19622

WiSe 13/14: Natural Languages Processing

Mariana Lara Neves

Comments

Seminar "Natural Language Processing (NLP)"

Natural language processing (NLP) is the task of automatically processing textual documents written on natural language (e.g., English). Applications of NLP includes a variety of tasks, such as document classification (assignment of a predefined list of topics to documents), named-entity recognition (identification of particular entities in the text, such as people, place, etc.), relationships and events extraction (between two or more entities, including identification of the relationship type, aslo called slot-filling), document or multi-document summarization (automatic generation of summaries based on one or more original documents) and question answering (automatic processing of questions posed on natural language and presenting a simple answer or summary in return).

The biomedical domain present some particular challenges to NLP, such as the high amount of scientific literature (millions of abstracts in Medline and thousands of full texts in PubMed Central), a variety of named entities (species, genes/protein, diseases, chemicals, drugs, etc.), the variability of their nomenclature (new genes or drugs are always being discovered ore developed), the complexity of the relationships between these entities (protein-protein interactions, drug-drug interactions, genedisease, gene expression, gene regulation, etc.). However, it also provides good resources to support the above tasks, such as a variety of ontologies (cf. BioPortal, UMLS), manually annotated corpora (e.g., Genia, BioCreative, etc.), frequently community-based challenges and shared tasks (e.g, BioCreative, BioNLP, etc.) and on-line biological databases with manually curated data (EntrezGene, CTD, etc.).

For this seminar, I have selected a couple of papers for each of the tasks above. These publications involve a variety of entities, relationships and ontologies and focus on both biological and medical domains. Finally, methods proposed in the papers below range from rule-based to supervised learning approaches.