Publications

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

Computer models of cultural evolution have shown language properties emerging on interacting agents with a brain that lacks dedicated, nativist language modules. Notably, models using Bayesian agents provide a precise specification of (extra-)liguististic factors (e.g., genetic) that shape language through iterated learning (biases on language), and demonstrate that weak biases get expressed more strongly over time (bias amplification). Other models attempt to lessen assumption on agents’ innate predispositions even more, and emphasize self-organization within agents, highlighting glossogenesis (the development of language from a nonlinguistic state). Ultimately however, one also has to recognize that biology and culture are strongly interacting, forming a coevolving system. As such, computer models show that agents might (biologically) evolve to a state predisposed to language adaptability, where (culturally) stable language features might get assimilated into the genome via Baldwinian niche construction. In summary, while many questions about language evolution remain unanswered, it is clear that it is not to be completely understood from a purely biological, cognitivist perspective. Language should be regarded as (partially) emerging on the social interactions between large populations of speakers. In this context, agent models provide a sound approach to investigate the complex dynamics of genetic biasing on language and speech

Abstract

The present paper describes the development of the languagesindanger.eu interactive website as an example of including multimedia language resources to disseminate knowledge and create educational material on less-resourced languages. The website is a product of INNET (Innovative networking in infrastructure for endangered languages), European FP7 project. Its main functions can be summarized as related to the three following areas: (1) raising students' awareness of language endangerment and arouse their interest in linguistic diversity, language maintenance and language documentation; (2) informing both students and teachers about these topics and show ways how they can enlarge their knowledge further with a special emphasis on information about language archives; (3) helping teachers include these topics into their classes. The website has been localized into five language versions with the intention to be accessible to both scientific and non-scientific communities such as (primarily) secondary school teachers and students, beginning university students of linguistics, journalists, the interested public, and also members of speech communities who speak minority languages

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

A wealth of studies has shown that more proficient monolingual speakers are better at predicting upcoming information during language comprehension. Similarly, prediction skills of adult second language (L2) speakers in their L2 have also been argued to be modulated by their L2 proficiency. How exactly language proficiency and prediction are linked, however, is yet to be systematically investigated. One group of language users which has the potential to provide invaluable insights into this link is bilingual children. In this paper, we compare bilingual children’s prediction skills with those of monolingual children and adult L2 speakers, and show how investigating bilingual children’s prediction skills may contribute to our understanding of how predictive processing works.

Abstract

There is a strong relation between children’s exposure to
spatial terms and their later memory accuracy. In the current
study, we tested whether the production of spatial terms by
children themselves predicts memory accuracy and whether
and how language modality of these encodings modulates
memory accuracy differently. Hearing child speakers of
Turkish and deaf child signers of Turkish Sign Language
described pictures of objects in various spatial relations to each
other and later tested for their memory accuracy of these
pictures in a surprise memory task. We found that having
described the spatial relation between the objects predicted
better memory accuracy. However, the modality of these
descriptions in sign, speech, or speech-plus-gesture did not
reveal differences in memory accuracy. We discuss the
implications of these findings for the relation between spatial
language, memory, and the modality of encoding.

Abstract

We describe a reaction time study in which listeners detected word or nonword syllable targets (e.g. zoo, trel) in sequences consisting of the target plus a consonant or syllable residue (trelsh, trelshek). The pattern of responses differed from an earlier word-spotting study with the same material, in which words were always harder to find if only a consonant residue remained. The earlier results should thus not be viewed in terms of syllabic parsing, but in terms of a universal role for syllables in speech perception; words which are accidentally present in spoken input (e.g. sell in self) can be rejected when they leave a residue of the input which could not itself be a word.

Abstract

In this paper we present a definition of Performance Grammar (PG), a psycholinguistically motivated syntax formalism, in declarative terms. PG aims not only at describing and explaining intuitive judgments and other data concerning the well–formedness of sentences of a language, but also at contributing to accounts of syntactic processing phenomena observable in language comprehension and language production. We highlight two general properties of human sentence generation, incrementality and late linearization,which make special demands on the design of grammar formalisms claiming psychological plausibility. In order to meet these demands, PG generates syntactic structures in a two-stage process. In the first and most important ‘hierarchical’ stage, unordered hierarchical structures (‘mobiles’) are assembled out of lexical building blocks. The key operation at work here is typed feature unification, which also delimits the positional options of the syntactic constituents in terms of so-called topological features. The second, much simpler stage takes care of arranging the branches of the mobile from left to right by ‘reading–out’ one positional option of every constituent. In this paper we concentrate on the structure assembly formalism in PG’s hierarchical component. We provide a declarative definition couched in an HPSG–style notation based on typed feature unification. Our emphasis throughout is on linear order constraints.

Abstract

Requesting and offering are closely related, insofar as they are activities associated with someone’s need for assistance. It has been supposed (e.g., Schegloff 2007) that requests and offers are not equivalent actions – specifically that offers are preferred actions and requests are dispreferred. We review the evidence for this claim across a corpus of requests and offers and demonstrate that the empirical evidence does not support the claim for a putative preference for offers over requests. Further consideration of the often symbiotic relationships between requesting and offering, particularly in face-to-face interactions, reveals a more complex picture of the ways in which people recruit others to help, or in which others are mobilized to help.

Abstract

The VALID Data Archive is an open multimedia data archive (under construction) with data from speakers suffering from language impairments. We report on a pilot project in the CLARIN-NL framework in which five data resources were curated. For all data sets concerned, written informed consent from the participants or their caretakers has been obtained. All materials were anonymized. The audio files were converted into wav (linear PCM) files and the transcriptions into CHAT or ELAN format. Research data that consisted of test, SPSS and Excel files were documented and converted into CSV files. All data sets obtained appropriate CMDI metadata files. A new CMDI metadata profile for this type of data resources was established and care was taken that ISOcat metadata categories were used to optimize interoperability. After curation all data are deposited at the Max Planck Institute for Psycholinguistics Nijmegen where persistent identifiers are linked to all resources. The content of the transcriptions in CHAT and plain text format can be searched with the TROVA search engine

Abstract

The distinction between finite and non-finite verb forms is well-established but not particularly well-defined. It cannot just be a matter of verb morphology, because it is also made when there is hardly any morphological difference: by far most English verb forms can be finite as well as non-finite. More importantly, many structural phenomena are clearly associated with the presence or absence of finiteness, a fact which is clearly reflected in the early stages of first and second language acquisition. In syntax, these include basic word order rules, gapping, the licensing of a grammatical subject and the licensing of expletives. In semantics, the specific interpretation of indefinite noun phrases is crucially linked to the presence of a finite element. These phenomena are surveyed, and it is argued that finiteness (a) links the descriptive content of the sentence (the 'sentence basis') to its topic component (in particular, to its topic time), and (b) it confines the illocutionary force to that topic component. In a declarative main clause, for example, the assertion is confined to a particular time, the topic time. It is shown that most of the syntactic and semantic effects connected to finiteness naturally follow from this assumption.

Abstract

It is generally assumed that verbs have an ‘argument structure’, which imposes various constraints on the noun phrases that can or must go with the verb, and an ‘event structure’, which characterises the particular temporal characteristics of the ‘event’ which the verb relates to: this event may be a state, a process, an activity, an ‘event in the narrow sense’, and others. In this paper, it is argued that that argument structure and event structure should be brought together. The lexical content of a verb assigns descriptive properties to one or more arguments at one or more times, hence verbs have an ‘argument time-structure’ (AT-structure). Numerous morphological and syntactical operations, such as participle formation or complex verb constructions, modify this AT-structure. This is illustrated with German recipient constructions such as ein Buch geschenkt bekommen or das Fenster geöffnet kriegen.

Abstract

Voice assimilation in Dutch is an optional phonological rule which changes the surface forms of words and in doing so may violate the otherwise obligatory phonological rule of syllablefinal devoicing. We report two experiments examining the influence of voice assimilation on phoneme processing, in lexical compound words and in noun-verb phrases. Processing was not impaired in appropriate assimilation contexts across morpheme boundaries, but was impaired when devoicing was violated (a) in an inappropriate non-assimilatory) context, or (b) across a syntactic boundary.

Abstract

Like Einstein's general theory of relativity is concerned with explaining the basics of an observable experience – i.e., gravity – most people take for granted that Chomsky's theory of generative grammar (GG) is concerned with the basic nature of language. This chapter highlights a mere subset of central constructs in GG, showing how they have featured prominently and thus shaped formal linguistic studies in multilingualism. Because multilingualism includes a wide range of nonmonolingual populations, the constructs are divided across child bilingualism and adult third language for greater coverage. In the case of the former, the chapter examines how poverty of the stimulus has been investigated. Using the nascent field of L3/Ln acquisition as the backdrop, it discusses how the GG constructs of I-language versus E-language sit at the core of debates regarding the very notion of what linguistic transfer and mental representations should be taken to be.

Abstract

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

Abstract

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

Abstract

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

Abstract

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

Abstract

Evolution and changes of all modern languages is a wellknown fact. However, recently it is reaching dynamics never seen before, which results in loss of the vast amount of information encoded in every language. In order to preserve such rich heritage, and to carry out linguistic research, properly annotated recordings of world languages are necessary. Since creating those annotations is a very laborious task, reaching times 100 longer than the length of the annotated media, innovative video processing algorithms are needed, in order to improve the efficiency and quality of annotation process. This is the scope of the AVATecH project presented in this article

Abstract

People's language is shaped by the input from the environment. The environment, however, offers a range of linguistic inputs that differ in their reliability. We test whether listeners accordingly weigh input from sources that differ in reliability differently. Using a perceptual learning paradigm, we show that listeners adjust their representations according to linguistic input provided by native but not by non-native speakers. This is despite the fact that listeners are able to learn the characteristics of the speech of both speakers. These results provide evidence for a disassociation between adaptation to the characteristic of specific speakers and adjustment of linguistic representations in general based on these learned characteristics. This study also has implications for theories of language change. In particular, it cast doubts on the hypothesis that a large proportion of non-native speakers in a community can bring about linguistic changes

Abstract

There has been a recent spate of work on recursion as a central design feature of language and specifically of syntax. This short report points out that there is little evidence that unlimited recursion, understood as centre embedding, is typical of natural language syntax. Nevertheless, embedded pragmatic construals seem available in every language. Further, much deeper centre embedding can be found in dialogue or conversation structure than can be found in syntax. Existing accounts for the ‘performance’ limitations on centre embedding are thus thrown in doubt. Dialogue materials suggest that centre embedding is perhaps a core part of the human interaction system and is for some reason much more highly restricted in syntax than in other aspects of cognition

Abstract

This study investigates prosodic marking of focus in Bai, a Sino-Tibetan language spoken in the Southwest of China, by adopting a semi-spontaneous experimental approach. Our data show that Bai speakers increase the duration of the focused constituent and reduce the duration of the post-focus constituent to encode focus. However, duration is not used in Bai to distinguish focus types differing in size and contrastivity. Further, pitch plays no role in signaling focus and differentiating focus types. The results thus suggest that Bai uses prosody to mark focus, but to a lesser extent, compared to Mandarin Chinese, with which Bai has been in close contact for centuries, and Cantonese, to which Bai is similar in the tonal system, although Bai is similar to Cantonese in its reliance on duration in prosodic focus marking.

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

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

What is the relationship between the language we speak and the way we think? Researchers working at the interface of language and cognition hope to understand the complex interplay between linguistic structures and the way the mind works. This is thorny territory in anthropology and its closely allied disciplines, such as linguistics and psychology.

Abstract

Readers experience a number of sensations during reading. They do
not – or do not only – process words and sentences in a detached, abstract
manner. Instead they “perceive” what they read about. They see descriptions of
scenery, feel what characters feel, and hear the sounds in a story. These sensa-
tions tend to be grouped under the umbrella terms “mental simulation” and
“mental imagery.” This chapter provides an overview of empirical research on
the role of mental simulation during literary reading. Our chapter also discusses
what mental simulation is and how it relates to mental imagery. Moreover, it
explores how mental simulation plays a role in leading models of literary read-
ing and investigates under what circumstances mental simulation occurs dur-
ing literature reading. Finally, the effect of mental simulation on the literary
reader’s experience is discussed, and suggestions and unresolved issues in this
field are formulated.

Abstract

Visual and auditory channels have different affordances and
this is mirrored in what information is available for linguistic
encoding. The visual channel has high spatial acuity, whereas
the auditory channel has better temporal acuity. These
differences may lead to different conceptualizations of events
and affect multimodal language production. Previous studies of
motion events typically present visual input to elicit speech and
gesture. The present study compared events presented as audio-
only, visual-only, or multimodal (visual+audio) input and
assessed speech and co-speech gesture for path and manner of
motion in Turkish. Speakers with audio-only input mentioned
path more and manner less in verbal descriptions, compared to
speakers who had visual input. There was no difference in the
type or frequency of gestures across conditions, and gestures
were dominated by path-only gestures. This suggests that input
modality influences speakers’ encoding of path and manner of
motion events in speech, but not in co-speech gestures.

Abstract

Models are a flourishing and indispensable area of research in language evolution. Here we
highlight critical issues in using and interpreting models, and suggest viable approaches. First,
contrasting models can explain the same data and similar modelling techniques can lead to
diverging conclusions. This should act as a reminder to use the extreme malleability of
modelling parsimoniously when interpreting results. Second, quantitative techniques similar to
those used in modelling language evolution have proven themselves inadequate in other
disciplines. Cross-disciplinary fertilization is crucial to avoid mistakes which have previously
occurred in other areas. Finally, experimental validation is necessary both to sharpen models'
hypotheses, and to support their conclusions. Our belief is that models should be interpreted as
quantitative demonstrations of logical possibilities, rather than as direct sources of evidence.
Only an integration of theoretical principles, quantitative proofs and empirical validation can
allow research in the evolution of language to progress.

Abstract

Two Siberian languages, Tundra Nenets and Tundra Yukaghir, do not obey strong island constraints in questioning: any sub-constituent of a relative or adverbial clause can be questioned. We argue that this has to do with how focusing works in these languages. The focused sub-constituent remains in situ, but there is abundant morphosyntactic evidence that the focus feature is passed up to the head of the clause. The result is the formation of a complex focus structure in which both the head and non head daughter are overtly marked as focus, and they are interpreted as a pairwise list such that the focus background is applicable to this list, but not to other alternative lists

Abstract

This volume is dedicated to exploring the crossroads where complex sentences and information management – more specifically information structure (IS) and reference tracking (RT) – come together. Complex sentences are a highly relevant but understudied domain for studying notions of IS and RT. On the one hand, a complex sentence can be studied as a mini-unit of discourse consisting of two or more elements describing events, situations, or processes, with its own internal information-structural and referential organisation. On the other hand, complex sentences can be studied as parts of larger discourse structures, such as narratives or conversations, in terms of how their information-structural characteristics relate to this wider context.We first focus on the interrelatedness of IS and RT (Section 1) and then define and discuss the notion of complex sentences and their subtypes in Section 2. Section 3 surveys issues of IS in complex sentences, while Section 4 focuses on RT in complex sentences. Sections 5 and 6 briefly consider IS and RT in a wider discourse context. Section 5 discusses the interaction between IS, RT, and other discourse factors, and Section 6 focuses on ways in which a specific RT system, switch reference, can function as an RT device beyond the sentence.

Abstract

No island effects are observable in Tundra Yukaghir questions, which are possible in virtually all syntactic environments. It is argued that this feature of Tundra Yukaghir relates to its capability of explicitly marking focus domains. If a question word occurs in a syntactic island, the whole island is morphologically treated as a focus domain. In order to take scope and function as question markers, question words must remain within the focus domain, i.e. in the island clause. This syntactic configuration is reflected in the semantics of question islands, which are used to inquire about the identity of the whole island, not merely the denotation of the question word.

Abstract

This paper presents an analysis of semantic association norms for German nouns. In contrast to prior studies, we not only collected associations elicited by written representations of target objects but also by their pictorial representations. In a first analysis, we identified systematic differences in the type and distribution of associate responses for the two presentation forms. In a second analysis, we applied a soft cluster analysis to the collected target-response pairs. We subsequently used the clustering to predict noun ambiguity and to discriminate senses in our target nouns.

Abstract

Recurrent neural networks (RNNs) have long been an architecture of interest for computational models of human sentence processing. The recently introduced Transformer architecture outperforms RNNs on many natural language processing tasks but little is known about its ability to model human language processing. We compare Transformer- and RNN-based language models’ ability to account for measures of human reading effort. Our analysis shows Transformers to outperform RNNs in explaining self-paced reading times and neural activity during reading English sentences, challenging the widely held idea that human sentence processing involves recurrent and immediate processing and provides evidence for cue-based retrieval.

Abstract

This study addresses the question whether visually grounded speech recognition (VGS) models learn to capture sentence semantics without access to any prior linguistic knowledge. We produce synthetic and natural spoken versions of a well known semantic textual similarity database and show that our VGS model produces embeddings that correlate well with human semantic similarity judgements. Our results show that a model trained on a small image-caption database outperforms two models trained on much larger databases, indicating that database size is not all that matters. We also investigate the importance of having multiple captions per image and find that this is indeed helpful even if the total number of images is lower, suggesting that paraphrasing is a valuable learning signal. While the general trend in the field is to create ever larger datasets to train models on, our findings indicate other characteristics of the database can just as important.

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

In this paper, we present two attempts to replicate a widely-cited but never fully published experiment in which German and Spanish speakers were asked to associate adjectives with nouns of masculine and feminine grammati-cal gender (Boroditsky et al. 2003). The researchers claim that speakers associ-ated more stereotypically female adjectives with grammatically feminine nouns and more stereotypically male adjectives with grammatically masculine nouns. We were not able to replicate the results either in a word association task or in an analogous primed lexical decision task. This suggests that the results of the original experiment were either an artifact of some non-documented aspect of the experimental procedure or a statistical fluke. The question whether speakers assign sex-based interpretations to grammatical gender categories at all cannot be answered definitively, as the results in the published literature vary consider-ably. However, our experiments show that if such an effect exists, it is not strong enough to be measured indirectly via the priming of adjectives by nouns.

Abstract

The goal of speech perception is understanding a speaker's message. To achieve this, listeners must recognize the words that comprise a spoken utterance. This in turn implies distinguishing these words from other minimally different words (e.g., word from bird, etc.), and this involves making phonemic distinctions. The article summarizes research on the perception of phonemic distinctions, on how listeners cope with the continuity and variability of speech signals, and on how phonemic information is mapped onto the representations of words. Particular attention is paid to theories of speech perception and word recognition.