A metaphor is a cognitive operation involving usage of natural language and cross-domain conceptual mapping. Its ontological interest is in principle purely cognitive and linguistic, i.e. how to model what humans do when a metaphorical mapping is activated by speech or text. However, ontology-based extraction and representation of knowledge needs to make the semantics of natural language explicit, establishing the referential aspects of a natural language construction as used in dialogues, descriptions, memorisation, fiction, poetry, instructions, emotional expression, etc. – in other words, for any function of language [jakobson].
In this paper we took the bull by the horns, and straightforwardly designed an OWL ontology for metaphors, mappings, blending, etc., and populated it with data from Berkeley’s MetaNet [W15-1405]. MetaNet is the reference repository of conceptual metaphors, developed as a Semantic Wiki111https://metaphor.icsi.berkeley.edu/pub/en/, maintained through collaborative editing by multiple conceptual metaphor experts (cf. Sect. 2).
The new ontology is called Amnestic Forgery222This is a recursive name, since FORGERY IS AMNESIA is a new metaphor generated by means of the ontology itself, cf. Sect. LABEL:generating.. We deploy it as an extension of Framester333http://etna.istc.cnr.it/framester2/sparql [Gangemi:2016:FWC:3092960.3092977]
a knowledge graph represented as Linked Open Data (LOD), which integrates heterogeneous linguistic resources (OntoWordNet[fellbaum98wordnet], VerbNet [KipperDP00], FrameNet-OWL [shortBaker1998, Nuzzolese:2011:GLL:1999676.1999685], BabelNet [Navigli:2012:BAC:2397213.2397579], etc.), factual datasets (DBpedia [LehmannEtAl09], YAGO [Suchanek:2008:YLO:1412759.1412998], etc.), and foundational ontologies, by providing them a unified formal semantics. We give a brief introduction to MetaNet and FrameNet in Sect. 2 and then discuss the problem of creating a metaphor representation (Sect. LABEL:metaphor), extracting MetaNet data and their schema, and aligning them to Framester (Sect. LABEL:amnestic), as well as envisaging multiple use cases with practical examples (Sect. LABEL:generating). We also discuss referential problems of metaphorically filtered situations (Sect. LABEL:referential). We complement the paper with a survey of computational metaphor studies (Sect. LABEL:related), and conclusions.
2 MetaNet and FrameNet
MetaNet [W15-1405] is a repository of manually curated metaphors. It has its roots dug into the linguistic frames as available on FrameNet [shortBaker1998]. FrameNet is a resource containing conceptual frames, where each frame consists of a description of a situation as denoted by a text; e.g., the frame Studying depicts the situation where a student is performing an act of studying in some institution (student and institution are the “semantic roles” used to encode a studying situation).
In [W15-1405], the authors describe an automated system for extracting metaphors using the information from a manually built metaphor repository. The repository contains metaphors along with their conceptual frames, metaphor constructions, and metaphoric relational patterns. Figure LABEL:fig:metanet shows an example of a metaphor Memorization is Writing from MetaNet, which is related to another metaphor Thinking is Linguistic Activity. The other neighbouring metaphors such as Simple Ideas Are Words and Thinking Is Speaking are also connected to the same generic metaphor. The ellipses represent the source and the target frames of one of the metaphors i.e., Memorization Is Writing. ‘‘Memorization” frame already exists in FrameNet while ‘‘Writing” frame is a MetaNet specific frames.
In Sect. LABEL:amnestic we describe how the informal encoding of MetaNet has been extracted, refactored, and integrated into the Framester knowledge graph.