Part III: Natural Language Processing for Linguists

Part III introduces basic approaches to processing and representing language computationally.

After reading through this section, you should be able to:

• create linguistic annotations for diverse languages

• search linguistic annotations for structural patterns

• know how to quantify linguistic information

• understand the distributional hypothesis

• understand the basics of word embeddings

• know how to work with common annotation schemas

Tutorials on YouTube

Table of Contents

• Processing diverse languages
  • Introduction
  • Stanza – a Python library for processing many languages
    • Loading a language model into Stanza
    • Processing text using Stanza
    • Processing multiple texts using Stanza
  • Interfacing Stanza with spaCy
• Universal Dependencies
  • A brief introduction to Universal Dependencies
  • Basic assumptions behind Universal Dependencies
    • Nominals
    • Clauses
    • Modifiers
  • Understanding the annotation schema
  • Exploring syntactic dependencies using spaCy
  • A final word on evaluation
• Finding linguistic patterns using spaCy
  • Finding patterns using spaCy Matchers
    • Matching words or phrases
    • Matching morphological features
    • Matching syntactic dependencies
  • Examining matches in context using concordances
• Introducing word embeddings
  • Background: the distributional hypothesis and word embeddings
  • Exploring the distributional hypothesis
    • A syntagmatic perspective
    • A paradigmatic perspective
  • Learning word embeddings
    • One-hot encoding
    • Creating a neural network
    • Training a neural network
• Word embeddings in spaCy
  • Using word embeddings in spaCy
  • Visualising word embeddings
  • Contextual word embeddings from Transformers
• Working with discourse-level annotations
  • Introducing the CoNLL-U annotation schema
    • The components of the CoNLL-U annotation schema
    • Interacting with CoNLL-U annotations in Python
  • Adding discourse-level annotations to Doc objects
    • Adding information on sentence mood
    • Adding information on discourse relations
  • Converting CoNLL-U annotations into Doc objects