Towards a Morphological Analyzer for Old Norse

1
Towards a Morphological Analyzer for Old Norse
2
Introduction

• Goal a computer program that analyzes
  morphological structure of Old Norse words and
  generates declension tables
• Two analyzers A1, A2 both output all possible
  declension paradigms for inputted word
• A1 input headwords from a dictionary database or
  manual input
• A2 input inflected words from saga texts
• Show sample query A1 without details

3
Broader Context (1)
Input (mss)
Marked-up transcript
Normalized text
Analyzer
Output (declension tables)
4
Broader Context (2)

• CLHT Project
• Scandinavian Section, UCLA (Prof. Timothy
  Tangherlini)
• Developing Old Norse morphological analyzer
• Det Arnamagnæanske Institut, Københavns
  Universitet (Matthew Driscoll)
• XML markup of Old Norse texts

5
Computational Environment

• Written in Perl
• Database MySQL
• Server Apache
• Running on Linux machine
• http//ecampusdev.humnet.ucla.edu/curban

6
Linguistic Environment

• Zoegas Dictionary of Old Norse - Icelandic
• augmented with additional headwords from the Old
  Norse dictionary project, Ordbog over det norrøne
  prosasprog(ONP) at Københavns Universitet
• Fornrit normalization
• Verification of performance comparison with
  forms in Bower (1994), Gordon (1956)
• Focus on non-poetic lexicon

7
Analyzer Structure (General)
In (head)word
MySQL database

• Analyzer
• Find root
• Find endings
• Apply sound changes

Out declension(s)
8
Analyzer Structure (Database)

• Tables exist for
• adjectives (regular endings, exceptions)
• articles (free, suffixed)
• dictionary
• nouns (regular endings, exceptions)
• possessive pronouns
• verbs (regular endings, exceptions, anomalous,
  strong_ablaut)

9
A1 Structure (Specific 1)

• Input
• Head word
• Declension information
• Part of speech
• Translation

10
A1 Structure (Specific 2-1)

• A1 pseudocode (nouns)
• Translate declension info into MySQL format
• Extract most likely endings from words in
  declension info
• Determine root of head word
• Create MySQL statement

11
A1 Structure (Specific 2-2)

• Receive all declension paradigms that fit
  declension information
• Apply regular sound changes
• Replace exceptional forms
• Output results
• Show sample queries with details

12
Outlook (1) Accomplishments

• Zoega in electronic, parsable format
• Show sample of complex Zoega entry
• A1 outputs paradigms for all parts of speech in
  Zoega

13
Outlook (2) A1 Performance
14
Outlook (3) Next Steps

• Improve A1 performance general, compound words,
  etc.
• Expand databases of exceptions
• Improve verification method
• Implement A2 beyond experimental stage
• Connect analyzers to XML-tagged text