vacancy/SceneGraphParser

Parses dependency trees using human-written rules to extract noun phrases as entities and their semantic relations, outputting structured graphs with entity spans, lemmatized forms, modifiers, and subject-object-relation tuples. Built as a pure Python alternative to Stanford's Scene Graph Parser, it uses spaCy as the backend for dependency parsing and provides a pluggable architecture for supporting additional NLP backends. Originally developed for vision-language research bridging visual embeddings with symbolic meaning representations.

592 stars. No commits in the last 6 months.