Cross-Lingual Annotation Projection Models for Role-Semantic by Eingereicht von Sebastian Pado

By Eingereicht von Sebastian Pado

Show description

Read or Download Cross-Lingual Annotation Projection Models for Role-Semantic Information PDF

Similar nonfiction_6 books

The Diversity of Social Theories (Current Perspectives in Social Theory, Volume 29)

Because the time whilst Talcott Parsons pursued the undertaking of 1 overarching basic concept of society, the panorama of social concept has drastically replaced, and the pluralism and multidimensionality elevated greatly. this present day, with such a lot of assorted techniques in and to social thought, and a number of methods of defining and describing their courting to and relevance for the social sciences, there was a turning out to be possibility of range and pluralism tipping into fragmentation, making the chance of social scientists and sociologists having the ability to converse with the expectancy of achieving a few type of knowing, ever much less most likely.

Protective Measures for Housing on Gas-contaminated Land: (BR 414)

A pragmatic advisor to stable perform for the detailing and building of passive soil gasoline protecting measures for brand new and present residential improvement. the most gases thought of are methane, carbon dioxide and combos of the 2.

Extra info for Cross-Lingual Annotation Projection Models for Role-Semantic Information

Sample text

These imbalances are problems which EUROPARL shares with many other, commonly used corpora such as the Penn Treebank (Marcus et al. 1993). , as opposed to noise-prone internetharvested corpora), and is commonly used for multilingual research such as statistical Machine Translation (Koehn 2005). In addition, we find that our application of EUROPARL – annotation projection of semantic class and role information – is comparatively robust to EUROPARL’s imbalances. In particular, the induction of semantic classes (Chapter 4) only requires an open domain corpus, which EUROPARL provides.

A high degree of concept-level parallelism is therefore crucial for the feasibility of annotation projection. Fortunately, it appears that at least for some levels of linguistic analysis, a fairly high degree of parallelism can be achieved, provided that the description is situated at an appropriate level of granularity. 2. Instance-level Parallelism While concept-level parallelism is an important property of linguistic theories, it is insufficient for any investigations which are concerned with actual multilingual data such as parallel corpora: Concept-level parallelism is only concerned with the inventory of categories, not with the analysis of concrete utterances.

This makes their integration into more general statistical processing frameworks problematic, since this usually means expressing the output as the optimal solution to a joint optimisation problem. Statistical alignment models, on the other hand, do have an optimisation criterion, but require a large bilingual corpus for training, and external linguistic information is difficult to integrate. Both model families share the problem of choosing from a wealth of possible instantiations. For the purpose of this thesis, we chose to use a statistical word alignment model.

Download PDF sample

Rated 4.46 of 5 – based on 25 votes