engleski

Separating (non-)figurative weeds from wheat

A number of approaches have been developed in cognitive linguistic research that promise to recognize and/or identify figurative expressions in discourse, chiefly conceptual metaphors, but also conceptual metonymies (Berber Sardinha 2008, 2012 ; Goatly 1997 ; Kövecses et al 2015 ; Markert & Nissim 2006 ; Nissim & Markert 2003 ; Shutova & Sun 2013 Shutova, Teufel & Korhonen 2013 ; Stefanowitsch 2004, 2006 ; Steen 2007 ; Steen et al 2010, Wallington et al 2003). Some of these have advertised themselves as being able to achieve high success rate, in certain cases even surprisingly high ones, in the more or less automatic and (un)supervised retrieval of figurative expressions in comprehensive texts of various size or in corpora. While they widely differ with respect to the complexity of the formal infrastructure underlying them, such as the algorithm they rely on in the process of training, the knowledge base, etc., they seem to have one thing in common – they basically aim at separating figurative wheat from non-figurative weeds. In other words, researchers are intent on getting their hands on figurative expressions in as direct fashion as possible. It is, however, possible to envisage an approach going in a diametrically opposed direction. This is exactly what this presentation is all about – turning our back to figurative wheat and attending to non-figurative weeds first, identifying it and subsequently eliminating it from further consideration. This may at first sight seem to be a counterintuitive proposal, considering the proportion of the non-figurative weeds in running texts or corpora, i.e. in view of the huge number of literal expressions, exceeding many times the number of potential figurative expressions surrounded by the former. While it should certainly be allowed as an interesting, and perhaps entertaining intellectual enterprise, and certainly legitimate if we do not want to leave any stone unturned, i.e. leave no alley of potential research unchecked, it would no doubt be considered by many to be a waste of time, a sort of doing science for its own sake, but hardly promising any worthwhile insights. However, we claim in this presentation on the basis of a series of small-scale case studies involving English, German, Croatian and Hungarian material (e.g. philosophy, oasis, patient) that by engaging in this unusual type of exercises, approaching conceptual metaphors in a negative way we can achieve a success rate that is comparable to the best ones described in the relevant literature, while making use of a considerably leaner tool, basically just a hierarchically organized, FrameNet- or WordNet-like knowledge database, the decisive factor being whether or not a given expression is surrounded in its more or less immediate context by a sufficient number of lexical items denoting concepts belonging to the same frame that the key item belongs to. The likelihood of its being nonfigurative weed, i.e. not a metaphorical expression, is directly proportional to the number of co-occurring items from the same frame. The material retrieved this way can be submitted to some further tests in order to refine the results. The situation with conceptual metonymies seems to be just the opposite, they can only be approached in a positive way, i.e. not by automatically excluding potential candidates. Here a low number of co-occurring items from the same frame indicates that it is quite likely to be metonymic weed. However, in order to determine that a given item is indeed used metonymically, we expect to find additional semanto-syntactic anomaly, i.e. misalignment between the valency frame of the item in question and of the valency frame of the predicate or an argument, depending on the situation. On top of that, it seems that a metonymy typology or hierarchy that is harnessed with an ontology-based FrameNet is necessary, so that a metonymy can be identified as such. This is illustrated on lexical items primarily denoting plants and animals as well as on geographical nouns. This difference between metaphors and metonymies – their tendency to co-occur or to fail to co-occur with items denoting concepts from the same frame is only to be expected in light of how the two phenomena are usually characterized in literature, i.e. as intra- or inter-domain phenomena.

figurative language ; metaphor ; metonymy ; figurative language recognition ; corpus linguistics ; ontology

nije evidentirano