1 year ago

science podcast free energy principle free energy free energy principle active inference active inference theory

A Neuro-Symbolic Approach to Language Understanding: Using an Ontology to
Create A Language Model Adam Pease https://www.adampease.org/
https://www.ontologyportal.org/ Adam's YouTube channel: / @adampease Formal
linguistic semantics has a long tradition of using logic as a way to capture
the objective semantics of language. We know an interpretation is wrong if
apparently consistent sets of sentences yield a contradiction when a
deterministic method is applied. But creating those models has been largely
limited to academic works in which a theory is embodied in rules on paper.
Computational models, in contrast, have had very shallow or approximate
interpretations of semantics. In this current work, I describe how to generate
a very large set (1TB or larger) set of language-logic pairs that is then used
to train a neural-network based language model. Because the sentences are
created synthetically, we can use a representation that can also generate a
consistent logical interpretation of the sentence automatically. Because this
model uses the Suggested Upper Merged Ontology (SUMO), tens of thousands of
concepts are used. Because SUMO uses a higher order logic, we can generate
sentences far more interesting than can be captured in simple triples. We can
use quantifiers, modals, epistemic and teleological operators and arbitrary
levels of embedded clauses. Because SUMO has been mapped to all of the WordNet
lexicon, we can also freely include synonyms for all concepts, if they exist.
This leads to a data set that can cover a diverse and interesting subset of
English. Active Inference Institute information: Website:
https://activeinference.org/ Twitter: https://twitter.com/InferenceActive
Discord: https://discord.gg/8VNKNp4jtx YouTube: / activeinference Active
Inference Livestreams: https://coda.io/@active-inference-ins...