Skip to content
This repository has been archived by the owner on Jan 10, 2023. It is now read-only.
/ sling Public archive

SLING - A natural language frame semantics parser

License

Notifications You must be signed in to change notification settings

google/sling

SLING - A natural language frame semantics parser

The SLING Project has moved

The SLING project has moved to https://github.com/ringgaard/sling.

Please refer to the above repo for the latest developments on SLING.

Background

The aim of the SLING project is to learn to read and understand Wikipedia articles in many languages for the purpose of knowledge base completion, e.g. adding facts mentioned in Wikipedia (and other sources) to the Wikidata knowledge base. We use frame semantics as a common representation for both knowledge representation and document annotation. The SLING parser can be trained to produce frame semantic representations of text directly without any explicit intervening linguistic representation.

The SLING project is still work in progress. We do not yet have a full system that can extract facts from arbitrary text, but we have built a number of the subsystems needed for such a system. The SLING frame store is our basic framework for building and manipulating frame semantic graph structures. The Wiki flow pipeline can take a raw dump of Wikidata and convert this into one big frame graph. This can be loaded into memory so we can do fast graph traversal for inference and reasoning over the knowledge base. The Wiki flow pipeline can also take raw Wikipedia dumps and convert these into a set of documents with structured annotations extracted from the Wiki markup. This also produces phrase tables that are used for mapping names to entities. There is a SLING Python API for accessing all this information and we also have a bot for uploading extracted facts to Wikidata.

The SLING Parser

The SLING parser is used for annotating text with frame semantic annotations. It is a general transition-based frame semantic parser using bi-directional LSTMs for input encoding and a Transition Based Recurrent Unit (TBRU) for output decoding. It is a jointly trained model using only the text tokens as input and the transition system has been designed to output frame graphs directly without any intervening symbolic representation.

SLING neural network architecture.

The SLING framework includes an efficient and scalable frame store implementation as well as a neural network JIT compiler for fast training and parsing.

A more detailed description of the SLING parser can be found in this paper:

More information ...

Credits

Original authors of the code in this package include:

  • Michael Ringgaard
  • Rahul Gupta
  • Anders Sandholm