Semantically Relatable Sequences in the Context of Interlingua Based Machine Translation
msra(2008)
摘要
In Interlingua based machine translation source lan- guage sentences have to be converted to a semantic representation- often a meaning graph with concept nodes and semantic relations- from which the target language sentences are produced. We argue that to- wards the meaning graph generation, a necessary step is to detect the sentence constituents which participate in semantic linkages. Semantic linkages are of the form, relation(entity1, entity2). Before creating the semantic linkages it is necessary to detect (entity1, en- tity2) which we call a Semantically Relatable Sequence (SRS). SRS computation makes use of NLP tools like the parser and NLP resources like the WordNet and OALD. Once SRSs are generated, we have covered a considerable distance to the translation. For evaluating the efficacy of the SRS generation sy stem, we show that the system accurately produces the shallow semantic role labels of 92,310 sentences of the FrameNet corpus. It is emphasized that the system ultimately is designed to produce deep semantic role labels in the framework of Universal Networking Lan- guage (UNL) which is a recently proposed interlingua. An important by-product of the work is the fact that the costly resource of semantically role labeled corpus can be obtained at least partially automatically through our system.
更多查看译文
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络