Automatic extraction of lexico-semantic information on shape words

Abstract

Background: Metaphors are complex linguistic forms that transcend their meanings from the original sense of the words they derive from. For instance, language users describe, in the expression "the political system is well-oiled", a political system with the same terminology for machinery. This research analyzes metaphors by taking into account the Lakoffian approach, which foresees the distinction of two domains, a source (machinery, as for the example above) and a target (political system). The investigation proposes different interpretations of Lakoff's Embodiment Theory, and it expands from the research of the Taiwanese School (comprising the work of Ahrens, Chung and Huang on metaphors), which was able to identify source domain information from designated targets and corpora and to map sources and targets into the knowledge repository SUMO via upper concepts. The following research is inspired by the work of the Taiwanese School, but it diverts from it in that sources were assumed to be SHAPE, words were analyzed by means of corpora as well as a lexico-semantic and frame-oriented analysis, and the retrieved metaphorical meanings were mapped to SUMO ontology via upper as well as mid-level concepts. The goal of the investigation is to study: a) whether shape words have metaphorical meanings and b) whether the ontological representation of metaphor meanings is satisfactory through the use of only upper concepts. The shape words considered were circle, triangle, square, point,line and shape. Method: The semi-automatic search followed different steps across the consulted sources. First, a lexico-semantic analysis of shape words as POS noun was conducted on the Princeton WordNet. The shape words were analyzed as mapped to and mapped as synsets. Their hyponymic-hypernymic structure was also studied through the introduction of semantic classes. Shape words were then analyzed in the context of frames in FrameNet, an open source which also provides lexico-semantic information on shape words combined with ad hoc ontological concepts inscribed in the frames. Finally, shape words were investigated within the context of tropes (including collocations, idioms, phrasal verbs), as retrieved from different corpora and dictionaries across several languages (Sino-Tibetan, Germanic and Romanic ones). WordNet and FrameNet were accessed through the NLTK module in Python; the corpora and dictionaries were searched online and in paperback format. Results: Each of the three adopted methodological approaches enabled the retrieval of metaphors of shape words. The targets were then mapped to SUMO by means of upper as well as mid-level concepts. Despite the evident difference of purpose of the repositories, the targets of the retrieved metaphors partially share upper and middle concepts among the three. In particular, the distinction between CONCRETEOBJECT, ABSTRACTOBJECT and CONCRETEEVENT, ABSTRACTEVENTS for the targets of SHAPE counted among the mostly shared mapping. Conclusions: This investigation aims at laying the groundwork for an ontology of shape, i.e. for a representation of knowledge on and related to SHAPE based upon universally accepted concepts. It also aims at providing diverse approaches to the retrieval and disambiguation of metaphorical meanings.