Guiding Theme C3: Deep Learning embeddings for adaptive language processing

Deep Learning embeddings are low-dimensional vector representations of words and phrases. They are able to capture both semantic and syntactic regularities of words and phrases. Embeddings are particularly useful as features for adaptive language processing, because they can be learned in an unsupervised fashion from large amounts of domain-specific text data, as well as from lexical knowledge bases such as WordNet. For an example of how to jointly learn embeddings from text and from a knowledge base for the task of relation extraction, read Connecting Language and Knowledge Bases with Embedding Models for Relation Extraction by J. Weston, A. Bordes, O. Yakhnenko and N. Usunier, in: Proc. of the Conference on Empirical Methods in Natural Language Processing (EMNLP 2013).

A Ph.D. project that primarily follows this guiding theme will develop embedding models to align textual information to linked lexical resources for the task of semantic role labeling in German. This is a more general problem than relation extraction:  Semantic role labeling is the task of annotating actions and events and their participants in text, thus providing answers to the question “who does what to whom”.  Domain- and genre-adaptation of such tools is an open-research issue, and the tools for processing the German data are very scarce. The student will closely collaborate with the guiding theme A2: Identification  of  complex  event  structures  in  discourse on this topic.

Poster (in German)

Example thesis topics

  • Joint learning of embeddings from text and linked lexical resource for semantic role labeling.
  • Deep Learning embeddings as features in genre and domain adaptive multi-document summarization.
  • Minimally supervised Deep Learning architectures for genre and domain adaptive multi-document summarization.


Ronan Collobert, Jason Weston, Léon Bottou, Michael Karlen, Koray Kavukcuoglu, Pavel P. Kuksa:
Natural Language Processing (Almost) from Scratch. Journal of Machine Learning Research 12: 2493-2537 (2011).


A A A | Drucken Print | Impressum Impressum | Sitemap Sitemap | Suche Search | Kontakt Contact | Webseitenanalyse: Mehr Informationen
zum Seitenanfangzum Seitenanfang