Computational Exploration of Metaphor Comprehension Processes Using a Semantic Space Model

JWSAN: Japanese word similarity and association norm

We present a new Japanese dataset, Japanese Word Similarity and Association Norm (JWSAN), comprising human rating scores of similarity and association for 2145 word pairs, with a clear distinction between word similarity and word association. Computational models of human semantic memory or mental lexicon, such as distributed semantic models, must predict not only association but also similarity. People can distinguish between word similarity and association. However, although the SimLex-999 dataset is publicly available for English, there is no Japanese similarity dataset with a clear distinction between the two types of word relatedness. JWSAN is the first large Japanese dataset with similarity and association ratings, containing noun, verb, and adjective word pairs. It is also characterized by data collection from a sufficient number of age- and-gender-controlled assessors, with similarity and association ratings obtained via a web-based survey conducted of 6450 native speakers of Japanese. In addition, the effects of the gender and age of the raters were also examined; these factors were only given scant consideration in the past. This dataset can act as a benchmark for improving distributed semantic models in Japanese.

The RK processor: A program for analysing metaphor and word feature-listing data

Feature-listing tasks are an invaluable resource for exploring how words, categories, and metaphors are represented. However, manually coding the generated features is time-consuming and expensive, and involves subjective judgments from the experimenter. The purpose of this paper is to introduce the "RK processor", a program that was developed in our lab to analyse metaphor feature data but which can also be applied to other feature-listing data. After detailing the steps of processing, we demonstrate that the processed feature data align with previous findings in which metaphor features were processed manually and that the processed features predict dimensions of metaphor judgments pertaining to comprehensibility and metaphor goodness. Lastly, we present several other applications for research on word similarity, compound words, categories and concepts, semantic ambiguity, incongruity resolution and computational modelling. The RK processor offers researchers a valuable tool to save time and resources and to maintain consistency in processing.

Evaluating the predication model of metaphor comprehension: Using word2vec to model best/worst quality judgments of 622 novel metaphors

In this paper our goal is to undertake a systematic assessment of the first, most widely known, and simplest computational model of metaphor comprehension, the predication model developed by Kintsch (Cognitive Science, 25(2), 173-202, 2000). 622 metaphors of the form "x is a y" were selected from a much larger set generated randomly. The metaphors were judged for quality using best/worst judgments, which asks judges to pick the best and worst metaphor from among four presented metaphors. The metaphors and their judgments have been publicly released. We modeled the judgments by extending Kintsch's predication model (2000) by systematically walking through the parameter space of that model. Our model successfully differentiated metaphors rated as good (> 1.5z) from metaphors rated as bad (< -1.5z; Cohen's d = 0.72) and was able to successfully classify good metaphors with an accuracy of 82.9%. However, it achieved a true negative rate below chance at 36.3% and had a resultantly low kappa of 0.037. The model could not distinguish unselected random metaphors from those selected by humans as having metaphorical potential. In a follow-up study we showed that the model's quality estimates reliably predict metaphor decision times, with better metaphors being judged more quickly than worse metaphors.

Normative Data for Novel Nominal Metaphors, Novel Similes, Literal, and Anomalous Utterances in Polish and English

The two studies reported in the article provide normative measures for 120 novel nominal metaphors, 120 novel similes, 120 literal sentences, and 120 anomalous utterances in Polish (Study 1) and in English (Study 2). The presented set is ideally suited to addressing methodological requirements in research on metaphor processing. The critical (sentence-final) words of each utterance were controlled for in terms of their frequency per million, number of letters and syllables. For each condition in each language, the following variables are reported: cloze probability, meaningfulness, metaphoricity, and familiarity, whose results confirm that the sentences are well-matched. Consequently, the present paper provides materials that can be employed in order to test the new as well as existing theories of metaphor comprehension. The results obtained from the series of normative tests showed the same pattern in both studies, where the comparison structure present in similes (i.e., A is like B) facilitated novel metaphor comprehension, as compared to categorical statements (i.e., A is B). It therefore indicates that comparison mechanisms might be engaged in novel meaning construction irrespectively of language-specific syntactic rules.

Re-Representing Metaphor: Modeling Metaphor Perception Using Dynamically Contextual Distributional Semantics

In this paper, we present a novel context-dependent approach to modeling word meaning, and apply it to the modeling of metaphor. In distributional semantic approaches, words are represented as points in a high dimensional space generated from co-occurrence statistics; the distances between points may then be used to quantifying semantic relationships. Contrary to other approaches which use static, global representations, our approach discovers contextualized representations by dynamically projecting low-dimensional subspaces; in these ad hoc spaces, words can be re-represented in an open-ended assortment of geometrical and conceptual configurations as appropriate for particular contexts. We hypothesize that this context-specific re-representation enables a more effective model of the semantics of metaphor than standard static approaches. We test this hypothesis on a dataset of English word dyads rated for degrees of metaphoricity, meaningfulness, and familiarity by human participants. We demonstrate that our model captures these ratings more effectively than a state-of-the-art static model, and does so via the amount of contextualizing work inherent in the re-representational process.

Metaphor in Sign Languages

Metaphor abounds in both sign and spoken languages. However, in sign languages, languages in the visual-manual modality, metaphors work a bit differently than they do in spoken languages. In this paper we explore some of the ways in which metaphors in sign languages differ from metaphors in spoken languages. We address three differences: (a) Some metaphors are very common in spoken languages yet are infelicitous in sign languages; (b) Body-part terms are possible in very specific types of metaphors in sign languages, but are not so restricted in spoken languages; (c) Similes in some sign languages are dispreferred in predicative positions in which metaphors are fine, in contrast to spoken languages where both can appear in these environments. We argue that these differences can be explained by two seemingly unrelated principles: the Double Mapping Constraint (Meir, 2010), which accounts for the interaction between metaphor and iconicity in languages, and Croft's (2003) constraint regarding the autonomy and dependency of elements in metaphorical constructions. We further argue that the study of metaphor in the signed modality offers novel insights concerning the nature of metaphor in general, and the role of figurative speech in language.

Tone matters for Cantonese-English bilingual children's English word reading development: A unified model of phonological transfer

Languages differ considerably in how they use prosodic features, or variations in pitch, duration, and intensity, to distinguish one word from another. Prosodic features include lexical tone in Chinese and lexical stress in English. Recent cross-sectional studies show a surprising result that Mandarin Chinese tone sensitivity is related to Mandarin-English bilingual children's English word reading. This study explores the mechanism underlying this relation by testing two explanations of these effects: the prosodic hypothesis and segmental phonological awareness transfer. We administered multiple measures of Cantonese tone sensitivity, English stress sensitivity, segmental phonological awareness in Cantonese and English, nonverbal ability, and English word reading to 123 Cantonese-English bilingual children ages 7 and 8 years. Structural equation modeling revealed a longitudinal prediction of Cantonese tone sensitivity to English word reading between 8 and 9 years of age. This relation was realized through two parallel routes. In one, Cantonese tone sensitivity predicted English stress sensitivity, and English stress sensitivity, in turn, significantly predicted English word reading, as postulated by the prosodic hypothesis. In the second, Cantonese tone sensitivity predicted English word reading through the transfer of segmental phonological awareness between Cantonese and English, as predicted by segmental phonological transfer. These results support a unified model of phonological transfer, emphasizing the role of tone in English word reading for Cantonese-English bilingual children.

Discourse Goals Affect the Process and Product of Nominal Metaphor Production

Although a large number of studies have addressed metaphor comprehension, only a few attempts have so far been made at exploring the process of metaphor production. Therefore, in this paper, we address the problem of how people generate nominal metaphors or identify an apt vehicle for a given topic of nominal metaphors. Specifically, we examine how the process and product of metaphor production differ between two discourse goals of metaphor, namely an explanatory purpose (e.g., to clarify) and a literary purpose (e.g., to aesthetically pleasing). Experiment 1 analyzed the metaphors (or vehicles) generated in the metaphor production task, and demonstrated that people identified more prototypical exemplars of the property to be attributed to the topic as a vehicle for explanatory metaphors than for literary metaphors. In addition, it was found that metaphors generated for the explanatory purpose were more apt and conventional, and had high topic-vehicle similarity than those generated for the literary purpose, while metaphors generated for the literary purpose were more familiar and imageable than those for the explanatory purpose. Experiment 2 used a priming paradigm to assess the online availability of prototypical and less prototypical members of the topic property during metaphor production. The result was that both prototypical and less prototypical members were activated in producing literary metaphors, while neither members were activated in the production of explanatory metaphors. These findings indicate that the process of metaphor production is affected by discourse goals of metaphor; less prototypical members of the category are searched for a vehicle during the production of literary metaphors, and thus literary metaphors are generated with less prototypical vehicles than explanatory metaphors.

A Complex Network Approach to Distributional Semantic Models

A number of studies on network analysis have focused on language networks based on free word association, which reflects human lexical knowledge, and have demonstrated the small-world and scale-free properties in the word association network. Nevertheless, there have been very few attempts at applying network analysis to distributional semantic models, despite the fact that these models have been studied extensively as computational or cognitive models of human lexical knowledge. In this paper, we analyze three network properties, namely, small-world, scale-free, and hierarchical properties, of semantic networks created by distributional semantic models. We demonstrate that the created networks generally exhibit the same properties as word association networks. In particular, we show that the distribution of the number of connections in these networks follows the truncated power law, which is also observed in an association network. This indicates that distributional semantic models can provide a plausible model of lexical knowledge. Additionally, the observed differences in the network properties of various implementations of distributional semantic models are consistently explained or predicted by considering the intrinsic semantic features of a word-context matrix and the functions of matrix weighting and smoothing. Furthermore, to simulate a semantic network with the observed network properties, we propose a new growing network model based on the model of Steyvers and Tenenbaum. The idea underlying the proposed model is that both preferential and random attachments are required to reflect different types of semantic relations in network growth process. We demonstrate that this model provides a better explanation of network behaviors generated by distributional semantic models.

The role of working memory and divided attention in metaphor interpretation

Although several types of figurative language exist, neuropsychological tests of non-literal language have focused on proverbs. Metaphors in the form X is (a) Y (e.g., The body's immunological response is a battle against disease.) place a lower demand on language skills and are more easily manipulated for novelty than proverbs. Forty healthy participants completed the Metaphor Interpretation Test (developed by the authors). The task includes 20 items chosen from a list of metaphors that were rated on several scales (e.g. imagery, aptness) in a study by Katz et al. (Metaphor Symb Act 3(4):191-214, 1988). Participants were asked to rate the familiarity and provide an explanation of each metaphor. A scoring system was developed to categorize answers into: abstract complete (AC), abstract partial (AP), concrete (CT), and other/unrelated (OT) types. Participants also completed short-term memory and divided attention tests. Overall, participants produced 56 % AC, 25.38 % AP, 7.88 % CT, and 10.88 % OT responses. It was found that a measure of verbal short-term memory span was the best predictor of performance on this task (adjusted R(2) = .369). It appears that short-term memory span, not working memory or divided attention, contributes most to providing abstract responses in explaining metaphors. This is in line with the idea that when one accesses the semantic network associated with a novel metaphor, one must hold this information in mind long enough to search for and link similar cognitive networks.

Latent semantic analysis

This article reviews latent semantic analysis (LSA), a theory of meaning as well as a method for extracting that meaning from passages of text, based on statistical computations over a collection of documents. LSA as a theory of meaning defines a latent semantic space where documents and individual words are represented as vectors. LSA as a computational technique uses linear algebra to extract dimensions that represent that space. This representation enables the computation of similarity among terms and documents, categorization of terms and documents, and summarization of large collections of documents using automated procedures that mimic the way humans perform similar cognitive tasks. We present some technical details, various illustrative examples, and discuss a number of applications from linguistics, psychology, cognitive science, education, information science, and analysis of textual data in general. WIREs Cogn Sci 2013, 4:683-692. doi: 10.1002/wcs.1254 CONFLICT OF INTEREST: The author has declared no conflicts of interest for this article. For further resources related to this article, please visit the WIREs website.

Detecting order-disorder transitions in discourse: implications for schizophrenia

Several psychiatric and neurological conditions affect the semantic organization and content of a patient's speech. Specifically, the discourse of patients with schizophrenia is frequently characterized as lacking coherence. The evaluation of disturbances in discourse is often used in diagnosis and in assessing treatment efficacy, and is an important factor in prognosis. Measuring these deviations, such as "loss of meaning" and incoherence, is difficult and requires substantial human effort. Computational procedures can be employed to characterize the nature of the anomalies in discourse. We present a set of new tools derived from network theory and information science that may assist in empirical and clinical studies of communication patterns in patients, and provide the foundation for future automatic procedures. First we review information science and complex network approaches to measuring semantic coherence, and then we introduce a representation of discourse that allows for the computation of measures of disorganization. Finally we apply these tools to speech transcriptions from patients and a healthy participant, illustrating the implications and potential of this novel framework.