Distributional Semantic Classes for Hypernymy Extraction and Denoising

A collection of induced word senses, a hypernym database, and clustered sense labels supporting hypernymy extraction research. The data was created by Alexander Panchenko of Universität Hamburg to demonstrate a method for improving hypernym extraction using distributionally-induced semantic classes. It includes semantic classes at two granularities, coarse and fine-grained, as described in the associated paper.

Use Cases

• Denoising automatically extracted hypernym relations based on distributionally similar senses.
• Inducing sense-aware semantic classes using distributional semantics.
• Performing label propagation to infer missing hypernyms for cluster terms.
• Benchmarking taxonomy induction methods using the provided state-of-the-art results.

Strengths

• Methodology validated by a large-scale crowdsourcing study showing improved precision and recall.
• Achieved state-of-the-art results on a benchmarking dataset for taxonomy induction.
• Includes semantic classes at two distinct granularities (coarse and fine-grained).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file formats are unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

Alexander Panchenko, Universität Hamburg

Collection Method

Induced via distributional semantics and crowdsourced validation, as described in the paper.