Publications

Abstract

Language is a social phenomenon. The author learns, processes, and uses it in social contexts. In other words, the social environment shapes the linguistic knowledge and use of the knowledge. To a degree, this is trivial. A child exposed to Japanese will become fluent in Japanese, whereas a child exposed to only Spanish will not understand Japanese but will master the sounds, vocabulary, and grammar of Spanish. Language is a structured system. Sounds and words do not occur randomly but are characterized by regularities. Learners are sensitive to these regularities and exploit them when learning language. People differ in the sizes of their social networks. Some people tend to interact with only a few people, whereas others might interact with a wide range of people. This is reflected in people’s holiday greeting habits: some people might send cards to only a few people, whereas other would send greeting cards to more than 350 people.

Abstract

How can one conceive of the neuronal implementation of the processing model we proposed in our target article? In his commentary (Pulvermüller 1999, reprinted here in this issue), Pulvermüller makes various proposals concerning the underlying neural mechanisms and their potential localizations in the brain. These proposals demonstrate the compatibility of our processing model and current neuroscience. We add further evidence on details of localization based on a recent meta-analysis of neuroimaging studies of word production (Indefrey & Levelt 2000). We also express some minor disagreements with respect to Pulvermüller’s interpretation of the “lemma” notion, and concerning his neural modeling of phonological code retrieval. Branigan & Pickering discuss important aspects of syntactic encoding, which was not the topic of the target article. We discuss their well-taken proposal that multiple syntactic frames for a single verb lemma are represented as independent nodes, which can be shared with other verbs, such as accounting for syntactic priming in speech production. We also discuss how, in principle, the alternative multiple-frame-multiplelemma account can be tested empirically. The available evidence does not seem to support that account.

Abstract

During the second half of the 19th century, the psychology of language was invented as a discipline for the sole purpose of explaining the evolution of spoken language. These efforts culminated in Wilhelm Wundt’s monumental Die Sprache of 1900, which outlined the psychological mechanisms involved in producing utterances and considered how these mechanisms could have evolved. Wundt assumes that articulatory movements were originally rather arbitrary concomitants of larger, meaningful expressive bodily gestures. The sounds such articulations happened to produce slowly acquired the meaning of the gesture as a whole, ultimately making the gesture superfluous. Over a century later, gestural theories of language origins still abound. I argue that such theories are unlikely and wasteful, given the biological, neurological and genetic evidence.

Abstract

The purpose of this study was to relate a psycholinguistic processing model of picture naming to the dynamics of cortical activation during picture naming. The activation was recorded from eight Dutch subjects with a whole-head neuromagnetometer. The processing model, based on extensive naming latency studies, is a stage model. In preparing a picture's name, the speaker performs a chain of specific operations. They are, in this order, computing the visual percept, activating an appropriate lexical concept, selecting the target word from the mental lexicon, phonological encoding, phonetic encoding, and initiation of articulation. The time windows for each of these operations are reasonably well known and could be related to the peak activity of dipole sources in the individual magnetic response patterns. The analyses showed a clear progression over these time windows from early occipital activation, via parietal and temporal to frontal activation. The major specific findings were that (1) a region in the left posterior temporal lobe, agreeing with the location of Wernicke's area, showed prominent activation starting about 200 msec after picture onset and peaking at about 350 msec, (i.e., within the stage of phonological encoding), and (2) a consistent activation was found in the right parietal cortex, peaking at about 230 msec after picture onset, thus preceding and partly overlapping with the left temporal response. An interpretation in terms of the management of visual attention is proposed.

Abstract

The evolution of speech and language has been a returning topic in the language sciences since the so-called “cognitive revolution.”

Abstract

The topic of this paper is the way in which speakers order information in discourse. I will refer to this issue with the term "linearization", and will begin with two types of general remarks. The first one concerns the scope and relevance of the problem with reference to some existing literature. The second set of general remarks will be about the place of linearization in a theory of the speaker. The following, and main part of this paper, will be a summary report of research of linearization in a limited, but well-defined domain of discourse, namely the description of spatial networks.

Abstract

This commentary discusses whether abstract metrical frames are stored. For stress-assigning languages (e.g., Dutch and English), which have a dominant stress pattern, metrical frames are stored only for words that deviate from the default stress pattern. The majority of the words in these languages are produced without retrieving any independent syllabic or metrical frame.

Abstract

This paper considers the AGL literature from a psycholinguistic perspective. It first presents a taxonomy of the experimental familiarization test procedures used, which is followed by a consideration of shortcomings and potential improvements of the empirical methodology. It then turns to reconsidering the issue of grammar learning from the point of view of acquiring constraints, instead of the traditional AGL approach in terms of acquiring sets of rewrite rules. This is, in particular, a natural way of handling long‐distance dependences. The final section addresses an underdeveloped issue in the AGL literature, namely how to detect latent hierarchical structure in AGL response patterns.

Abstract

It is argued that the dominantly successive nature of language is largely mode-independent and holds equally for sign and for spoken language. A preliminary distinction is made between what is simultaneous or successive in the signal, and what is in the process; these need not coincide, and it is the successiveness of the process that is at stake. It is then discussed extensively for the word/sign level, and in a more preliminary fashion for the clause and discourse level that online processes are parallel in that they can simultaneously draw on various sources of knowledge (syntactic, semantic, pragmatic), but successive in that they can work at the interpretation of only one unit at a time. This seems to hold for both sign and spoken language. In the final section, conjectures are made about possible evolutionary explanations for these properties of language processing.

Abstract

Speakers tend to repeat materials from previous talk. This tendency is experimentally established and manipulated in various question-answering situations. It is shown that a question's surface form can affect the format of the answer given, even if this form has little semantic or conversational consequence, as in the pair Q: (At) what time do you close. A: “(At)five o'clock.” Answerers tend to match the utterance to the prepositional (nonprepositional) form of the question. This “correspondence effect” may diminish or disappear when, following the question, additional verbal material is presented to the answerer. The experiments show that neither the articulatory buffer nor long-term memory is normally involved in this retention of recent speech. Retaining recent speech in working memory may fulfill a variety of functions for speaker and listener, among them the correct production and interpretation of surface anaphora. Reusing recent materials may, moreover, be more economical than regenerating speech anew from a semantic base, and thus contribute to fluency. But the realization of this strategy requires a production system in which linguistic formulation can take place relatively independent of, and parallel to, conceptual planning.

Abstract

The core issue in the 19-century sources of psycholinguistics was the question, "Where does language come from?'' This genetic perspective unified the study of the ontogenesis, the phylogenesis, the microgenesis, and to some extent the neurogenesis of language. This paper makes the point that this original perspective is still a valid and attractive one. It is exemplified by a discussion of the genesis of spoken words.

Abstract

Survey article

Abstract

Philosophers, psychologists, and linguists have argued that spatial conception is pivotal to cognition in general, providing a general, egocentric, and universal framework for cognition as well as metaphors for conceptualizing many other domains. But in an aboriginal community in Northern Queensland, a system of cardinal directions informs not only language, but also memory for arbitrary spatial arrays and directions. This work suggests that fundamental cognitive parameters, like the system of coding spatial locations, can vary cross-culturally, in line with the language spoken by a community. This opens up the prospect of a fruitful dialogue between anthropology and the cognitive sciences on the complex interaction between cultural and universal factors in the constitution of mind.

Abstract

This Element tries to discern the known unknowns in the field
of pragmatics, the ‘Dark Matter’ of the title. We can identify a key
bottleneck in human communication, the sheer limitation on the speed
of speech encoding: pragmatics occupies the niche nestled between
slow speech encoding and fast comprehension. Pragmatic strategies
are tricks for evading this tight encoding bottleneck by meaning more
than you say. Five such tricks are reviewed, which are all domains where
we have made considerable progress. We can then ask for each of these
areas, where have we neglected to push the frontier forward? These are
the known unknowns of pragmatics, key areas, and topics for future
research. The Element thus offers a brief review of some central areas of
pragmatics, and a survey of targets for future research.

Abstract

Cognitive anthropology is in need of a theory that extends beyond cultural model theory and explains both how culture has transformed human cognition and the curious ontology of culture itself, for, as Durkheim insisted, culture cannot be reduced to psychology. This chapter promotes a framework that deals with both the evolutionary question and the ontological problem. It is argued that at least a central part of culture should be conceived of in terms of cognitive technology. Beginning with obvious examples of cognitive artifacts, like those used in measurement, way-finding, time-reckoning and numerical calculation, the chapter goes on to consider extensions to our communication systems, emotion-modulating systems and the cognitive division of labor. Cognitive artifacts form ‘coupled systems’ that amplify individual psychology, lying partly outside the head, and are honed by cultural evolution. They make clear how culture gave human cognition an evolutionary edge.

Abstract

The present paper discusses the benefits and challenges of token-based typology, which takes into account the frequencies of words and constructions in language use. This approach makes it possible to introduce new criteria for language classification, which would be difficult or impossible to achieve with the traditional, type-based approach. This point is illustrated by several quantitative studies of word order variation, which can be measured as entropy at different levels of granularity. I argue that this variation can be explained by general functional mechanisms and pressures, which manifest themselves in language use, such as optimization of processing (including avoidance of ambiguity) and grammaticalization of predictable units occurring in chunks. The case studies are based on multilingual corpora, which have been parsed using the Universal Dependencies annotation scheme.

Abstract

Our study contributes to the less explored domain of lexical typology, focusing on semantic prosody and connotation. Semantic derogation, or pejoration of nouns referring to women, whereby such words acquire connotations and further denotations of social pejoration, immorality and/or loose sexuality, has been a very prominent question in studies on gender and language (change). It has been argued that pejoration emerges due to the general derogatory attitudes toward female referents. However, the evidence for systematic differences in connotations of female- vs. male-related words is fragmentary and often fairly impressionistic; moreover, many researchers argue that expressed sentiments toward women (as well as men) often are ambivalent. One should also expect gender differences in connotations to have decreased in the recent years, thanks to the advances of feminism and social progress. We test these ideas in a study of positive and negative connotations of feminine and masculine term pairs such as woman - man, girl - boy, wife - husband, etc. Sentences containing these words were sampled from diachronic corpora of English, Chinese and Russian, and sentiment scores for every word were obtained using two systems for Aspect-Based Sentiment Analysis: PyABSA, and OpenAI’s large language model GPT-3.5. The Generalized Linear Mixed Models of our data provide no indications of significantly more negative sentiment toward female referents in comparison with their male counterparts. However, some of the models suggest that female referents are more infrequently associated with neutral sentiment than male ones. Neither do our data support the hypothesis of the diachronic convergence between the genders. In sum, results suggest that pejoration is unlikely to be explained simply by negative attitudes to female referents in general.

Abstract

Schizophrenia is a common and complex mental disorder with neuroimaging alterations. Recent neuroanatomical pattern recognition studies attempted to distinguish individuals with schizophrenia by structural magnetic resonance imaging (sMRI) and diffusion tensor imaging (DTI). 1, 2 Applications of cutting-edge machine learning approaches in structural neuroimaging studies have revealed potential pathways to classification of schizophrenia based on regional gray matter volume (GMV) or density or cortical thickness. 3–5 Additionally, cortical folding may have high discriminatory value in correctly identifying symptom severity in schizophrenia. 6 Regional GMV and cortical thickness have also been combined in attempts to differentiate individuals with schizophrenia from healthy controls (HCs). 7 Applications of machine learning algorithms to diffusion imaging data analysis to predict individuals with first-episode schizophrenia (FES) have achieved encouraging accuracy. 8–10 White matter (WM) abnormalities in schizophrenia as estimated by DTI appear to be present in the early stage of the disorder, most likely reflecting the developmental stage of the sample of interest.

Abstract

Identifying data-driven biotypes of major depressive disorder (MDD) has promise for the clarification of diagnostic heterogeneity. However, few studies have focused on white-matter abnormalities for MDD subtyping. This study included 116 patients with MDD and 118 demographically-matched healthy controls assessed by diffusion tensor imaging and neurocognitive evaluation. Hierarchical clustering was applied to the major fiber tracts, in conjunction with tract-based spatial statistics, to reveal white-matter alterations associated with MDD. Clinical and neurocognitive differences were compared between identified subgroups and healthy controls. With fractional anisotropy extracted from 20 fiber tracts, cluster analysis revealed 3 subgroups based on the patterns of abnormalities. Patients in each subgroup versus healthy controls showed a stepwise pattern of white-matter alterations as follows: subgroup 1 (25.9% of patient sample), widespread white-matter disruption; subgroup 2 (43.1% of patient sample), intermediate and more localized abnormalities in aspects of the corpus callosum and left cingulate; and subgroup 3 (31.0% of patient sample), possible mild alterations, but no statistically significant tract disruption after controlling for family-wise error. The neurocognitive impairment in each subgroup accompanied the white-matter alterations: subgroup 1, deficits in sustained attention and delayed memory; subgroup 2, dysfunction in delayed memory; and subgroup 3, no significant deficits. Three subtypes of white-matter abnormality exist in individuals with major depression, those having widespread abnormalities suffering more neurocognitive impairments, which may provide evidence for parsing the heterogeneity of the disorder and help optimize type-specific treatment approaches.

Abstract

The past year has seen a steep rise in generative AI systems that claim to be open. But how open are they really? The question of what counts as open source in generative AI is poised to take on particular importance in light of the upcoming EU AI Act that regulates open source systems differently, creating an urgent need for practical openness assessment. Here we use an evidence-based framework that distinguishes 14 dimensions of openness, from training datasets to scientific and technical documentation and from licensing to access methods. Surveying over 45 generative AI systems (both text and text-to-image), we find that while the term open source is widely used, many models are `open weight' at best and many providers seek to evade scientific, legal and regulatory scrutiny by withholding information on training and fine-tuning data. We argue that openness in generative AI is necessarily composite (consisting of multiple elements) and gradient (coming in degrees), and point out the risk of relying on single features like access or licensing to declare models open or not. Evidence-based openness assessment can help foster a generative AI landscape in which models can be effectively regulated, model providers can be held accountable, scientists can scrutinise generative AI, and end users can make informed decisions.

Abstract

Measures of ‘humanness’, ‘coherence’ or ‘fluency’ are the mainstay of dialogue system evaluation, but they don’t target system capabilities and rarely offer actionable feedback. Reviewing recent work in this domain, we identify an opportunity for evaluation at the level of action sequences, rather than the more commonly targeted levels of whole conversations or single responses. We introduce interactive probes, an evaluation framework inspired by empirical work on social interaction that can help to systematically probe the capabilities of dialogue systems. We sketch some first probes in the domains of tellings and repair, two sequence types ubiquitous in human interaction and challenging for dialogue systems. We argue interactive probing can offer the requisite flexibility to keep up with developments in interactive language technologies and do justice to the open-endedness of action formation and ascription in interaction.

Abstract

Infants point for various motives. Classically, one such motive is declarative, to share attention and interest with adults to events. Recently, some researchers have questioned whether infants have this motivation. In the current study, an adult reacted to 12-month-olds' pointing in different ways, and infants' responses were observed. Results showed that when the adult shared attention and interest (i.e. alternated gaze and emoted), infants pointed more frequently across trials and tended to prolong each point – presumably to prolong the satisfying interaction. However, when the adult emoted to the infant alone or looked only to the event, infants pointed less across trials and repeated points more within trials – presumably in an attempt to establish joint attention. Results suggest that 12-month-olds point declaratively and understand that others have psychological states that can be directed and shared.

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

* These authors contributed equally/shared first author
Deep convolutional neural networks (DCNNs) are able to partially predict brain activity during object categorization tasks, but factors contributing to this predictive power are not fully understood. Our study aimed to investigate the factors contributing to the predictive power of DCNNs in object categorization tasks. We compared the activity of four DCNN architectures with EEG recordings obtained from 62 human participants during an object categorization task. Previous physiological studies on object categorization have highlighted the importance of figure-ground segregation—the ability to distinguish objects from their backgrounds. Therefore, we investigated whether figure-ground segregation could explain the predictive power of DCNNs. Using a stimulus set consisting of identical target objects embedded in different backgrounds, we examined the influence of object background versus object category within both EEG and DCNN activity. Crucially, the recombination of naturalistic objects and experimentally controlled backgrounds creates a challenging and naturalistic task, while retaining experimental control. Our results showed that early EEG activity (< 100 msec) and early DCNN layers represent object background rather than object category. We also found that the ability of DCNNs to predict EEG activity is primarily influenced by how both systems process object backgrounds, rather than object categories. We demonstrated the role of figure-ground segregation as a potential prerequisite for recognition of object features, by contrasting the activations of trained and untrained (i.e., random weights) DCNNs. These findings suggest that both human visual cortex and DCNNs prioritize the segregation of object backgrounds and target objects to perform object categorization. Altogether, our study provides new insights into the mechanisms underlying object categorization as we demonstrated that both human visual cortex and DCNNs care deeply about object background.

Abstract

This study aims to advance our understanding of the nature and source(s) of individual differences in pragmatic language behavior over the adult lifespan. Across four story continuation experiments, we probed adults’ (N = 496 participants, ages 18–82) choice of referential forms (i.e., names vs. pronouns to refer to the main character). Our manipulations were based on Fossard et al.’s (2018) scale of referential complexity which varies according to the visual properties of the scene: low complexity (one character), intermediate complexity (two characters of different genders), and high complexity (two characters of the same gender). Since pronouns signal topic continuity (i.e., that the discourse will continue to be about the same referent), the use of pronouns is expected to decrease as referential complexity increases. The choice of names versus pronouns, therefore, provides insight into participants’ perception of the topicality of a referent, and whether that varies by age and cognitive capacity. In Experiment 1, we used the scale to test the association between referential choice, aging, and cognition, identifying a link between older adults’ switching skills and optimal referential choice. In Experiments 2–4, we tested novel manipulations that could impact the scale and found both the timing of a competitor referent’s presence and emphasis placed on competitors modulated referential choice, leading us to refine the scale for future use. Collectively, Experiments 1–4 highlight what type of contextual information is prioritized at different ages, revealing older adults’ preserved sensitivity to (visual) scene complexity but reduced sensitivity to linguistic prominence cues, compared to younger adults.

Abstract

This study investigated how adults over the lifespan flexibly adapt their use of prosocial speech acts when conveying bad news to communicative partners. Experiment 1a (N = 100 Scottish adults aged 18–72 years) assessed whether participants’ use of prosocial speech acts varied according to audience design considerations (i.e., whether or not the recipient of the news was directly affected). Experiment 1b (N = 100 Scottish adults aged 19–70 years) assessed whether participants adjusted for whether the bad news was more or less severe (an index of general knowledge). Younger adults displayed more flexible adaptation to the recipient manipulation, while no age differences were found for severity. These findings are consistent with prior work showing age-related decline in audience design but not in the use of general knowledge during language production. Experiment 2 further probed younger adults (N = 40, Scottish, aged 18–37 years) and older adults’ (N = 40, Scottish, aged 70–89 years) prosocial linguistic behavior by investigating whether health (vs. nonhealth-related) matters would affect responses. While older adults used prosocial speech acts to a greater extent than younger adults, they did not distinguish between conditions. Our results suggest that prosocial linguistic behavior is likely influenced by a combination of differences in audience design and communicative styles at different ages. Collectively, these findings highlight the importance of situating prosocial speech acts within the pragmatics and aging literature, allowing us to uncover the factors modulating prosocial linguistic behavior at different developmental stages.

Abstract

Speakers and listeners show an informativity bias in the use and interpretation of color modifiers. For example, speakers use color more often when referring to objects that vary in color than to objects with a prototypical color. Likewise, listeners look away from objects with prototypical colors upon hearing that color mentioned. Here we test whether speakers and listeners account for another factor related to informativity: the strength of the association between lexical categories and color. Our results demonstrate that speakers and listeners' choices are indeed influenced by this factor; as such, it should be integrated into current pragmatic theories of informativity and computational models of color reference.

Abstract

Attention-deficit/hyperactivity disorder (ADHD) and reading disability (RD) are common highly heritable disorders of childhood, which frequently co-occur. Data from twin and family studies suggest that this overlap is, in part, due to shared genetic underpinnings. Here, we report the first genome-wide linkage analysis of measures of reading ability in children with ADHD, using a sample of 233 affected sibling pairs who previously participated in a genome-wide scan for susceptibility loci in ADHD. Quantitative trait locus (QTL) analysis of a composite reading factor defined from three highly correlated reading measures identified suggestive linkage (multipoint maximum lod score, MLS>2.2) in four chromosomal regions. Two regions (16p, 17q) overlap those implicated by our previous genome-wide scan for ADHD in the same sample: one region (2p) provides replication for an RD susceptibility locus, and one region (10q) falls approximately 35 cM from a modestly highlighted region in an independent genome-wide scan of siblings with ADHD. Investigation of an individual reading measure of Reading Recognition supported linkage to putative RD susceptibility regions on chromosome 8p (MLS=2.4) and 15q (MLS=1.38). Thus, the data support the existence of genetic factors that have pleiotropic effects on ADHD and reading ability--as suggested by shared linkages on 16p, 17q and possibly 10q--but also those that appear to be unique to reading--as indicated by linkages on 2p, 8p and 15q that coincide with those previously found in studies of RD. Our study also suggests that reading measures may represent useful phenotypes in ADHD research. The eventual identification of genes underlying these unique and shared linkages may increase our understanding of ADHD, RD and the relationship between the two.

Abstract

The lack of diversity in the language sciences has increasingly been criticized as it holds the potential for producing flawed theories. Research on (i) geographically diverse language communities and (ii) on sign languages is necessary to corroborate, sharpen, and extend existing theories. This study contributes a case study of adapting a well-established paradigm to study the acquisition of sign phonology in Kata Kolok, a sign language of rural Bali, Indonesia. We conducted an experiment modeled after the familiarization paradigm with child signers of Kata Kolok. Traditional analyses of looking time did not yield significant differences between signing and non-signing children. Yet, additional behavioral analyses (attention, eye contact, hand behavior) suggest that children who are signers and those who are non-signers, as well as those who are hearing and those who are deaf, interact differently with the task. This study suggests limitations of the paradigm due to the ecology of sign languages and the sociocultural characteristics of the sample, calling for a mixed-methods approach. Ultimately, this paper aims to elucidate the diversity of adaptations necessary for experimental design, procedure, and analysis, and to offer a critical reflection on the contribution of similar efforts and the diversification of the field.

Abstract

Minimal expressions are at the heart of interaction: Interjections like "Huh?" and "Mhm" keep conversations flowing by establishing and reinforcing intersubjectivity among interlocutors. Crosslinguistic research has identified that similar interactional pressures can yield structurally similar words (e.g., to initiate repair across languages). While crosslinguistic comparisons that include signed languages remain uncommon, recent work has revealed similarities in discourse management strategies among signers and speakers that share much of their cultural background. This study contributes a crossmodal comparison of repair initiators and continuers in speakers of English and signers of British Sign Language (BSL). We combine qualitative and quantitative analyses of data from sixteen English speakers and sixteen BSL signers, resulting in the following: First, the interactional infrastructure drawn upon by speakers and signers overwhelmingly relies on behaviors of the head, face, and body; these are used alone or sometimes in combination with verbal elements (i.e., spoken words or manual signs), while verbal strategies alone are rare. Second, discourse management strategies are remarkably similar in form across the two languages: A held eye gaze or freeze-look is the predominant repair initiator and head nodding the main continuer. These results suggest a modality-agnostic preference for visual strategies that do not occupy the primary articulators, one that we propose is founded in recipiency; people maintain the flow of communication following principles of minimal effort and minimal interruption.

Abstract

Nyelv és/vagy evolúció: Lehetséges-e a nyelv evolúciós magyarázata? [Derek Bickerton: Nyelv és evolúció] (Magyari Lilla); Történelmi olvasókönyv az agyról [Charles G. Gross: Agy, látás, emlékezet. Mesék az idegtudomány történetéből] (Garab Edit Anna); Művészet vagy tudomány [Margitay Tihamér: Az érvelés mestersége. Érvelések elemzése, értékelése és kritikája] (Zemplén Gábor); Tényleg ésszerűek vagyunk? [Herbert Simon: Az ésszerűség szerepe az emberi életben] (Kardos Péter); Nemi különbségek a megismerésben [Doreen Kimura: Női agy, férfi agy]. (Hahn Noémi);

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

Spoken language comprehension requires abstraction of linguistic information from speech, but the interaction between auditory and linguistic processing of speech remains poorly understood. Here, we investigate the nature of this abstraction using neural responses recorded intracranially while participants listened to conversational English speech. Capitalizing on multiple, language-specific patterns where phonological and acoustic information diverge, we demonstrate the causal efficacy of the phoneme as a unit of analysis and dissociate the unique contributions of phonemic and spectrographic information to neural responses. Quantitive higher-order response models also reveal that unique contributions of phonological information are carried in the covariance structure of the stimulus-response relationship. This suggests that linguistic abstraction is shaped by neurobiological mechanisms that involve integration across multiple spectro-temporal features and prior phonological information. These results link speech acoustics to phonology and morphosyntax, substantiating predictions about abstractness in linguistic theory and providing evidence for the acoustic features that support that abstraction.

Abstract

The human body is unique: it is both an object of perception and the source of human experience. Its universality makes it a perfect resource for asking questions about how cognition, brain and typology relate to one another. For example, we can ask how speakers of different languages segment and categorize the human body. A dominant view is that body parts are “given” by visual perceptual discontinuities, and that words are merely labels for these visually determined parts (e.g., Andersen, 1978; Brown, 1976; Lakoff, 1987). However, there are problems with this view. First it ignores other perceptual information, such as somatosensory and motoric representations. By looking at the neural representations of sesnsory representations, we can test how much of the categorization of the human body can be done through perception alone. Second, we can look at language typology to see how much universality and variation there is in body-part categories. A comparison of a range of typologically, genetically and areally diverse languages shows that the perceptual view has only limited applicability (Majid, Enfield & van Staden, in press). For example, using a “coloring-in” task, where speakers of seven different languages were given a line drawing of a human body and asked to color in various body parts, Majid & van Staden (in prep) show that languages vary substantially in body part segmentation. For example, Jahai (Mon-Khmer) makes a lexical distinction between upper arm, lower arm, and hand, but Lavukaleve (Papuan Isolate) has just one word to refer to arm, hand, and leg. This shows that body part categorization is not a straightforward mapping of words to visually determined perceptual parts.

Abstract

Many studies in cognitive science address how people categorize objects, but there has been comparatively little research on event categorization. This study investigated the categorization of events involving material destruction, such as “cutting” and “breaking”. Speakers of 28 typologically, genetically, and areally diverse languages described events shown in a set of video-clips. There was considerable cross-linguistic agreement in the dimensions along which the events were distinguished, but there was variation in the number of categories and the placement of their boundaries.

Abstract

Frames of reference are coordinate systems used to compute and specify the location of objects with respect to other objects. These have long been thought of as innate concepts, built into our neurocognition. However, recent work shows that the use of such frames in language, cognition and gesture varies crossculturally, and that children can acquire different systems with comparable ease. We argue that language can play a significant role in structuring, or restructuring, a domain as fundamental as spatial cognition. This suggests we need to rethink the relation between the neurocognitive underpinnings of spatial cognition and the concepts we use in everyday thinking, and, more generally, to work out how to account for cross-cultural cognitive diversity in core cognitive domains.

Abstract

People engage in simulation when reading literary narratives. In this study, we tried to pinpoint how different kinds of simulation (perceptual and motor simulation, mentalising) affect reading behaviour. Eye-tracking (gaze durations, regression probability) and questionnaire data were collected from 102 participants, who read three literary short stories. In a pre-test, 90 additional participants indicated which parts of the stories were high in one of the three kinds of simulation-eliciting content. The results show that motor simulation reduces gaze duration (faster reading), whereas perceptual simulation and mentalising increase gaze duration (slower reading). Individual differences in the effect of simulation on gaze duration were found, which were related to individual differences in aspects of story world absorption and story appreciation. These findings suggest fundamental differences between different kinds of simulation and confirm the role of simulation in absorption and appreciation.

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

Prediction of upcoming input is thought to be a main characteristic of language processing (e.g. Altmann & Mirkovic, 2009; Dell & Chang, 2014; Federmeier, 2007; Ferreira & Chantavarin, 2018; Pickering & Gambi, 2018; Hale, 2001; Hickok, 2012; Huettig 2015; Kuperberg & Jaeger, 2016; Levy, 2008; Norris, McQueen, & Cutler, 2016; Pickering & Garrod, 2013; Van Petten & Luka, 2012). One of the main pillars of experimental support for this notion comes from studies that have attempted to measure electrophysiological markers of prediction when participants read or listened to sentences ending in highly predictable words. The N400, a negative-going and centro-parietally distributed component of the ERP occurring approximately 400ms after (target) word onset, has been frequently interpreted as indexing prediction of the word (or the semantic representations and/or the phonological form of the predicted word, see Kutas & Federmeier, 2011; Nieuwland, 2019; Van Petten & Luka, 2012; for review). A major difficulty for interpreting N400 effects in language processing however is that it has been difficult to establish whether N400 target word modulations conclusively reflect prediction rather than (at least partly) ease of integration. In the present exploratory study, we attempted to distinguish lexical prediction (i.e. ‘top-down’ activation) from lexical integration (i.e. ‘bottom-up’ activation) accounts of ERP N400 modulations in language processing.

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

Human thought and language have extraordinary expressive power because meaningful parts can be assembled into more complex semantic structures. This partly underlies our ability to compose meanings into endlessly novel configurations, and sets us apart from other species and current computing devices. Crucially, human behaviour, including language use and linguistic data, indicates that composing parts into complex structures does not threaten the existence of constituent parts as independent units in the system: parts and wholes exist simultaneously yet independently from one another in the mind and brain. This independence is evident in human behaviour, but it seems at odds with what is known about the brain's exquisite sensitivity to statistical patterns: everyday language use is productive and expressive precisely because it can go beyond statistical regularities. Formal theories in philosophy and linguistics explain this fact by assuming that language and thought are compositional: systems of representations that separate a variable (or role) from its values (fillers), such that the meaning of a complex expression is a function of the values assigned to the variables. The debate on whether and how compositional systems could be implemented in minds, brains and machines remains vigorous. However, it has not yet resulted in mechanistic models of semantic composition: how, then, are the constituents of thoughts and sentences put and held together? We review and discuss current efforts at understanding this problem, and we chart possible routes for future research.

Abstract

Neither neurobiological nor process models of meaning composition specify the operator through which constituent parts are bound together into compositional structures. In this paper, we argue that a neurophysiological computation system cannot achieve the compositionality exhibited in human thought and language if it were to rely on a multiplicative operator to perform binding, as the tensor product (TP)-based systems that have been widely adopted in cognitive science, neuroscience and artificial intelligence do. We show via simulation and two behavioural experiments that TPs violate variable-value independence, but human behaviour does not. Specifically, TPs fail to capture that in the statements fuzzy cactus and fuzzy penguin, both cactus and penguin are predicated by fuzzy(x) and belong to the set of fuzzy things, rendering these arguments similar to each other. Consistent with that thesis, people judged arguments that shared the same role to be similar, even when those arguments themselves (e.g., cacti and penguins) were judged to be dissimilar when in isolation. By contrast, the similarity of the TPs representing fuzzy(cactus) and fuzzy(penguin) was determined by the similarity of the arguments, which in this case approaches zero. Based on these results, we argue that neural systems that use TPs for binding cannot approximate how the human mind and brain represent compositional information during processing. We describe a contrasting binding mechanism that any physiological or artificial neural system could use to maintain independence between a role and its argument, a prerequisite for compositionality and, thus, for instantiating the expressive power of human thought and language in a neural system.

Abstract

Humans learn to represent complex structures (e.g. natural language, music, mathematics) from experience with their environments. Often such structures are latent, hidden, or not encoded in statistics about sensory representations alone. Accounts of human cognition have long emphasized the importance of structured representations, yet the majority of contemporary neural networks do not learn structure from experience. Here, we describe one way that structured, functionally symbolic representations can be instantiated in an artificial neural network. Then, we describe how such latent structures (viz. predicates) can be learned from experience with unstructured data. Our approach exploits two principles from psychology and neuroscience: comparison of representations, and the naturally occurring dynamic properties of distributed computing across neuronal assemblies (viz. neural oscillations). We discuss how the ability to learn predicates from experience, to represent information compositionally, and to extrapolate knowledge to unseen data is core to understanding and modeling the most complex human behaviors (e.g. relational reasoning, analogy, language processing, game play).

Abstract

Active communication between researchers and society is necessary for the scientific community’s involvement in developing sciencebased
policies. This need is recognized by governmental and funding agencies that compel scientists to increase their public engagement
and disseminate research findings in an accessible fashion. Storytelling techniques can help convey science by engaging people’s imagination
and emotions. Yet, many researchers are uncertain about how to approach scientific storytelling, or feel they lack the tools to
undertake it. Here we explore some of the techniques intrinsic to crafting scientific narratives, as well as the reasons why scientific
storytellingmaybe an optimal way of communicating research to nonspecialists.Wealso point out current communication gaps between
science and society, particularly in the context of neurodiverse audiences and those that include neurological and psychiatric patients.
Present shortcomings may turn into areas of synergy with the potential to link neuroscience education, research, and advocacy

Abstract

Listeners are known to track statistical regularities in speech. Yet, which temporal cues
are encoded is unclear. This study tested effects of talker-specific habitual speech rate
and talker-independent average speech rate (heard over a longer period of time) on
the perception of the temporal Dutch vowel contrast /A/-/a:/. First, Experiment 1
replicated that slow local (surrounding) speech contexts induce fewer long /a:/
responses than faster contexts. Experiment 2 tested effects of long-term habitual
speech rate. One high-rate group listened to ambiguous vowels embedded in `neutral'
speech from talker A, intermixed with speech from fast talker B. Another low-rate group
listened to the same `neutral' speech from talker A, but to talker B being slow.
Between-group comparison of the `neutral' trials showed that the high-rate group
demonstrated a lower proportion of /a:/ responses, indicating that talker A's habitual
speech rate sounded slower when B was faster. In Experiment 3, both talkers
produced speech at both rates, removing the different habitual speech rates of talker A
and B, while maintaining the average rate differing between groups. This time no
global rate effect was observed. Taken together, the present experiments show that a
talker's habitual rate is encoded relative to the habitual rate of another talker, carrying
implications for episodic and constraint-based models of speech perception.

Abstract

Speech can be produced at different rates. Listeners take this rate variation into account by normalizing vowel duration for contextual speech rate: An ambiguous Dutch word /m?t/ is perceived as short /mAt/ when embedded in a slow context, but long /ma:t/ in a fast context. Whilst some have argued that this rate normalization involves low-level automatic perceptual processing, there is also evidence that it arises at higher-level cognitive processing stages, such as decision making. Prior research on rate-dependent speech perception has only used explicit recognition tasks to investigate the phenomenon, involving both perceptual processing and decision making. This study tested whether speech rate normalization can be observed without explicit decision making, using a cross-modal repetition priming paradigm. Results show that a fast precursor sentence makes an embedded ambiguous prime (/m?t/) sound (implicitly) more /a:/-like, facilitating lexical access to the long target word "maat" in a (explicit) lexical decision task. This result suggests that rate normalization is automatic, taking place even in the absence of an explicit recognition task. Thus, rate normalization is placed within the realm of everyday spoken conversation, where explicit categorization of ambiguous sounds is rare.

Abstract

Human conversations are inherently multimodal, including auditory speech, visual articulatory cues, and hand gestures. Recent studies demonstrated that the timing of a simple up-and-down hand movement, known as a beat gesture, can affect speech perception. A beat gesture falling on the first syllable of a disyllabic word induces a bias to perceive a strong-weak stress pattern (i.e., “CONtent”), while a beat gesture falling on the second syllable combined with the same acoustics biases towards a weak-strong stress pattern (“conTENT”). This effect, termed the “manual McGurk effect”, has been studied in both in-lab and online studies, employing standard experimental sessions lasting approximately forty minutes. The present work tests whether the manual McGurk effect can be observed in an online short version (“mini-test”) of the original paradigm, lasting only ten minutes. Additionally, we employ two different response modalities, namely a two-alternative forced choice and a visual analog scale. A significant manual McGurk effect was observed with both response modalities. Overall, the present study demonstrates the feasibility of employing a ten-minute manual McGurk mini-test to obtain a measure of gesture-speech integration. As such, it may lend itself for inclusion in large-scale test batteries that aim to quantify individual variation in language processing.

Abstract

Classical galactosaemia (CG) is a hereditary disease in galactose metabolism that despite dietary treatment is characterized by a wide range of cognitive deficits, among which is language production. CG brain functioning has been studied with several neuroimaging techniques, which revealed both structural and functional atypicalities. In the present study, for the first time, we compared the oscillatory dynamics, especially the power spectrum and time–frequency representations (TFR), in the electroencephalography (EEG) of CG patients and healthy controls while they were performing a language production task. Twenty-one CG patients and 19 healthy controls described animated scenes, either in full sentences or in words, indicating two levels of complexity in syntactic planning. Based on previous work on the P300 event related potential (ERP) and its relation with theta frequency, we hypothesized that the oscillatory activity of patients and controls would differ in theta power and TFR. With regard to behavior, reaction times showed that patients are slower, reflecting the language deficit. In the power spectrum, we observed significant higher power in patients in delta (1–3 Hz), theta (4–7 Hz), beta (15–30 Hz) and gamma (30–70 Hz) frequencies, but not in alpha (8–12 Hz), suggesting an atypical oscillatory profile. The time-frequency analysis revealed significantly weaker event-related theta synchronization (ERS) and alpha desynchronization (ERD) in patients in the sentence condition. The data support the hypothesis that CG language difficulties relate to theta–alpha brain oscillations.

Abstract

Poor recognition of other-race faces is ubiquitous around the world. We resolve a longstanding contradiction in the literature concerning whether interracial social contact improves the other-race effect. For the first time, we measure the age at which contact was experienced. taking advantage of
unusual demographics allowing dissociation of childhood from adult contact, results show sufficient childhood contact eliminated poor other-race recognition altogether (confirming inter-country adoption
studies). Critically, however, the developmental window for easy acquisition of other-race faces closed by approximately 12 years of age and social contact as an adult — even over several years and involving many other-race friends — produced no improvement. Theoretically, this pattern of developmental change in plasticity mirrors that found in language, suggesting a shared origin grounded in the
functional importance of both skills to social communication. Practically, results imply that, where parents wish to ensure their offspring develop the perceptual skills needed to recognise other-race people easily, childhood experience should be encouraged: just as an English-speaking person who moves to France as a child (but not an adult) can easily become a native speaker of French, we can easily
become “native recognisers” of other-race faces via natural social exposure obtained in childhood, but not later

Abstract

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

Abstract

This thesis addressed the spoken production of complex numerals for time and space. The production of complex numerical expressions like those involved in telling time (e.g., 'quarter to four') or producing house numbers (e.g., 'two hundred forty-five') has been almost completely ignored. Yet, adult speakers produce such expressions on a regular basis in everyday communication. Thus, no theory on numerical cognition or speech production is complete without an account of the production of multi-morphemic utterances such as complex numeral expressions. The main question of this thesis is which particular speech planning levels are involved in the naming and reading of complex numerals for time and space. More specifically, this issue was investigated by examining different modes of response (clock times versus house numbers), alternative input formats (Arabic digit versus alphabetic format; analog versus digital clock displays), and different expression types (relative 'quarter to four' versus absolute 'three forty-five' time expressions).

Abstract

This study investigates how speakers of Dutch compute and produce relative time expressions. Naming digital clocks (e.g., 2:45, say ‘‘quarter to three’’) requires conceptual operations on the minute and hour information for the correct relative time expression. The interplay of these conceptual operations was investigated using a repetition priming paradigm. Participants named analog clocks (the primes) directly before naming digital clocks (the targets). The targets referred to the hour (e.g., 2:00), half past the hour (e.g., 2:30), or the coming hour (e.g., 2:45). The primes differed from the target in one or two hour and in five or ten minutes. Digital clock naming latencies were shorter with a five- than with a ten-min difference between prime and target, but the difference in hour had no effect. Moreover, the distance in minutes had only an effect for half past the hour and the coming hour, but not for the hour. These findings suggest that conceptual facilitation occurs when conceptual transformations are shared between prime and target in telling time.

Abstract

Recent work has claimed that (non-tonal) phonological patterns are subregular (Heinz 2011a,b, 2018; Heinz and Idsardi 2013), occupying a delimited proper subregion of the regular functions—the weakly deterministic (WD) functions (Heinz and Lai 2013; Jardine 2016). Whether or not it is correct (McCollum et al. 2020a), this claim can only be properly assessed given a complete and accurate definition of WD functions. We propose such a definition in this article, patching unintended holes in Heinz and Lai’s (2013) original definition that we argue have led to the incorrect classification of some phonological patterns as WD. We start from the observation that WD patterns share a property that we call unbounded semiambience, modeled after the analogous observation by Jardine (2016) about non-deterministic (ND) patterns and their unbounded circumambience. Both ND and WD functions can be broken down into compositions of deterministic (subsequential) functions (Elgot and Mezei 1965; Heinz and Lai 2013) that read an input string from opposite directions; we show that WD functions are those for which these deterministic composands do not interact in a way that is familiar from the theoretical phonology literature. To underscore how this concept of interaction neatly separates the WD class of functions from the strictly more expressive ND class, we provide analyses of the vowel harmony patterns of two Eastern Nilotic languages, Maasai and Turkana, using bimachines, an automaton type that represents unbounded bidirectional dependencies explicitly. These analyses make clear that there is interaction between deterministic composands when (and only when) the output of a given input element of a string is simultaneously dependent on information from both the left and the right: ND functions are those that involve interaction, while WD functions are those that do not.

Abstract

This paper aims to determine whether iconic tracing gestures produced while speaking constitute part of the speaker’s communicative intention. We used a picture description task in which speakers must communicate the spatial and color information of each picture to an interlocutor. By establishing the necessary minimal content of an intended message, we determined whether speech produced with concurrent gestures is less explicit than speech without gestures. We argue that a gesture must be communicatively intended if it expresses necessary information that was nevertheless omitted from speech. We found that speakers who produced iconic gestures representing spatial relations omitted more required spatial information from their descriptions than speakers who did not gesture. These results provide evidence that speakers intend these gestures to communicate. The results have implications for the cognitive architectures that underlie the production of gesture and speech.

Abstract

Scientific convergence is a phenomenon where the distance between hitherto distinct scientific fields narrows and the fields gradually overlap over time. It is creating important potential for research, development, and innovation. Although scientific convergence is crucial for the development of radically new technology, the identification of emerging scientific convergence is particularly difficult since the underlying knowledge flows are rather fuzzy and unstable in the early convergence stage. Nevertheless, novel scientific publications emerging at the intersection of different knowledge fields may reflect convergence processes. Thus, in this article, we exploit the growing number of research and digital libraries providing bibliographic metadata to propose an automated analysis of science dynamics. We utilize and adapt machine-learning methods (DeepWalk) to automatically learn a similarity measure between scientific fields from graphs constructed on bibliographic metadata. With a time-based perspective, we apply our approach to analyze the trajectories of evolving similarities between scientific fields. We validate the learned similarity measure by evaluating it within the well-explored case of cholesterol-lowering ingredients in which scientific convergence between the distinct scientific fields of nutrition and pharmaceuticals has partially taken place. Our results confirm that the similarity trajectories learned by our approach resemble the expected behavior, indicating that our approach may allow researchers and practitioners to detect and predict scientific convergence early.

Abstract

While it is well established that grammar learning success varies with age, the cause of this developmental change is largely unknown. This study examined functional MRI activation across a broad developmental sample of 165 Dutch-speaking individuals (8-25 years) as they were implicitly learning a new grammatical system. This approach allowed us to assess the direct effects of age on grammar learning ability while exploring its neural correlates. In contrast to the alleged advantage of children language learners over adults, we found that adults outperformed children. Moreover, our behavioral data showed a sharp discontinuity in the relationship between age and grammar learning performance: there was a strong positive linear correlation between 8 and 15.4 years of age, after which age had no further effect. Neurally, our data indicate two important findings: (i) during grammar learning, adults and children activate similar brain regions, suggesting continuity in the neural networks that support initial grammar learning; and (ii) activation level is age-dependent, with children showing less activation than older participants. We suggest that these age-dependent processes may constrain developmental effects in grammar learning. The present study provides new insights into the neural basis of age-related differences in grammar learning in second language acquisition.

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

Current approaches to learning semantic representations of sentences often use prior word-level knowledge. The current study aims to leverage visual information in order to capture sentence level semantics without the need for word embeddings. We use a multimodal sentence encoder trained on a corpus of images with matching text captions to produce visually grounded sentence embeddings. Deep Neural Networks are trained to map the two modalities to a common embedding space such that for an image the corresponding caption can be retrieved and vice versa. We show that our model achieves results comparable to the current state of the art on two popular image-caption retrieval benchmark datasets: Microsoft Common Objects in Context (MSCOCO) and Flickr8k. We evaluate the semantic content of the resulting sentence embeddings using the data from the Semantic Textual Similarity (STS) benchmark task and show that the multimodal embeddings correlate well with human semantic similarity judgements. The system achieves state-of-the-art results on several of these benchmarks, which shows that a system trained solely on multimodal data, without assuming any word representations, is able to capture sentence level semantics. Importantly, this result shows that we do not need prior knowledge of lexical level semantics in order to model sentence level semantics. These findings demonstrate the importance of visual information in semantics.

Abstract

Spatial memory deficits are core features of aging-related changes in cognitive abilities. The neural correlates of these deficits are largely unknown. In the present study, we investigated the neural underpinnings of age-related differences in spatial memory by functional MRI using a navigational memory task with route encoding and route recognition conditions. We investigated 20 healthy young (18 - 29 years old) and 20 healthy old adults (53 - 78 years old) in a random effects analysis. Old subjects showed slightly poorer performance than young subjects. Compared to the control condition, route encoding and route recognition showed activation of the dorsal and ventral visual processing streams and the frontal eye fields in both groups of subjects. Compared to old adults, young subjects showed during route encoding stronger activations in the dorsal and the ventral visual processing stream (supramarginal gyrus and posterior fusiform/parahippocampal areas). In addition, young subjects showed weaker anterior parahippocampal activity during route recognition compared to the old group. In contrast, old compared to young subjects showed less suppressed activity in the left perisylvian region and the anterior cingulate cortex during route encoding. Our findings suggest that agerelated navigational memory deficits might be caused by less effective route encoding based on reduced posterior fusiform/parahippocampal and parietal functionality combined with diminished inhibition of perisylvian and anterior cingulate cortices correlated with less effective suppression of task-irrelevant information. In contrast, age differences in neural correlates of route recognition seem to be rather subtle. Old subjects might show a diminished familiarity signal during route recognition in the anterior parahippocampal region.

Abstract

Earlier work has shown that speakers naming several objects usually look at each of them before naming them (e.g., Meyer, Sleiderink, & Levelt, 1998). In the present study, participants saw pictures and described them in utterances such as "The chair next to the cross is brown", where the colour of the first object was mentioned after another object had been mentioned. In Experiment 1, we examined whether the speakers would look at the first object (the chair) only once, before naming the object, or twice (before naming the object and before naming its colour). In Experiment 2, we examined whether speakers about to name the colour of the object would look at the object region again when the colour or the entire object had been removed while they were looking elsewhere. We found that speakers usually looked at the target object again before naming its colour, even when the colour was not displayed any more. Speakers were much less likely to fixate upon the target region when the object had been removed from view. We propose that the object contours may serve as a memory cue supporting the retrieval of the associated colour information. The results show that a speaker's eye movements in a picture description task, far from being random, depend on the available visual information and the content and structure of the planned utterance.