Creation of a Russian-English Translation Program

1
Creation of a Russian-English Translation Program
Karen Shiells
2
Purpose

• Object-oriented approach
• Interactive machine translation
• Designed for aid, not independent translation
• Explore algorithms used in machine translation
• Identify grammatical obstacles to translation
• Create a base to expand later

3
Scope of Study

• Machine translation is and will be imperfect
• Modern translation uses statistical methods
• Project is limited to
• Separating base words from morphological endings
• Constructing syntax trees from source text
• Generating simple English output from tree
• Identifying words already known to the program

4
Other Research

• Part-of-speech tagging
• Uses probability to identify parts of speech
• Applied to unknown words and structures
• Complex labeling systems, beyond conventional
• Translation algorithms
• Massive dictionaries store words and information
• Aided by verb categorization
• Omit unknown words and translate without
• Usually comprehensible, but require human revision

5
Old Methods

• Direct Translation
• First method
• Rearranges sentences without parsing
• Based on rules of transfer for specific languages
• Interlingua
• From era of international languages
• Uses one representation as an intermediary
• Intermediary is usually a constructed language
• Easier to add language pairs

6
Syntactic Transfer

• Similar to interlingua
• Generates syntax tree using specific parser
• Rearranges tree to fit target structure
• Uses specific generation method to form output
• Entire algorithm specific to one language pair
• Best quality translations
• Relatively new
• Not as common in commercial software

7
Alternative Structures

• Valency
• Stores number of complements for each word
• Type of complements not specified
• Occupies less space in dictionary
• Phrase-Structure Representation
• Most familiar noun phrase, verb phrase, etc.
• Breaks sentence into superstructures
• Puts terminal symbols only in leaves
• Non-terminal symbols for branches

8
Dependency Trees

• Uses words as nodes, not just leaves
• Examples
• Verb dependent on subject
• Objects dependent on verb
• Adjectives dependent on nouns
• Prepositions vary by type of prepositional phrase
• Easier to verify agreement between words
• Occupies less space

9
Object Orientation

• Object-oriented approach allows more flexibility
• Endings, cases, and declensions are classes
• Fewer hard-coded rules
• Methods for locating dependents are in classes
• Modular design allows gradual changes
• Changes in lexical analysis do not affect parsing
• Changes in dictionary do not affect translation

10
Verb Typing

• Divides verbs into categories, for example
• Transitive
• Intransitive
• Directional or Non-directional motion
• Condenses structure storage
• Dictionary stores only type of a verb
• Particular structures taken from general
• Code can apply to general structures, not specific

11
Dictionary

• Open, save, add, remove, and search functions
• Stores
• Russian nominative
• English nominatives
• Part of speech
• Noun/pronoun attributes
• Verb types

12
Translator

• Uses transliteration for ease of testing
• Can be easily converted to Unicode Cyrillic
• Debugging output to terminal window

13
Results

• Subject, verb, direct object translated
• Subject is first nominative
• Verb matched by gender, number, and person
• Direct object is first accusative
• Adjectives matched to nouns
• Matched by case, number, and gender
• Word order not considered
• Word order should be accounted for, but aren't
• Adjectives to nearest, not matching
• Prepositional objects should be nearby

14
Conclusions

• Part-of-speech guessing could be added easily
• When a subordinate is not found, add to list
• For each unmatched word, prompt user
• Allow selection between subordinates not found
• Verb typing would be harder, but helpful
• Restricting complements makes more precise
• More efficient, not searching for all possible
• Prepositions could be associated with nouns
• Even in inflecting languages, word order matters
• Subordinates should be located by proximity
• Multiple functions use the same inflections

15
Bibliography

• Allen, James. Natural Language Understanding.
  New York Benjamin/Cummings Publishing Company,
  1995.
• Arnold, Doug, Lorna Balkan, Siety Meijer, R. Lee
  Humphreys, and Louisa Sandler. Machine
  Translation An Introductory Guide. London NCC
  Blackwell, 1994. Available Online
  http//www.essex.ac.uk/linguistics/clmt/MTbook/Pos
  tScript.
• Barber, Charles. The English Language A
  Historical Introduction. Cambridge Cambridge
  University Press, 1993.
• Beard, Robert. Russian An Interactive On-Line
  Reference Grammar. November 1, 2005. Available
  Online http//www.alphadictionary.com/rusgrammar/
  .
• Comrie, Bernard, ed. The World's Major
  Languages. Oxford Oxford University Press,
  1990.
• Hutchins, John and Harold Somers. An
  Introduction to Machine Translation. London
  Academic Press, 1992. Available Online
  http//ourworld.compuserve.com/hompages/WJHutchins
  /IntroMT-TOC.htm.