← Back to Publications

Learning Language from a Large (Unannotated) Corpus

Authors:

Year: 2014
Venue: arXiv:1401.3372
Links: arXiv abstract

Summary

Proposes a fully automated, unsupervised method to extract dependency grammars and syntax-to-semantics mappings from large unannotated corpora, building on the authors' Link Grammar, RelEx, and OpenCog work. The goal is to mine the linguistic content needed to power natural-language comprehension and generation directly from raw text, avoiding both hand-coded rules and human-annotated corpora.

Relevance to Hyperon

A foundational source for the unsupervised grammar-induction program (Link Grammar / RelEx / learn lineage) of the Semantic Parsing Deep Dive — the Vepstas–Goertzel basis for later interpretable language-learning work.

Key References

• RawData source
• Semantic Parsing Deep Dive