Computational linguistics

1
Computational linguistics

• September 9, 2002
• Overview
• Morphology
• CL applications
• inflectional vs. derivational
• morphological parsing, finite state transducers

2
Applications

• lexicons
• stemming
• generating correct surface word forms

3
Terminology

• Morpheme smallest meaning-bearing unit
• Free independent words
• Bound affix
• prefix (un-)
• suffix (-ing)
• circumfix (German ge-t sagen ? gesagt)
• infix (Bontoc Igorot -in- kayu ? kinayu)
• example from Linguistics, second edition, by
  Akmajian, Demers and Harnish, page 59

4
Concatenative vs. nonconcatenative

• Concatenative morphology combines morphemes by
  concatentation (prefixes and suffixes demonstrate
  this)
• Non-concatentative morphology combines in a
  non-concatenative manner
• circumfixes and infixes
• templatic morphology

5
Templatic morphology

• Semitic languages (Arabic, Hebrew)
• stem (root), e.g. ktb (write)
• consonant-vowel (CV) template CVCCVC (causative)
• vocalization ui (perfect passive)
• Combination consonants in stem map onto Cs in
  template, vowels in vocalization map onto Vs to
  yield surface form kuttib

6
Try it out!

• http//www.xrce.xerox.com/competencies/content-ana
  lysis/arabic/input/keyboard_input.html

7
Morphology

• Inflectional morphology
• usually does not change grammatical category
• Ex cat(N) /plural/ cats(N)
• Derivational morphology
• typically changes grammatical category
• Ex program(N) able programmable(ADJ)

8
Parsing

• refers to the recovery of structure from analysis
  of input
• often refers to the processing of sentences
• can also refer to the processing of words
• Stemming refers to the recovery of a word stem
  given a surface form of the word
• uncharacteristically un character istic
  ally

9
Lexicons

• One approach list all words
• difficult in English because some morphology is
  productive
• (table adapted from page 62 of text)

10
Options

• Not only do some affixes attach to large numbers
  of stems,
• they also attach to new words in the language
• spam, spams, spamming, spammed, spammer
• Idea encode morphological rules to generate all
  forms of words from a minimal set of word stems.
• Figure 3.4, 3.5
• Exercise 3.1, 3.2