DelphinOslo

DELPH-IN at the University of Oslo

The Oslo Machine Translation Research Group is part of the Department of Nordic Studies and Linguistics at the University of Oslo (UiO). The group is the co-ordinator of the national LOGON initiative, a five-year basic research project working towards high-quality Norwegian -- English machine translation. The LOGON architecture incorporates both deep LFG and HPSG grammars, more shallow NLP components (including a PoS tagger, chunker, and NE recognizer), as well as stochastic processes for disambiguation and robustness at all processing levels. The group currently has seven members, of which four are graduate students, and participates in an array of national and international training and research initiatives.

One of the central research questions in LOGON is on the utility of current language technology for machine translation, as the project is essentially assembling its translation pipeline from general-purpose NLP resources, many of them from the DELPH-IN repository. While Norwegian analysis is accomplished using the (proprietary) LFG resource NorGram, the transfer and English realization components are both open-source. LOGON adopts a semantic transfer approach and uses MRS as its meaning representation and interface language. The transfer module was realized as a general-purpose MRS rewriting engine (in most respects a generalization of the older `munging' machinery available in the LKB) and is publicly available. Target language realization uses the LinGO English Resource Grammar (ERG) and the generation component in the LKB. In close collaboration with partners at Sussex, Stanford, and Cambridge Universities, the LKB generator was significantly improved in its efficiency and robustness for use in LOGON. On-going research at UiO revolves around aspects of (cross-linguistic) meaning representation in MRS, the acquisition of transfer knowledge, and further refinements of the English realization module (including work on the LinGO ERG and the underlying software). A particular focus in this second phase of the LOGON project (as of early in 2005) is on stochastic disambiguation in all three phases of the MT pipeline.