Idioms in example-based machine translation

Abstract

Machine Translation (MT) has progressed in parallel with idiom research throughout the years, since they are both interdisciplinary fields. However, most researchers and MT systems regard idioms as a thorn in MT';s flesh. When it comes to idiom translation, it becomes really a difficult task for human translators, let alone for MT systems. The construction of an idiom database is complex and time-consuming, since there are not idiom corpora widely available and must be either manually constructed or consist of real examples carefully filtered. We incorporated both cases into our data sets and proved that idiom processing based on syntactic patterns of the topological field model is thoroughly feasible.