geniatagger

Geniatagger annotates English biomedical sentences with linguistic labels—base forms, part-of-speech tags, chunk tags, and named entity tags—to support processing of MEDLINE abstracts.

Key Features:

• Base Forms: Identifies the base forms of words in a sentence.
• Part-of-Speech (POS) Tags: Assigns part-of-speech tags to each word.
• Chunk Tags: Labels chunks or phrases to identify meaningful multiword units.
• Named Entity Tags: Recognizes and tags named entities including gene names, diseases, and chemicals.
• Bidirectional Inference for Sequence Labeling: Employs a bidirectional inference algorithm tailored for sequence labeling tasks such as POS tagging, named entity recognition, and chunking.
• Decomposition and Highest-Probability Sequence: Enumerates possible decomposition structures and identifies the highest-probability sequence together with its corresponding structure.
• Easiest-First Decoding: Incorporates an easiest-first decoding algorithm that achieves performance comparable to full bidirectional inference with reduced computational cost.
• Performance Comparison: Demonstrates performance that outperforms traditional unidirectional inference and is on par with kernel support vector machines.

Scientific Applications:

• MEDLINE Abstract Annotation: Annotates MEDLINE abstracts and similar biomedical documents for downstream text mining.
• Biomedical Named Entity Recognition: Extracts domain-specific entities such as gene names, diseases, and chemicals from biomedical text.
• Linguistic Preprocessing for NLP: Provides lemmas, POS tags, and chunk labels to support downstream biomedical natural language processing pipelines.

Methodology:

Uses bidirectional inference for sequence labeling by enumerating decomposition structures and selecting the highest-probability sequence, and employs an easiest-first decoding algorithm for efficient decoding.