I'm Michael Ringgaard, founder and CEO of Ringgaard Research. Ringgaard Research is a not-for-profit company that develops and maintains the SLING system as well as doing research relating to knowledge management and natural language processing. After 13 year as a software engineer at Google Research working with natural language processing, I resigned from Google in September 2020 to start for myself.

History

I started development of SLING at Google back in 2016, although the earliest prototypes date back to 2014. I wanted to design a system that was efficient for working with knowledge bases as well as general semantic annotation of text. We used this for building a general frame semantic parser. The parser was made open source in 2017 under an Apache license and published on GitHub, together with a paper describing the parser.

The parser used an academic corpus (OntoNotes) which was not very useful from a practical point of view, so in 2019 I started working on a way of using Wikipedia and Wikidata for constructing training data for the parser. Part of this work involved taking Wikidata, which is the knowledge base behind Wikipedia, and converting it into SLING frames. These are then combined into a SLING store which allows you to efficiently read the whole knowledge base into memory to do fast inference over the entire knowledge graph.

SLING

Since I left Google in September 2020, I have been working on making a pipeline that collects updates to Wikidata in real-time, as well as downloads and processes Wikipedia in many different languages. These datasets are reconciled with other data sources into one single federated knowledge base and published every night. This also produces alias tables, cross references, and search indices, which are used for searching and navigating the knowledge base.