Band 4, Heft 2

In this paper I want to re-examine the key corpus-linguistic notion of semantic preference. This is defined here as the collocation of a lexical item with items from a specific (more or less general) semantic subset. The article aims to throw some light on the term semantic preference, and to examine in more detail some aspects of semantic preference that are frequently neglected in research. It also discusses how semantic preference interacts with syntax and meaning, and what happens when semantic preferences are not ‘realized’ in context. Finally, it seeks to illuminate the distinction between semantic preference and semantic prosody, and points to future research in this area.

Two research questions are examined in this work regarding the uses of ‘market’ in Mandarin, Malay and English. The first question asks whether the use of ‘market’ in these three languages is similar or different. The second question asks whether the collocates of the ‘market’ are similar or different across these languages when used in different grammatical relations. Implications of the similarities and differences will be discussed. In order to answer these two questions, ‘market’ metaphors used by different communities are laid out based on the frequency counts of its source domains and the collocates according to different grammatical roles (subject, object, modifier, etc.) of ‘market.’ The results show that certain source domains have preferences for different grammatical roles for ‘market.’ In addition to this finding, the choice of source domains by different speech communities may also reflect their perspectives regarding their country's economy. Therefore, through using quantitative data, this paper is able to infer the perspectives of these speech communities when referring to ‘market’ in their languages. This can be done not only through analyzing the semantic meanings of the metaphors, but also through their interface with grammatical relations.

This article addresses the problem of the inconsistency of the frequencies reported by the commercial search engines Google and Yahoo, assessing their reliability for academic use and seeking ways to use them more safely for language research and learning. The main part of the article is the presentation of three tests on the self-consistency and relational consistency of search engine frequencies and their results showing some important patterns. It is found that despite occasional frequency shifts of large magnitude, the engines (especially Yahoo) generally maintain a reasonable degree of self-consistency. However, some of the Google frequencies were apparently incorrect as indicated by their relational inconsistency. The tests are preceded by an overview of search engine uses and problems and a discussion of ‘query cache’, which is an essential component of the frequency mechanism providing a possible explanation of search engine behaviors. In conclusion, a search strategy and ways of carrying this work further to develop a model of search engine use are proposed.

While the small but growing field of corpus-based work on metaphor offers numerous examples of fine-grained analyses of metaphor as it occurs in discourse, such work has generally been extremely narrow in focus, examining the distribution either of particular, searchable lexical items or of instantiations of a certain conceptual metaphor within a given discourse domain. Aiming for a broader approach, this paper addresses some of the methodological issues associated with using corpora as a tool for metaphor research, and proposes criteria for identifying metaphor in spontaneous discourse. Based on approximately 500 metaphorical utterances drawn from the Santa Barbara Corpus of Spoken American English (DuBois 2000; DuBois et. al. 2003), this study assesses the overall frequency of metaphor in discourse, as well as the relative frequencies with which various types of topics are referred to metaphorically, towards the goal of a better understanding of how and why speakers use metaphor in common language. It is discovered that the frequency of each of these categories of metaphor corresponds closely with its degree of concreteness, as verified independently with a survey of college students. In addition, this paper explores the role of metaphor as a discourse-level structuring tool. The overall methodology is offered as a model for future research.

In the past decades language corpora have become indispensable tools for linguistic research and the development of linguistic theory. However, it is not yet widely acknowledged that the quality of corpus-based research and theories depends crucially on the quality of the corpora, not only in terms of their content and size but especially as far as the accuracy and richness of the annotations are concerned. Neither has much systematic thought gone into the effectiveness of the traditional corpus creation process regarding this problem. This paper proposes a novel approach to corpus creation – agile corpus creation – that addresses the problem of simultaneously maximizing corpus size as well as the quality and quantity of manual and automatic annotations while minimizing the time and cost involved in corpus creation. The central aspects of agile corpus creation lie in the reorganization of the traditional linear and separate phases of corpus design, data collection, data annotation and corpus analysis and in the recognition of potential sources of errors during corpus creation.

Collostruction strength, i.e. the degree of attraction that a word C j exhibits to a construction C k , has been argued to be exploited in processes of online comprehension, for example, to parse ambiguous structures. There are, however, many ways to express this quantity and a large body of candidate measures can be found in the computational and corpus linguistic literature. The present study provides a comprehensive empirical evaluation of 47 competing (variants of) measures of association in order to assess their usefulness for models of sentence comprehension. To that end, the degree of adequacy of a given measure is evaluated against its performance in a task of predicting human behavior in an eye-tracking experiment that investigated the reading of a local syntactic complementation ambiguity (Kennison 2001). The analysis shows that individual measures in fact arrive at different estimations of degrees of attraction between verbs and the relevant complementation patterns, and hence differ in their power to predict human reading behavior. On the basis of the obtained results, it is suggested that minimum sensitivity (Pedersen and Bruce 1996) is best suited as an expression of collostruction strength.