstract

A crucial aspect of everyday conversational interactions is our ability to establish and maintain common ground. Understanding the relevant mechanisms involved in such social coordination remains an important challenge for cognitive science. While common ground is often discussed in very general terms, different contexts of interaction are likely to afford different coordination mechanisms. In this paper, we investigate the presence and relation of three mechanisms of social coordination – backchannels, interactive alignment and conversational repair – across free and task-oriented conversations. We find significant differences: task-oriented conversations involve higher presence of repair – restricted offers in particular – and backchannel, as well as a reduced level of lexical and syntactic alignment. We find that restricted repair is associated with lexical alignment and open repair with backchannels. Our findings highlight the need to explicitly assess several mechanisms at once and to investigate diverse activities to understand their role and relations.

Abstract

This project is part of a collaborative project with the research group “Reciprocals across languages” led by Nick Evans. One goal of this project is to develop a typology of reciprocals. This questionnaire is designed to help field workers get an overview over the type of markers used in the expression of reciprocity in the language studied.

Abstract

We conduct the first systematic comparison of automated semantic
annotation based on either the full-text or only on the title metadata
of documents. Apart from the prominent text classification baselines
kNN and SVM, we also compare recent techniques of Learning
to Rank and neural networks and revisit the traditional methods
logistic regression, Rocchio, and Naive Bayes. Across three of our
four datasets, the performance of the classifications using only titles
reaches over 90% of the quality compared to the performance when
using the full-text.

Abstract

We assess the suitability of word embeddings for practical information retrieval scenarios. Thus, we assume that users issue ad-hoc short queries where we return the first twenty retrieved documents after applying a boolean matching operation between the query and the documents. We compare the performance of several techniques that leverage word embeddings in the retrieval models to compute the similarity between the query and the documents, namely word centroid similarity, paragraph vectors, Word Mover’s distance, as well as our novel inverse document frequency (IDF) re-weighted word centroid similarity. We evaluate the performance using the ranking metrics mean average precision, mean reciprocal rank, and normalized discounted cumulative gain. Additionally, we inspect the retrieval models’ sensitivity to document length by using either only the title or the full-text of the documents for the retrieval task. We conclude that word centroid similarity is the best competitor to state-of-the-art retrieval models. It can be further improved by re-weighting the word frequencies with IDF before aggregating the respective word vectors of the embedding. The proposed cosine similarity of IDF re-weighted word vectors is competitive to the TF-IDF baseline and even outperforms it in case of the news domain with a relative percentage of 15%.

Abstract

This paper reports on a multilingual investigation into the effects of different masker types on native and non-native perception in a VCV consonant recognition task. Native listeners outperformed 7 other language groups, but all groups showed a similar ranking of maskers. Strong first language (L1) interference was observed, both from the sound system and from the L1 orthography. Universal acoustic-perceptual tendencies are also at work in both native and non-native sound identifications in noise. The effect of linguistic distance, however, was less clear: in large multilingual studies, listener variables may overpower other factors.