Lost in Translation - Found in Meaning

Samenvatting

Translating from one language into another is a complex task and meaning, undoubtly, plays a crucial role. For instance, literally translating ``having a chat'' into the Dutch phrase ``een praatje hebben'' changes the meaning and is therefore considered a bad translation. Paradoxically, often slight differences in meaning are allowed in translations: ``voetballer'' could be translated as ``player'' in football contexts. The aim of this
proposal is to find out what the conditions are that determine good or bad translations, and investigate what role meaning plays in this process.

The idea is to automatically analyze large collections of translated texts (a technique that drastically improved current machine translation systems in the last decade). What is groundbreaking in my proposal is that I will assign cross-lingual formal meaning representations to the
translations. This has never been done on a large scale for multiple languages. I expect that this computational approach will offer us a completely new view on the relationship between meaning and translation. Specifically, I aim to:

o Develop a cross-lingual semantic formalism and apply
this to very large parallel text corpora;

o Investigate under what linguistic conditions literal
translations are unsuitable;

o Determine which linguistic and contextual features
allow for meaning differences in translation.

The knowledge gained in this project can be used to advance language learning and help human translators. It could also be used to improve machine translation by developing translation quality metrics based on meaning rather than plain words.

This proposal is interdisciplinary in nature and its goals reflect my uniquely combined background in linguistics, logic, computational semantics and language technology. Building on my previous work on large-scale formal semantic analysis for English, the objectives of this project are a logical next step in this area of research, moving to cross-lingual research on big data.

Output

Wetenschappelijk artikel

Kenmerken

Projectnummer

277-89-003

Hoofdaanvrager

Prof. dr. J. Bos

Verbonden aan

Rijksuniversiteit Groningen, Faculteit der Letteren, Centre for Language and Cognition Groningen (CLCG)

Uitvoerders

Dr. L. Abzianidze, Prof. dr. J. Bos, Dr. T.D. Deoskar, Dr. K. Evang MA, Dr. K. Evang MA, H. Haagsma MSc BA, R.I.K. van Noord

Looptijd

01/08/2015 tot 31/07/2020