A LONG-STANDING GOAL of artificial intelligence (AI)
is to build systems capable of understanding natural
language. To focus the notion of “understanding” a bit,
let us say the system must produce an appropriate action
upon receiving an input utterance from a human. For
Context: knowledge of mathematics
Utterance: What is the largest prime less than 10?
Context: knowledge of geography
Utterance: What is the tallest mountain in Europe?
Action: Mt. Elbrus
Context: user’s calendar
Utterance: Cancel all my meetings after 4pm tomorrow.
Action: (removes meetings from calendar)
We are interested in utterances such
as the ones listed here, which require
deep understanding and reasoning.
This article focuses on semantic parsing, an area within the field of natural language processing (NLP), which
has been growing over the last decade.
for Natural Language
˽ Natural language understanding
can be factored into mapping utterances
to logical forms (semantic parsing)
and executing logical forms to produce
˽ It is possible to learn a semantic parser
from utterance-action pairs alone by
automatically inferring the hidden
˽ The semantic parsing framework is
highly modular, consisting of an executor,
grammar, model, parser, and learner,
each of which can be developed and
improved separately, but the system is
trained to maximize end-to-end accuracy.
Semantic parsing is a rich fusion of the logical
and the statistical worlds.
BY PERCY LIANG