[Comp Linguistics] Sementic Representations and Parsing

Authored by Tony Feng

Created on Nov 29th, 2022

Last Modified on Nov 29th, 2022

Intro

This sereis of posts contains a summary of materials and readings from the course CSCI 1460 Computational Linguistics that I’ve taken @ Brown University. The class aims to explore techniques regarding recent advances in NLP with deep learning. I posted these “Notes” (what I’ve learnt) for study and review only.

“Sementic” Tasks

Semantics & Syntax

Syntax = the study of form
Semantics = the study of meaning

The general goal is to describe precisely how we get from form to meaning.

Semantics & Pragmatics

Semantics = “sentence meaning” = tied to form/grammar
Pragmatics = “speaker meaning” = tied more generally to context

In NLP, we tend to blur the distinction, and focus on specific tasks. For example, “Siri, I can’t hear the music” should be received as an instruction.

Executable Forms

Explicit representation of natural language in formal language
- Question Answering over Databases: SELECT * FROM flights WHERE city = “BOS”
- Robotics: move(a) ^ pick_up(obj)
- Digital Personal Assistants: set_alarm(07:00, GMT-5)

Natural Language Inference

Also called Recognizing Textual Entailment (RTE)
Given a premise $p$ and a hypothesis $h$, does $p$ entail $h$?
Now widely used as a general-purpose evaluation task for systems of “understanding”

Formal Sementics

Model Theory

A premise $p$ entails a hypothesis $h$ iff, in every possible world in which $p$ is true, $h$ is also true.

Truth Conditions and Truth Values

Truth Conditional Semantics: the meaning of a sentence is its truth conditions
- Understanding “I have a pen in my pocket” does not require knowing whether I have a pen in my pocket.
Contrast with Denotational Semantics: the meaning of a sentence is its truth value
- “3 is half of 6” == “one week has 7 days”
Formal semantics uses truth conditions