Contextualized Usage-Based Material Selection

Contextualized Usage-Based Material Selection Dirk De Hertog, Piet Desmet ITEC, KU Leuven, imec Etienne Sabbelaan 51, 8500 Kortrijk, Belgium {dirk.dehertog, piet.desmet}@kuleuven.be Abstract In this paper, we combine several NLP-functionalities to organize examples drawn from corpora. The application’s primary target audience are language learners. Currently, authentic linguistic examples for a given keyword search are often organized alphabetically according to context. From this, it is not always clear which contextual regularities actually exist on a syntactic, collocational and semantic level. Showing information at different levels of abstraction will help with the discovery of linguistic regularities and thus improve linguistic understanding. Practically this translates in a system that groups retrieved results on syntactic grounds, after which the examples are further organized at the hand of semantic similarity within certain phrasal slots. Visualization algorithms are then used to show focused information in phrasal slots, laying bare semantic restrictions within the construction. Keywords: Language Learning, Constructions, Corpus-based examples, Distributional Analysis 1. Introduction We present the integration of NLP-functionalities to help with the selection of contextualized usage-based examples to assist language learners and other end users that could benefit from a distributional linguistic analysis. Usage- based material, in the form of examples drawn from corpora, like the Keyword in Context (KWIC) method, are thought of as valuable resources for studying languages. KWIC allows for the user to search for a keyword, retrieves a set of examples from a corpus and presents them ordered alphabetically according to context words. However, from examples alone, it is not always immediately clear which contextual regularities exist at the syntactic, collocational and semantic level. The lack of such additional structure and the required effort to derive it single-handedly have proven demotivational in practical settings. We propose to use several NLP-methods to provide additional information by which the examples could be further organized in a meaningful way. This would add value as it would clarify linguistic regularities at a glance. Practically this translates in a system that looks for a certain keyword and groups the retrieved examples first according to syntactic grounds, then to semantic similarity within certain contextual phrasal slots. By doing so, we aim to lessen the gap between corpus- based examples and cognitive understanding of linguistic regularities with the end-user. We start from the basic premise that learning presumes understanding (Krashen, 1981; Chapelle, 1996); in this respect, we argue that full disambiguation facilitates linguistic understanding and thus enhances word knowledge and improves vocabulary retention. The procedure is inspired by collostructional analysis (Anatol and Gries, 2003), which treats syntactic constructions as a disambiguating factor. For the semantic understanding of words, word sense is arguably linked to the actual construction in which it is used. Conversely, the intended word sense presupposes correct syntactic and/or collocational use for it to be understood. A retrieval that links both types of information should therefore prove highly informative. We first explain what we interpret as cognitive understanding from a pedagogical and linguistic point of view. We then offer technical operationalizations to achieve the intended semantic …

Content truncated to 500 words