Cross-Language Information Retrieval (CLIR)

1
Cross-Language Information Retrieval (CLIR)

• 614 Information Retrieval Theory
• YooJin Ha
• 12/17/2004

2
Content

• Definition
• Users need
• General problem
• Organization of CLIR model
• Matching methods
• Experiments
• Future expectation

3
What is CLIR?

• CLIR (Cross Language Information Retrieval) is a
  retrieval system that operates with queries in
  one language to retrieve documents in other
  languages. -- allow users to access information
  written in the users languages of choice.
• CLIR can be used to enhance the ability of users
  to search and retrieve documents in many languages

4
Who will be users of a CLIR system?

• 1. Bilingual users who have good reading skills
  in their second language may have poor language
  productive skills and thus cannot express their
  information need in their 2nd language as well as
  they can in their first language.
• 2. Monolingual users who have interest or need
  such information for their research but some
  reality issues (such as cost and time before
  coming resources for a full translation.
• (Ogden et al., 1999 Abdelali et al., 2002)

5
Users need in query formulationAccording
to my survey (about OPAC, WorldCat usage), most
users want to search with their own language.
Americans take it for granted that they can
present the system with queries in English. This
then forces all users to enter queries the same
way for example, Korean would want to present
queries in their native language, just as Chinese
and Japanese would also want. If WorldCat can
morph into a cross language system, then
catalogers and the general public would have
greater access and use of this system.
6
Multilanguage examples

• Rutgers University Libraries WorldCat
• Arctos Interactive Multilingual Search with Ursa

7
History

• 1964 International Road Research Documentation
  system used controlled vocabulary thesaurus in
  English, French, and German
• 1970 Salton conducted to retrieve other
  languages than a query, augmenting the SMART
  system using with hand-constructed bilingual
  term, thesauri.
• 1978 ISO adopted ISO standard 5964 on
  constructing of multilingual thesauri
• (Oard Diekema,
  1998)

8
Problems

• Various structures by different languages. (I.e.
  stemming is widely used to increase recall
  performance)
• Difficulties in normalization (I.e. Korean- too
  morphological, so it is difficult to decompose)
• CLIR system must be adapted to the
  characteristics of whichever languages it will
  use.
• Disambiguation (difficult to find a correct term
  
• often several possible meanings for the term
  in that query)
• None of them has been fully successful yet.

9
Some solutions

• In stemming, statistical dependence based on
  co-occurrence of terms (I.e. Spanish has more
  forms of each verb)(Croft al et, 1996)
• Dictionary combined with a mechanism to string
  word meaning together (I.e. German) (Sheridan
  Ballerini, 1996)
• Automatic query enrichment by using local
  feedback and local context analysis (Ballestores
  Croft, 1998)

10
Machine translation (MT)

• Many people easily think the MT can be applied to
  any system. It often makes errors and only few
  available examples of research done on MT with
  specific regard to CLIR.
• MT only can work well for specific domains, such
  as containing specific terminology. (Oard Dorr,
  1996)
• SYSTRAN Language Translation Technology
• AltaVista - Babel Fish Translation

11
Organization of CLIR system - Oard (1997)
system model

• 1. Document pre-processing (collection)
• 2. Query formulation
• 3. Matching
• 4. Selection
• 5. Examination
• 6. Relevance feedback
• (cited in Oard Diekema, 1998)

12
Matching strategies (Oard Diekema, 1998)
13
Organization of CLIR system

• Matching
• Cognate matching
• Translation query translation,
  documenttranslation, interlingual techniques
• Translation knowledge ontology,
  bilingualdictionary, machine translation
  lexicons,corpora
• (Oard and Diekema,
  1998 )

14
Matching - Cognate matching

• Guess the meaning based on similarities in
  spelling or pronunciation (i.e. proper nouns and
  technological terms)
• - Buckley et al. (1998) letter sequences
  with similar sounds (I.e., c, k, qu)
• - Davis extended this technique using fuzzy
  matching to discover Spanish cognates for
  English.
• This method is mostly combined with other major
  translation methods.

15
Matching - Translation

• Query translation
• Document translation,
• Interlingual techniques

16
Matching - Translation

• Most system apply query translation first due to
  its efficiency compared to document translation
  in time and cost.
• Risk of query translation queries often do not
  contain enough context to permit good translation
  
• Some experimental systems tried to provide
  translation of summary for small amount of
  retrieved documents. (I.e. Keizai, MULINEX)

17
Matching - Translation

• Interlingual techniques
• Based on multilingual thesauri,
  controlled-vocabulary techniques applied to
  convert both the query and the documents into a
  unified language independent representation.
  (I.e. Lexical Semantic Indexing (LSI) Dumais al.
  et.,1997)

18
Matching - Translation Interlingua approach
(example)

• TextWises CINDOR (Conceptual Interlingua
  Approach
• Both documents and queries are mapped into the
  Conceptual Interlingua
• Permits matching and retrieval based on any
  combination of languages involved, rather than
  relying on pairwise translations
• Enables language-independent retrieval based on
  natural language concepts and matching to
  synonyms in all languages
• (Liddy and Sheridan,
  1999)
• Cindorsearch

19
TextWises CINDOR(Liddy and Sheridan, 1999)
20
Matching

• 3. Translation knowledge
• - Ontology,
• - Bilingual dictionary,
• - Machine translation lexicons,
• - Corpora

21
Matching 3. Translation knowledge Ontology

• The most dominant and effective source for CLIR
  (i.e. thesauri- support controlled-vocabulary
  and free-text retrieval, hierarchical
  relationships (broader, narrow terms), synonymy,
  and related term)
• (I.e. SPIDER system multilingual thesaurus
  and query expansion with Italian and German)
• (Sheridan
  Ballerini, 1996)
• It help users to define better queries but needs
  users understanding of those structure. If it is
  given easy explanations or examples of this, it
  can be a good aid tool.(user must create the
  query using only vocabulary from the thesaurus)

22
Matching 3. Translation knowledge Bilingual
dictionary

• It has been widely used. It is useful because it
  can enhance users query formulation.
• Problems ambiguity (selecting correct terms)
• a limited scope
• - Abdelali et al. (2002) attempt to present
  visual dictionaries for who does not have
  knowledge of the target language. It is a useful
  try of query-interactive approach

23
Translation knowledge Machine Translation
Lexicon

• To overcome the limitations of general purpose
  transfer dictionary, Salton (1971)
• used tuned lexicons and thesauri built from
  controlled vocabulary to good success in specific
  text retrieval problems.
• - Preparing special purpose of lexical
• resources remains a daunting task.
• (cited on Ogden et. al., 1999)

24
Translation knowledge parallel/comparable corpora

• Text corpora contains examples of usage patterns
  in the query language that can be matched to
  examples in the target language if the sentence
  or paragraphs of texts are aligned to one
  another.
• Braschler and Schäuble (2000) applied this
  approach, and obtained useful alignment sets
  among German, French, English, and Italian
  language documents. One aspect of the theory is
  that end users can be best served by efficient,
  inexpensive systems

25
Example of an alignment pair(Corpus-based
approaches)

• German and French documents, only titles are
  shown.
• Takeshita zu Antrittsbesuch in Bonn
  eingetroffen
• (Takashita has arrived in Bonn for first
  visit)
• Arrivee en RFA du premier ministre japonais
• (Arrival of the Japanese prime minister in
  the FRG)
• Alignments are produced by indicators to
  find similarities between pairs of documents.
• 
  (Brashler Schable, 2000)

26
Translation knowledge parallel/comparable corpora

• Generalized Vector Space Model
• use a bilingual training corpus to build
  matrices of document and term weights
• (Carbonell et al. 1997)
• Latent Semantic Indexing
• - Telcordia Latent Semantic Indexing (LSI)
  Demos
• - LSI by Dumais, Landauer Littman (SIGIR96)

27
Attempted experiment system

• ARCTOS An interactive search engine that
  illustrates a selection interface that shows how
  thumbnail images can be used to support selection
  without knowing the document's language. ARCTOS
  was developed by the Computing Research Lab at
  New Mexico State University.
• MULINEX(Multilingual Indexing, Navigation and
  Editing Extensions for the WWW) multilingual
  search engine for the WWW. It has collected
  around 5000 French, English and German web pages
  related to the European Currency Union
  (http//mulinex.dfki.de/demo.html)

28
Attempted experiment system

• MUNDIAL
• MINDS(Multilingual Interactive Document
  Summarization) Summarization in Spanish,
  Japanese, Korean, and English
• CANAL (Cataloging with Multilingual Natural
  Language Access/Linguistic Server)
• KEIZAI
• Unicode used. For who dont know the target
  language (Japanese, Korean, and English)

29
User Studies

• Keizai (1999) They focused on trying to
  understand how to create interfaces and systems
  that are useful to people. empirical user
  studies evaluating retrieval visualization
  interfaces.
• Keizai is a cross-language interactive retrieval
  that uses URSA (Unicode Retrieval System
  Architecture), developed at the Computing
  Research Laboratory at New Mexico State
  University.

30
System Evaluation

• TREC(Text retrieval conferences) US
• TREC-3 Spanish / TREC-4 Chinese/
• TREC-6 SMART - balance query
• English and French
• TREC-7 English, French, and German, and
• Italian)
• CLEF(Cross-language evaluation forum) in Europe
• NCCTR (National Institute of Informatics-NACSIS
  Test Collection for IR system) in Japan

31
System Evaluation

• Arctos project User involvement. The user was
  asked to interactively improve the query
  translation using bilingual translation
  resources. The retrieved documents were presented
  using document thumbnails and query highlighting
• TITAN Kikui et al, 1996, Japanese-English
• Multilingual Gisting Resnik, 1997

32
User Interactive CLIR

• Ogden et al. (1999) implemented more interactive
  approaches the system, Arctos, provides a user
  with a browser-based interface with which to
  enter English queries. After an initial query is
  entered, the query is translated using a simple
  word-for-word or phrasal translator. The user
  can then interactively improve the query
  translation using links to on-line bilingual
  translation resources and then submit the query
  for retrieval against document collections in the
  target.

33
User Interactive CLIR

• User-aided query translation
• user can disambiguate query term
• (I.e. Query expansion)
• Query formulation
• Davis and Ogden (1997) who implemented the
  QUILT system by providing a display of the
  Spanish translation of English query terms.
  Judging relevance with showing summary of
  English translation
• (Cited in Oard and Dikema, 1998, p.229)

34
Conclusion

• Needed for more CLIR user studies(users
  information seeking behavior and users particular
  domain characteristics) in order to reflect their
  real needs to provide updated information.
• CLIR evaluations are needed more attention of
  users interaction with the system as well as
  system performance.
• Needed more efficient interface in order to
  reflect users needs.

35
References

• Braschler, M. Schäuble, P. (2000). Using
  Corpus-Based Approaches in a System for
  Multilingual Information Retrieval. Information
  Retrieval, 3, 273-284.
• Grefenstette, G.. (Ed.). (1998). Cross-Language
  Information Retrieval. Norwell, MA Kluwer
• Lee, K., Kageura, K. Choi, K. (2004). Implicit
  ambiguity resolution using incremental clustering
  in cross-language information retrieval.
  Information Processing and Management, 40,
  145-159.
• Oard, D., Diekema, A. (1998). Cross-Language
  Information Retrieval. Annual Review of
  Information Science and Technology (ARIST), 33,
  223-256.

36
References

• Ogden, D., Cowie, J., Davis, M.,Ludovik, E.,
  Molina-Salado, H. Shin, H. (1999). Getting
  Information from Documents You Cannot Read An
  Interactive Cross-Language Tex Retrieval and
  Summarization System. Joint ACM Digital
  Library/SIGIR Workshop on Multilingual
  Information Discovery and AccesS (MIDAS).
• Peters, C. (Ed.) (2000). Cross-Language
  Information Retrieval and Evaluation Workshop of
  th Cross-Language Evaluation Forum, CLEF 2000,
  Lisbon, Portugal, September 2000. Berlin
  Springer-Verlag.
• Sheridan, P. Ballerini, J. (1996). Experiments
  in multilingual information retrieval using the
  SPIDER system. ACM, 58-65.