Grounding co-occurrence: Identifying features in a lexical co-occurrence model of semantic memory

Lexical co-occurrence models of semantic memory represent word meaning by vectors in a high-dimensional space. These vectors are derived from word usage, as found in a large corpus of written text. Typically, these models are fully automated, an advantage over models that represent semantics that are based on human judgments (e.g., feature-based models). A common criticism of co-occurrence models is that the representations are not grounded: Concepts exist only relative to each other in the space produced by the model. It has been claimed that feature-based models offer an advantage in this regard. In this article, we take a step toward grounding a cooccurrence model. A feed-forward neural network is trained using back propagation to provide a mapping from co-occurrence vectors to feature norms collected from subjects. We show that this network is able to retrieve the features of a concept from its co-occurrence vector with high accuracy and is able to generalize this ability to produce an appropriate list of features from the co-occurrence vector of a novel concept.     

How sublexical association strength modulates updating: Cognitive and strategic effects

In the current study, we investigated updating of long-term memory (LTM) associations. Specifically, we examined sublexical associations by manipulating preexisting LTM relations between consonant couplets (in encoding and updating phases), and explicitly instructed participants to engage with a specific strategy for approaching the task (item disjunction, grouping, or none). In two experiments, we used a multistep subject-based memory updating task in which we measured processing response times (RTs; Exp. 1, Exp. 2) and recognition RTs (Exp. 2). For the first time, in both experiments, we found costs in dismantling strong pre-existing associations from LTM and benefits in recreating strong preexisting associations. In addition, we found that control of irrelevant information was more difficult when this belonged to a strong association. Regarding task strategies, we showed that inducing a disjunction strategy enhanced updating, no matter the strength of the association. Results were discussed in the light of updating as a process of dismantling and recreating associations. The role of a specific strategic approach in enhancing the updating was also discussed.     

Toward a Unified Model of Lexical Acquisition and Lexical Access

Much effort has gone into constructing models of how children segment speech and thereby discover the words of their language. Much effort has also gone into constructing models of how adults access their mental lexicons and thereby segment speech into words. In this paper, I explore the possibility of a model that could account for both word discovery by children and on-line segmentation by adults. In particular, I discuss extensions to the distributional regularity (DR) model of Brent and Cartwright (1996) that could yield an account of on-line segmentation as well as word discovery.     

The co-occurrence of multisensory competition and facilitation

Previous studies of multisensory integration have often stressed the beneficial effects that may arise when information concerning an event arrives via different sensory modalities at the same time, as, for example, exemplified by research on the redundant target effect (RTE). By contrast, studies of the Colavita visual dominance effect (e.g., [Colavita, F. B. (1974). Human sensory dominance. Perception & Psychophysics, 16, 409–412]) highlight the inhibitory consequences of the competition between signals presented simultaneously in different sensory modalities instead. Although both the RTE and the Colavita effect are thought to occur at early sensory levels and the stimulus conditions under which they are typically observed are very similar, the interplay between these two opposing behavioural phenomena (facilitation vs. competition) has yet to be addressed empirically. We hypothesized that the dissociation may reflect two of the fundamentally different ways in which humans can perceive concurrent auditory and visual stimuli. In Experiment 1, we demonstrated both multisensory facilitation (RTE) and the Colavita visual dominance effect using exactly the same audiovisual displays, by simply changing the task from a speeded detection task to a speeded modality discrimination task. Meanwhile, in Experiment 2, the participants exhibited multisensory facilitation when responding to visual targets and multisensory inhibition when responding to auditory targets while keeping the task constant. These results therefore indicate that both multisensory facilitation and inhibition can be demonstrated in reaction to the same bimodal event.     

Bilingualism and Lexical Representation

Forty-eight Hindi-English bilinguals completed two blocks of trials where each trial involved presentation of a letter string requiring a lexical decision. In the first block subjects were exposed to 22 words and 11 non-words in either English or Hindi. In the second block the original words were repeated in either the same language or in the alternative language. In this block the old (repeated) words were mixed with 22 new words, and 22 non-words. Twelve subjects were included in each of the four groups given by the factorial combination of blocks and languages. Reaction time in the lexical decision task was facilitated when words were repeated in the same language (109 and 125 ms in the English-English and Hindi-Hindi groups respectively), but little or no facilitation was observed in the inter-lingual conditions (-22 and 23 ms in the Hindi-English and English-Hindi conditions respectively). The results support the view that lexical representation in bilinguals is language-specific.     

Selection demands versus association strength in the verb generation task

Neuroimaging and neuropsychological data from Thompson-Schill and colleagues (Thompson-Schill, D’Esposito, Aguirre, & Farah, 1997; Thompson-Schill et al., 1998) showed that in the verb generation task, the left inferior frontal gyrus (left IFG) was more involved in a high-selection condition, in which the noun had two verb responses of equal association strength, than in a low-selection condition, in which the noun had only one strongly associated verb. They proposed that the left IFG was involved in selecting semantic information from competing alternatives. The present study compared verb generation in two high-selection conditions that varied in the association strength of the most frequently produced verbs. The results from neurally intact participants and a patient with a lesion that included the left IFG indicated that association strength between nouns and the most frequently produced verb, rather than competition during verb selection, affected verb generation. The degree of involvement of the left IFG may depend on the difficulty of verb retrieval as reflected in association strength.     

Lexical derivation

A recent point of controversy in linguistics has been the correct treatment of derived nouns such asrefusal, destruction, andeageness. Two positions, the transformationalist approach and the lexicalist approach, have been proposed to account for the relationship between these nouns and their base forms (refuse, destroy, eager). The two approaches differ over how they resolve an incompatibility among the basic assumptions of transformational grammar. Positive linguistic arguments given by Chomsky (1970) favor the lexicalist account. Empirical predictions are suggested from the application of the two approaches to psycholinguistic situations. Experimental findings are reviewed, including those from a research program contrasting lexical derivation and the attributes of imageability and concreteness. The results fail to support either position, and throw doubt upon the idea that there is a genuine phenomenon to be investigated. This gives rise to a dilemma over the relationship between linguistics and psychology, which may be resolved either by rejecting the conception of linguistics as the explanation of linguistic intuitions or by rejecting the notion of a direct link between linguistics and psychology.     

Exploring lexical co-occurrence space using HiDEx

Hyperspace analog to language (HAL) is a high-dimensional model of semantic space that uses the global co-occurrence frequency of words in a large corpus of text as the basis for a representation of semantic memory. In the original HAL model, many parameters were set without any a priori rationale. We have created and publicly released a computer application, the High Dimensional Explorer (HiDEx), that makes it possible to systematically alter the values of these parameters to examine their effect on the co-occurrence matrix that instantiates the model. We took an empirical approach to understanding the influence of the parameters on the measures produced by the models, looking at how well matrices derived with different parameters could predict human reaction times in lexical decision and semantic decision tasks. New parameter sets give us measures of semantic density that improve the model’s ability to predict behavioral measures. Implications for such models are discussed.     

Attitude accessibility,attitude-behavior consistency,and the strength of the object-evaluation association

Two experiments were conducted in order to examine the accessibility of attitudes from memory as a function of the manner of attitude formation. The findings of the first experiment indicated that subjects could respond more quickly in a response-time task to inquiries about their attitudes when the attitudes were based upon direct behavioral experience with the attitude objects than when they were based upon nonbehavioral experience. It was suggested that, relative to indirect experience, behavioral experience may facilitate the attitude formation process and increase attitude accessibility once the attitude is formed. A second experiment found support for both of these notions. Two additional experiments indicated that repeated association of the attitude object and the attitudinal evaluation enhanced both attitude accessibility and attitude-behavior consistency. It was suggested that the strength of the object-evaluation association is a critical determinant of accessibility, which, in turn, acts as a central factor in the process by which attitudes affect later behavior. It was further suggested that the manner of attitude formation affects attitude-behavior consistency because direct experience produces a stronger object-evaluation association and, hence, a more accessible attitude than does indirect experience.     

Contrasting Grammatical and Lexical Determiners

This paper investigates the difference between the production of grammatical determiners and lexical determiners in the production of adjective-noun phrases (NPs) in Danish. Models of sentence processing (Garrett in Psychology of learning and motivation, Academic press, New York, pp 133–177, 1975; Bock in J Mem Lang 26(2):119–137, 1987) suggest that the phonological encoding stage of grammatical items can only be specified once lexical items have been phonologically encoded. In their usage-based linguistic theory of the grammar-lexicon distinction, Boye and Harder (Lang 88(1):1–44, 2012) propose that this later encoding of grammatical elements is motivated by two specific features of grammatical elements. The first feature, dependence, is that grammatical items (morphemes, words, constructions) cannot be produced in isolation, but are always dependent on a lexical host item. This feature entails a more complex processing which might lead to longer reaction times when comparing the production of NPs with a grammatical determiner to a lexical one. Additionally, a more complex processing might lead to a lower accuracy rate for the grammatical condition relative to the lexical one. The second feature, low prominence, is that grammatical items code background information and therefore cannot convey the main point of a linguistic message. Less focus on grammatical elements might lead to a lower accuracy rate for the production of grammatical elements relative to lexical ones. Those predictions were tested in a task comparing the production of Danish grammatical determiners (indefinite articles) with the production of lexical ones (numerals, which are homonymous with the articles except for a stress difference) in similar contexts. Group-based analyses were performed in order to take inter-individual differences into account. The results show that the two features as proposed by Boye and Harder (2012) are only revealed for the fastest speakers group but not the slower ones.     

Producing high-dimensional semantic spaces from lexical co-occurrence

A procedure that processes a corpus of text and produces numeric vectors containing information about its meanings for each word is presented. This procedure is applied to a large corpus of natural language text taken from Usenet, and the resulting vectors are examined to determine what information is contained within them. These vectors provide the coordinates in a high-dimensional space in which word relationships can be analyzed. Analyses of both vector similarity and multidimensional scaling demonstrate that there is significant semantic information carried in the vectors. A comparison of vector similarity with human reaction times in a single-word priming experiment is presented. These vectors provide the basis for a representational model of semantic memory, hyperspace analogue to language (HAL).     

Effects of Lexical Prosody and Word Familiarity on Lexical Access of Spoken Japanese Words

Lexical prosody (e.g., stress and pitch accent) has been shown to constrain lexical activation of spoken words in various languages. In the present study, whether or not the constraint of lexical prosody is affected by word familiarity in lexical access of Japanese words was examined using a cross-modal priming task. The stimuli were pairs of prosodically different homophones (minimal accent pairs). When the targets were more familiar members of minimal accent pairs, the responses were facilitated by prior presentations of primes that were prosodically different homophones of the targets, suggesting that lexical prosody did not constrain lexical activation. In contrast, when less familiar members of minimal accent pairs were used as the targets, the prosodically different homophones did not facilitate the responses to the targets. These results suggest that the constraint of lexical prosody is not so strong but is affected by the factor of word relative familiarity.     

Lexical Anomia: Or the Case of the Missing Lexical Entries

This paper reports the case of an aphasic patient, EE, with a problem in word retrieval. He is consistently unable to produce specific lexical items, which tend to be items of low rated familiarity. His retrieval of these words is not aided by the provision of phonemic cues or extra time for word retrieval. His errors consist primarily of failures to respond, and the provision of semantic information without any attempt at the target. It is argued that this pattern of performance is consistent with the loss of specific lexical items from a phonological lexicon for speech production.

EE is shown to have no impairment in auditory recognition and comprehension of the lexical items that are unavailable for naming. This dissociation is problematic for theories that propose a single phonological lexicon for both word recognition and production, but is easily accounted for by separate input and output lexicons.     

Lexical morphology and lexical access.

Research on morphology in word recognition has been plagued by conflicting results (McQueen & Cutler, 1998, give a recent review). Some findings suggest that words are accessed as full forms, while others suggest that words are accessed in terms of their component morphemes. The answer may lie in the properties of the affixes themselves: Kiparsky's (1982) Lexical Phonology and Morphology assigns affixes in English to different "levels" of attachment, based on their productivity, order of attachment, and phonological interaction with roots. We present data suggesting that productive, phonologically neutral, semantically transparent "Level 2" suffixes are "decomposed" for analysis in some cases, but that words with idiosyncratic, structure-changing, semantically opaque "Level 1" suffixes are not.     

Ontologies and Construals in Lexical Semantics

The purpose of this paper is to propose a framework of lexical meaning, broadly along the lines of Cognitive Semantics (Langacker 1987a). Within the proposed model, all aspects of meaning are to be explained in terms of properties of ontologies in conceptual space, i.e. properties of content ontologies and schematic ontologies and construals which are imposed on the conceptual structures on the occasion of use. It is through the operations of construals on ontological structures that different readings of lexical expressions arise. Lexical meanings are dynamic and sensitive to contextual demands, rather than fixed and stable. In a dynamic, usage-based model like this, polysemy and multiple readings emerge as a natural consequence of the human ability to think flexibly. Another more specific purpose of this paper is to draw attention to the usefulness of ontologies in linguistic research in general and semantic modelling in particular.     

Phonemes: Lexical access and beyond

Phonemes play a central role in traditional theories as units of speech perception and access codes to lexical representations. Phonemes have two essential properties: they are ‘segment-sized’ (the size of a consonant or vowel) and abstract (a single phoneme may be have different acoustic realisations). Nevertheless, there is a long history of challenging the phoneme hypothesis, with some theorists arguing for differently sized phonological units (e.g. features or syllables) and others rejecting abstract codes in favour of representations that encode detailed acoustic properties of the stimulus. The phoneme hypothesis is the minority view today. We defend the phoneme hypothesis in two complementary ways. First, we show that rejection of phonemes is based on a flawed interpretation of empirical findings. For example, it is commonly argued that the failure to find acoustic invariances for phonemes rules out phonemes. However, the lack of invariance is only a problem on the assumption that speech perception is a bottom-up process. If learned sublexical codes are modified by top-down constraints (which they are), then this argument loses all force. Second, we provide strong positive evidence for phonemes on the basis of linguistic data. Almost all findings that are taken (incorrectly) as evidence against phonemes are based on psycholinguistic studies of single words. However, phonemes were first introduced in linguistics, and the best evidence for phonemes comes from linguistic analyses of complex word forms and sentences. In short, the rejection of phonemes is based on a false analysis and a too-narrow consideration of the relevant data.     

Lexical decisions with homophones and homonyms

In two experiments homophones, homonyms, and control words were presented aurally in a lexical decision task. Response times for homophones were significantly faster than those for homonyms and control words, whereas the latter two did not differ. It was concluded that lexical decision times do not provide evidence for multiple entries in the internal lexicon for homonyms. The possibility is considered that the superior performance with homophonic words also does not reflect activation of multiple entries corresponding to multiple meanings, but instead the utilization of orthographic information. This study was supported by a grant from the Natural Sciences and Engineering Research Council of Canada.     

Lexical representation of fact and opinion

This research investigated the lexical properties underlying comprehension that a word represents fact or opinion. In Experiment 1 subjects reliably identified words as either fact or opinion. Bivariate correlations and multiple regression showed that fact/opinion judgments were predicted primarily by ratings of the ease of verifiability of a word's referent and secondarily by the word's literalness, but not by several other lexical attributes: abstractness-concreteness, vagueness-preciseness, and evaluation. Experiment 2 extended the results of the first experiment to an implicit fact/opinion judgment, i.e., the identification of headlines as originating from a newspaper's front page (presumably based primarily upon fact) or from the editorial page (presumably based upon opinion). The headline judgments, also made reliably by subjects, were predicted by the same variables found significant in the first experiment, i.e., by verifiability and literalness, but not by the other lexical properties. Thus, the results of both experiments indicate that the identification of a word as representing fact or opinion is rooted in a word's verifiability and literalness.