Abstract
Syntactic annotation is an indispensable input for many semantic NLP applications. For instance, Semantic Role Labelling algorithms almost invariably apply some form of syntactic parsing as preprocessing. The categories used for syntactic annotation in NLP generally reflect the formal patterns used to form the text. This results in complex annotation schemes, often tuned to one language or domain, and unintuitive to non-expert annotators. In this paper we propose a different approach and advocate substituting existing syntax-based approaches with semantics-based grammatical annotation. The rationale of this approach is to use manual labor where there is no substitute for it (i.e., annotating semantics), leaving the detection of formal regularities to automated statistical algorithms. To this end, we propose a simple semantic annotation scheme, UCCA for Universal Conceptual Cognitive Annotation. The scheme covers many of the most important elements and relations present in linguistic utterances, including verb-argument structure, optional adjuncts such as adverbials, clause embeddings, and the linkage between them. The scheme is supported by extensive typological crosslinguistic evidence and accords with the leading Cognitive Linguistics theories.
Original language | English |
---|---|
State | Published - 2013 |
Event | 10th International Conference on Computational Semantics, IWCS 2013 - Potsdam, Germany Duration: 19 Mar 2013 → 22 Mar 2013 |
Conference
Conference | 10th International Conference on Computational Semantics, IWCS 2013 |
---|---|
Country/Territory | Germany |
City | Potsdam |
Period | 19/03/13 → 22/03/13 |
All Science Journal Classification (ASJC) codes
- Computer Networks and Communications
- Information Systems
- Computer Science Applications