NLIWoD ISWC 2014 - Multilingual Retrieval Interface for Structured data on the Web
0 likes769 views
The document discusses a multilingual retrieval interface for structured data on the web, utilizing natural language queries to access information about items like the Mona Lisa. It presents a two-layered architecture supporting multiple languages and features a grammar framework for query generation, aiming for interoperability with ontologies. The system allows for efficient and effective retrieval across 15 languages, highlighting its scalability and potential for extension.
NLIWoD ISWC 2014 - Multilingual Retrieval Interface for Structured data on the Web
- 1. Multilingual Retrieval Interface for Structured Data on the Web Dana Dannells, Ramona Enache, Mariana Damova NLIWoD - ISWC’2014
- 2. Semantic Data Infrastructures Leonardo Mona Lisa RDF Repository SPARQL :Painter :Painting :painted Mona Lisa ? rdf:type rdf:type • Semantic Web • Linked data • SPARQL query language 2 10/19/2014
- 3. Natural Interface through NL 10/19/2014 3 EN FR DE Who painted Mona Lisa? Qui a paint Mona Lisa? Wer hat Mona Lisa gemahlt? Who is Mona Lisa’s painter? Qui est le paintre de Mona Lisa? Wer ist der Mahler von Mona Lisa ? Who created Mona Lisa? Qui a créé Mona Lisa? Wer hat Mona Lisa geschöpft? Leonardo da Vinci A: Q: NL to ontology interoperability
- 4. Multilingual Retrieval Interface 10/19/2014 4
- 5. GF – Grammar Framework •Type-theoretical grammar formalism supporting multilingual applications •Two-layered architecture –Abstract syntax - semantics –Concrete syntax – language dependent surface structure 10/19/2014 5 Abstract syntax: Concrete English syntax: Abstract representation: cat NP, VP, S; lincat NP, VP, S = {s: Str}; lin Mary = mkNP (mkPN "Mary"); fun Mary, John: NP; lin Mary = {s = "Mary"}; lin John = mkNP (mkPN "John"); fun Love: NP -> VP; lin John = {s = "John"}; lin Love o = mkVP (mkV2 "love") o; fun Pred : NP -> VP -> S; lin Love o = {s = "loves" ++ o.s}; lin Pred sub v = mkS (mkCl sub v); lin Pred sub v = {s = sub.s ++ v.s}; Ex: John loves Mary
- 6. Multilingual Aspect of GF 10/19/2014 6
- 7. YAQL (Yet Another Query language) •A common architecture with one base module and domain knowledge representation •Straightforward abstract syntax generation from ontology with just the minimum lexical types –Common noun – Kind –Noun phrase – Entity –Verb phrase – Property –Verb phrase with higher arity – Relation •Reusable generic grammar structure 10/19/2014 7 Abstract syntax: Concrete syntax: Move ; Move = Utt ; Query ; Query = QS ; MQuery : Query -> Move ; Query Command Answer
- 8. YAQL and the Semantic Web The category Kind gets coupled with OWL entities 10/19/2014 8 •Yet another NL two layers so that a new domain model can be easily integrated into the query module •Bidirectional translation in 15 languages (Bulgarian, Catalan, Danish, Dutch, English, Finnish, French, Hebrew, Italian, German, Norwegian, Romanian, Russian, Spanish, Swedish) Text Query Answer Data Lexicon RGL YAQL
- 9. NL to SPARQL 10/19/2014 9
- 10. Who painted Mona Lisa? English: Who painted t ? QPainter t = mkQS pastTense (mkQCl who_IP paint_V2 t) Finnish: Whose painting is t ?} QPainter t = mkQS (mkQCl (mkIP (E.GenIP who_IP) (mkN "maalaama")) t) French: By who is t ? QPainter t = mkQS (mkQCl (mkIAdv by8agent_Prep who_IP) t) 10/19/2014 10 $MQuery (QPainter (PTitle TMona_Lisa))$ Abstract syntax SPARQL MQuery q = "PREFIX painting:<http://spraakbanken.gu.se/rdf/owl/painting.owl#> PREFIX rdf:<http://www.w3.org/1999/02/22-rdf-syntax-ns#> PREFIX rdfs:<http://www.w3.org/2000/01/rdf-schema#>> SELECT distinct"++ q.wh1 ++ " WHERE { ?painting rdf:type painting:Painting; rdfs:label ?title; " ++ q.wh2 ++ q.prop++"}" ; Concrete syntax
- 11. Evaluation •User satisfaction •Efficiency in terms of time, effort and cost •Effectiveness, how the system scales up 10/19/2014 11 Coverage: 1159 query patterns in 15 languages 10 characteristics of CH objects Extendibility New query grammar - 150 lines of code Evaluation Random queries in 7 languages with very few native informants’ corrections
- 12. Conclusion •NL to ontology interoperability approach •Multilingual interface for retrieval of structured data from the Web •Easily extendable initial base of YAQL transformations •Great coverage of paraphrases •Expert language/information engineers required 10/19/2014 12
- 13. Thank you for your attention 10/19/2014 13 Contacts: dana.dannells@svenska.gu.se ra.monique@gmail.com mariana.damova@mozajka.co ?