Universal Conceptual Cognitive Annotation (UCCA)

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 228–238, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Universal Conceptual Cognitive Annotation (UCCA) Omri Abend∗ Institute of Computer Science The Hebrew University omria01@cs.huji.ac.il Ari Rappoport Institute of Computer Science The Hebrew University arir@cs.huji.ac.il Abstract Syntactic structures, by their nature, re- flect first and foremost the formal con- structions used for expressing meanings. This renders them sensitive to formal vari- ation both within and across languages, and limits their value to semantic ap- plications. We present UCCA, a novel multi-layered framework for semantic rep- resentation that aims to accommodate the semantic distinctions expressed through linguistic utterances. We demonstrate UCCA’s portability across domains and languages, and its relative insensitivity to meaning-preserving syntactic variation. We also show that UCCA can be ef- fectively and quickly learned by annota- tors with no linguistic background, and describe the compilation of a UCCA- annotated corpus. 1 Introduction Syntactic structures are mainly committed to rep- resenting the formal patterns of a language, and only indirectly reflect semantic distinctions. For instance, while virtually all syntactic annotation schemes are sensitive to the structural difference between (a) “John took a shower” and (b) “John showered”, they seldom distinguish between (a) and the markedly different (c) “John took my book”. In fact, the annotations of (a) and (c) are identical under the most widely-used schemes for English, the Penn Treebank (PTB) (Marcus et al., 1993) and CoNLL-style dependencies (Surdeanu et al., 2008) (see Figure 1). ∗ Omri Abend is grateful to the Azrieli Foundation for the award of an Azrieli Fellowship. Underscoring the semantic similarity between (a) and (b) can assist semantic applications. One example is machine translation to target languages that do not express this structural distinction (e.g., both (a) and (b) would be translated to the same German sentence “John duschte”). Question An- swering applications can also benefit from dis- tinguishing between (a) and (c), as this knowl- edge would help them recognize “my book” as a much more plausible answer than “a shower” to the question “what did John take?”. This paper presents a novel approach to gram- matical representation that annotates semantic dis- tinctions and aims to abstract away from specific syntactic constructions. We call our approach Uni- versal Conceptual Cognitive Annotation (UCCA). The word “cognitive” refers to the type of cate- gories UCCA uses and its theoretical underpin- nings, and “conceptual” stands in contrast to “syn- tactic”. The word “universal” refers to UCCA’s capability to accommodate a highly rich set of se- mantic distinctions, and its aim to ultimately pro- vide all the necessary semantic information for learning grammar. In order to accommodate this rich set of distinctions, UCCA is built as a multi- layered structure, which allows for its open-ended extension. This paper focuses on the foundational layer of UCCA, a coarse-grained layer that rep- resents some of the most important relations ex- pressed through linguistic utterances, including ar- gument structure of verbs, nouns and adjectives, and the inter-relations …

Content truncated to 500 words