Cross language semantic relations between English and Portuguese
•Download as PPT, PDF•
1 like•132 views
The presentation describes a first attempt to extract semantic relations (SR), namely cross-language SR between English and Portuguese.
More Related Content
Similar to Cross language semantic relations between English and Portuguese
Similar to Cross language semantic relations between English and Portuguese (17)
More from INESC-ID (Spoken Language Systems Laboratory - L2F)
More from INESC-ID (Spoken Language Systems Laboratory - L2F) (20)
Recently uploaded
Recently uploaded (20)
Cross language semantic relations between English and Portuguese
- 1. CROSS-LANGUAGE SEMANTIC RELATIONS BETWEEN ENGLISH AND PORTUGUESE ICL: Workshop on Iberian Cross-Language NLP Tasks 07 September, 2011 Anabela Barreiro Huelva, Spain Hugo Gonçalo Oliveira hroliv@dei.uc.pt
- 2. OUTLINE INTRODUCTION MOST STUDIED SEMANTIC RELATIONS & AUTOMATIC ACQUISITION OF MULTILINGUAL LEXICO-SEMANTIC RELATIONS LINGUISTIC-BASED METHOD TO GENERATE SEMANTIC RELATIONS LINGUISTIC RESOURCES EMPLOYED ESTABLISHMENT OF MORPHO-SEMANTICO-SYNTACTIC RELATIONS IN THE DICTIONARY AND CREATION OF GRAMMARS TO READ DICTIONARY INFORMATION AND GENERATE NEW SEMANTIC PAIRS CROSS-LANGUAGE RELATIONS RE-USING GRAMMARS PRELIMINARY RESULTS CONCLUSIONS AND FUTURE RESEARCH ICL: Workshop on Iberian Cross-Language NLP Tasks 07 September, 2011 Anabela Barreiro Huelva, Spain
- 3. SEMANTIC RELATIONS Synonymy - different lexical items have the same meaning e.g. car synonym-of automobile Homonymy - lexical items have the same orthographic form but different meanings e.g. bank, financial institution vs. slope Hyponymy - a lexical item is a subclass or a specific kind of another e.g. dog hyponym-of mammal Meronymy - a lexical item is a part, piece or member of another e.g. wheel part-of car. General and domain-oriented: process-of, result-of, etc. ICL: Workshop on Iberian Cross-Language NLP Tasks 07 September, 2011 Anabela Barreiro Huelva, Spain
- 4. MULTILINGUAL LEXICAL-SEMANTIC KNOWLEDGE DATABASES CREATED FROM THE WEB UWN/MENTA (de Melo & Weikum 2009, 2010): multilingual knowledge base with WordNet meaning connections for 1,500,000 words in over 200 languages (http://www.mpi-inf.mpg.de/yago-naga/uwn/query.html) PanDictionary (Mausam et al 2010): sense-distinguished, massively multilingual dictionary with translations in more than 1000 languages, created after exploiting free dictionaries, and inferring additional translations with a certain probability WikiNet (Nastase et al 2010): multilingual network of concepts obtained after exploiting Wikipedia, with lexicalizations, in several languages, as well as connections representing a variety of relations. Bergsma & Van Durme 2011: learning word-to-word translations after exploiting the visual similarity of labeled images on the Web ICL: Workshop on Iberian Cross-Language NLP Tasks 07 September, 2011 Anabela Barreiro Huelva, Spain
- 5. LINGUISTIC RESOURCES Eng4NooJ and Port4NooJ – linguistic knowledgeEng4NooJ and Port4NooJ – linguistic knowledge • OpenLogos dictionary (http://logos-os.dfki.de/) converted into NooJ format, and enhanced with new properties, including derivational and morpho-syntactic and semantic relations • Morphological systems • Contextual rules and grammars • Domain specific dictionaries ICL: Workshop on Iberian Cross-Language NLP Tasks 07 September, 2011 Anabela Barreiro Huelva, Spain LINGUISTIC-BASED METHOD TO GENERATE (CROSS-LANGUAGE) SEMANTIC RELATIONS Mapping morpho-syntactic and semantically related words
- 6. CONCEPTUAL SEMANTIC RELATIONS Semantico-syntactic Abstract Language (SAL) hierarchical taxonomic scheme over 1,000 elements or words (expandable) organized in Supersets, Sets, and Subsets distributed over all parts-of-speech Noun Supersets ICL: Workshop on Iberian Cross-Language NLP Tasks 07 September, 2011 Anabela Barreiro Huelva, Spain Concrete (CO) Mass (MA) Animate (AN) Place (PL) Information, (IN) Abstract (AB) Process (intransitive) (PI) Process (transitive) (PT) Measure (ME) Time (TI) Aspective (AS) Unknown (UN)
- 7. ICL: Workshop on Iberian Cross-Language NLP Tasks 07 September, 2011 Anabela Barreiro Huelva, Spain Abstract cancer circumstance coma loneliness poverty status fatherhood inequality category class kind make nature rank type reserve resource well-spring justice analogy idea truth Non –verbal abstracts [about persons, things] ABnonvb 40 general concepts ABgen 42 ABnonvb 40 ABprop 609 ABstate 736 ABclass 731 ABorig 723 undifferentiated non-verbals abs ABverb 41 verbal abstracts [about agents, processes] ABpur 748 ABmeth 733 ABqual 655 ABtime 732 ABcause 602 ABverb 41 ABnegc 764 ABcont 765 ABstrvb 749 [of v’g; to v] task problem objective function sources/origins states/conditions /relationships classifications barroque color design feature form likeableness profile trait shape properties/qual ities/nature acting technique computing task reading readiness navigation skill [for/of v’g] technique means mode pattern [of/in/for v’g] efficiency alacrity suitableness ease instance end holiday parade birthday [to v ; of/for v’g] basis motivation sense readiness talent ability cost [of v’g] danger threat risk catastrophe disaster accident blizzard navigation error noise abatement heat absorption purpose method/ procedure quality of action/agent time event cause/potential/ disposition undifferentiated verbal abstracts strong verbals process/result contrary event negative cause
- 8. ICL: Workshop on Iberian Cross-Language NLP Tasks 07 September, 2011 Anabela Barreiro Huelva, Spain DICTIONARY WITH IMPLICIT SEMANTIC RELATIONS impressionar,V+FLX=FALAR+SAL=PVPCpleasetype+EN=impress+VSUP=fazer+VSUP=causar +DRV=NDRV01:CANÇÃO adaptar,V+FLX=FALAR+Aux=1+SAL=INOP57+Subset=132+EN=adapt+VSUP=fazer +DRV=NDRV00:CANÇÃO azedar,V+FLX=LIMPAR+Aux=1+SAL=OBJTRundif98+Subset=740+EN=sour+VSUP=ficar +DRV=ADRV00:ALTO aesthetic,AFLX=NATURAL+SAL=AVstate+PT=estético+DRV=AVDRV03 skepticism,N+FLX=BOOK+SAL=ABcause+PT=cepticismo+DRV=NAVDRV02
- 9. ICL: Workshop on Iberian Cross-Language NLP Tasks 07 September, 2011 Anabela Barreiro Huelva, Spain Eng4NooJ Port4NooJ NDRV04 = <B>ion/Npred e.g. accelerate > acceleration NDRV02 = <B>nça/N+Npred e.g. mudar > mudança ADRV02 = <B>icable/ADJ e.g. apply > applicable ADRV02 = <B2>o/A+Apred e.g. azedar > azedo AVDRV01 = <E>ly/ADV e.g. frequent > frequently AVDRV00 = <B>zmente/ADV e.g. veloz > velozmente AVDRV04 = <B>tically/ADV e.g. realism > realistically AVDRV05 = <A> <B>amente/ADV e.g. rápido > rapidamente RULES TO TRANSFORM MORPHO-SYNTACTIC AND SEMANTICALLY RELATED WORDS OF DIFFERENT PoS TRANSFORMATIONAL RULES
- 10. ICL: Workshop on Iberian Cross-Language NLP Tasks 07 September, 2011 Anabela Barreiro Huelva, Spain ACTION-OF AND RESULT-OF SEMANTIC RELATIONS ACTION-OF abolishment IS ACTION OF abolish abolição É AÇÃO DE abolir abuse IS ACTION OF abuse abuso É AÇÃO DE abusar happening IS ACTION OF happen acontecimento É AÇÃO DE acontecer agreement IS ACTION OF agree acordo É AÇÃO DE acordar RESULT-OF lit IS RESULT OF light aceso É RESULTADO DE acender stuffed IS RESULT OF stuff embalsamado É RESULTADO DE embalsamar rotten IS RESULT OF rotten podre É RESULTADO DE apodrecer interdicted IS RESULT OF interdict interditado É RESULTADO DE interditar
- 11. Grammar to recognize adverbial compounds and transform them into equivalent single adverbs LOCAL GRAMMARS ICL: Workshop on Iberian Cross-Language NLP Tasks 07 September, 2011 Anabela Barreiro Huelva, Spain Grammar to generate cross- language relations between Portuguese support verb constructions and equivalent English single verbs Mechanism created to read the linguistic information in the dictionaries and generate the semantic relations
- 12. SPIDER (SYSTEM OF PARAPHRASING IN DOCUMENT EDITING AND REVISION)Suggestions for general language linguistic phenomena Compound adverbs > single adverbs Support verb constructions > single verbs Relatives > participial adjectives ICL: Workshop on Iberian Cross-Language NLP Tasks 07 September, 2011 Anabela Barreiro Huelva, Spain Barreiro, 2011
- 13. SEMANTIC EQUIVALENCE IN SPIDER Synonyms in context (ex: phrasal verbs equivalent expressions)↔ to clear up (weather) = (weather) to become better/brighter Support verb constructions single verbs stylistic variants↔ ↔ to make a decision = to decide; to make an audit = to perform an audit Aspectual constructions single verbs↔ to launch an attack = to attack Adverbials (compounds single adverbs)↔ in a constructive way = constructively; on purpose > purposely = deliberately Relatives participial adjectives↔ the president that was elected = the president elect Relatives possessives↔ the role that Europe plays/has = the role of Europe RelativesRelatives ↔ compound nouns (and vice-versa)compound nouns (and vice-versa) a container for the milk = a milk container; a bottle made of plastic = a plastic bottlea container for the milk = a milk container; a bottle made of plastic = a plastic bottle Agentive passives ↔ actives the man was released by the police officer = the police officer released the man CICLing 2011 February 20-26, 2011 Anabela Barreiro Tokyo, Japan
- 14. $EN CROSS-LANGUAGE RELATIONS RE- USING GRAMMARS a fazer um estágio para dar aulas de / tutor Religião a fazer um estágio para dar aulas de / lecture Religião a fazer um estágio para dar aulas de / teach Religião começa a dar exemplos / exemplify : sentia-se capaz de dar um murro em / punch quem quisesse detê-lo gostávamos de lhe dar uma palavrinha / speak . ICL: Workshop on Iberian Cross-Language NLP Tasks 07 September, 2011 Anabela Barreiro Huelva, Spain
- 15. PRELIMINARY RESULTS ICL: Workshop on Iberian Cross-Language NLP Tasks 07 September, 2011 Anabela Barreiro Huelva, Spain Relation Quantity Hyponymy 14963 Synonymy (between verbs, nouns, adj and adv) 10395 (5367, 20, 34, 5014) Action of 3773 Result of 283 Port4NooJ (publicly available at http://www.linguateca.pt/Repositorio/Port4Nooj/): 600 derivational rules most transforming verbs into predicate nouns (587) 119 productive (nominalizations) 486 verb relations between verbs and autonomous predicate nouns
- 16. CONCLUSIONS AND FUTURE RESEARCH ICL: Workshop on Iberian Cross-Language NLP Tasks 07 September, 2011 Anabela Barreiro Huelva, Spain The linguistic-based method is systematic, expandable (unlimited possibility to grow and improve) Dictionaries and grammars to generate monolingual semantic relations are easily adaptable and can be reused to generate cross-language relations, rules can be standardized and often re-used across close languages. Even though the methodology adopted was applied to the OpenLogos resources, it is compliant with the exploitation of other lexical resources with semantic relations, for any language besides English and Portuguese
- 17. CONCLUSIONS AND FUTURE RESEARCH ICL: Workshop on Iberian Cross-Language NLP Tasks 07 September, 2011 Anabela Barreiro Huelva, Spain Semantic relations - highly suitable for paraphrasing and cross- language tasks, including machine translation Need for robust semantic relations quality resources and semantic- driven NLP applications Future work - gather and combine open source available semantic resources, enhance properties on the existing resources, and enlarge the linguistic phenomena coverage.
- 18. CROSS-LANGUAGE SEMANTIC RELATIONS BETWEEN ENGLISH AND PORTUGUESE ICL: Workshop on Iberian Cross-Language NLP Tasks 07 September, 2011 Anabela Barreiro Huelva, Spain Hugo Gonçalo Oliveira hroliv@dei.uc.pt Acknowledgements Anabela Barreiro was partially supported by the UPV, award 1931, under the program Research Visits for Renowned Scientists (PAID-02-11). Hugo Gonçalo Oliveira is supported by the FCT scholarship grant SFRH/BD/44955/2008, co-funded by FSE.
Editor's Notes
- Good afternoon! My name is Anabela Barreiro from L2F INESC-Lisbon. Today, I will present some work in collaboration with my colleague Hugo Gonçalo Oliveira from the Centre for Informatics and Systems of the University of Coimbra. The presentation describes a first attempt to extract semantic relations (SR), namely cross-language SR between English and Portuguese.
- This short presentation is divided in 3 main parts: a short introduction, a content part and the conclusions. describes the state of the art in automatic acquisition of distinct types of lexico-semantic relations. presents the base linguistic resources used to attain semantic relations. describes the relations of synonymy, hyponymy, action-of, and result-of. presents the method for the extraction of the semantic relations. It describes, in particular, the morpho-syntactic and semantic relations established in the dictionary, how the grammars read this linguistic information, and how they use it to generate semantic pairs. This latter section also shows how to expand from monolingual to cross-language relations with minimal change in the local grammars. Section 6 presents some preliminary results. And finally, section 7 presents the conclusions and guidelines for future research work. ------------------------------------------------------------------------- In the introduction, I will describe paraphrases in general and emphasize their importance in language and communication. In the second part of this presentation, I will discuss paraphrasing functions, make a few comments on the importance of paraphrases in NLP tasks and the practical need for paraphrases in the industrial context, for professional purposes, including those of pedagogical nature. I will continue with some considerations on the importance of paraphrases for translation, and tools that use paraphrases. The core of the presentation is a new tool, called SPIDER, which is a system of paraphrasing in document editing and revision. Finally, I will present the conclusions. And I will finalize my presentation with a few notes of future applications and research.
- Some of the most studied lexico-semantic relations are: But in certain domains, the semantic relations: process-of, result-of, among others are very common.
- Interesting work is being developed on cross-language/multilingual lexical-semantic knowledge databases created from the web.
- Some of these SR were extracted from the lexical resources of the OpenLogos machine translation system. In combination with these resources, new resources were created, namely derivational rules and grammars to recognize and generate morpho-syntactic and semantically related words and multiword units. Semantic relations, obtained by means of local grammars developed within NooJ linguistic environment, cover larger number of items and can be extracted in a simple and easy way. This paper aims at showing how these resources combined can be used in cross-language tasks. This linguistic-based method of automatic generation of (cross-language) semantic relations consists of mapping morpho-syntactic and semantically related words The semantic relations were generated from 2 new linguistic knowledge resources: Eng4NooJ and Port4NooJ. These resources were built with the NooJ linguistic environment. Part of the linguistic knowledge on these resources came from the OpenLogos machine translation system. The adapted and enhanced linguistic knowledge consists of dictionary and local grammars, which include new properties and transformations allowing for the establishment of morpho-syntactic and semantic relations. Local grammars apply the new dictionary properties, permitting the transformations required in paraphrasing and MT.
- The conceptual semantic relations were symbolically represented in the OpenLogos lexicon as a hierarchical taxonomic scheme called semantico-syntactic abstraction language (SAL), with over 1,000 elements or words (expandable), organized in Supersets, Sets, and Subsets, distributed over all parts-of-speech. For example, SAL has 12 supersets for nouns: Semantic relations - generated automatically, based on the linguistic information associated with each lexical entry. This info was used to generate hierarchical hyponymy and hypernymy relations, among others.
- This slide shows the Abstract noun Superset. In the Abstract noun Superset, there are two principal Sets: the non-verbal Abstract nouns, and the verbal Abstract nouns, both with their own Subsets. The Subset Classifications is a member of the non-verbal Abstract noun Set. It includes nouns such as category, class, kind, make, nature, rank, type, among others. The Subset Methods/ Procedures is a member of the verbal Abstract noun Set. It includes nouns such as technique, means, mode, pattern, among others. The complete taxonomy can be viewed at the Logos Archives website http://logossystemarchives.homestead.com/.
- EXTRACTION OF CONCEPTUAL SEMANTIC RELATIONS FROM THE RESOURCES - Action-of, result-of, and synonymy relations between multiword units and single words, where there is a morpho-syntactic and semantic relation between words of distinct PoS.
- Local grammars - mechanism created to read the linguistic information in the dictionaries and generate the semantic relations. These SR have been used in paraphrasing and machine translation.
- SPIDER uses linguistically based automated paraphrasing and text-editing mechanisms to help users with their writing needs by providing suggestions for customized text authoring, so it can be used word processing applications. It also generates word and phrasal usage data to help guide decision-making. This slide illustrates SPIDER’s suggestions for several general language linguistic phenomena: Compound adverbs &gt; single adverbs Relatives &gt; participial adjectives SVCs &gt; single verbs
- SPIDER’s linguistic knowledge allows extensive transformation and re-writing, including recognition and rewriting of: words or multi-word units into their synonyms or paraphrases, in the appropriate contexts (to clear up (weather) &gt; to become better/brighter); support verb constructions into single verbs (to make a decision &gt; to decide; to give support to N(AN) &gt; to support N(AN); to go V-ing &gt; to continue V-ing; to get into contact with &gt; to contact; to turn on N(light) &gt; to extinguish N; to become acid &gt; to acidify); support verb constructions into their stylistic variants (to make an audit &gt; to perform an audit; to make an impression &gt; to cause an impression); aspectual constructions into verbs, (to launch an attack &gt; to attack); Multiword adverbs and adverbial phrases into single adverbs (in a constructive way &gt; constructively; on purpose &gt; purposely / deliberately); Relatives into possessives (the position that the Church defends &gt; the position of the Church; the role that the politicians play &gt; the role of the politicians); Relatives into participial adjectives (the president that was elected &gt; the president elected); Relatives into compound nouns (a container for the milk &gt; a milk container); phrases with “made of” (a bottle made of plastic &gt; a plastic bottle); Agentive passives into actives (the young man is released by the police officer &gt; the police officer releases the young man).
- ReWriter was inspired in ParaMT, a prototype of a multilingual paraphraser (or translation system). ParaMT uses a similar methodology, except for that it provides an equivalent in a language different from the one of the source text (paraphrases across language). ParaMT can be used directly in machine translation. At the current stage of development, ParaMT aims at translating multi-word units efficiently, handling considerably well the translation of Portuguese support verb constructions into English verbs, as illustrated in the table on this slide. Because Eng4NooJ resources contain Portuguese transfers for each entry, any grammar used to obtain monolingual transformations can be reused to obtain bilingual (or multilingual) transformations. The recycling of grammars is minimal, since the only parameter that needs to be added is the specification of the output language, as $EN for English or $PT for Portuguese (meaning, “retrieve the output in English”, etc.). For monolingual transformations, no output language is specified.
- In theory, the exploitation of the lexicon in combination with SAL allows the establishment of numerous relations between words and expressions. For the current paper, we focused only on a few of those relations which cover a larger number of items and could be extracted in a simple and easy way. The result of extraction for Portuguese (not yet reviewed) is publicly available. Currently, Port4NooJ contains more than 30,000 morpho-syntactic relations between semantically related elements. Table 6 presents some preliminary results, which do not refer to paraphrasing capabilities, but simply to relations between lexical items. The total results for paraphrasing are significantly higher. Local grammars, applied to information (properties) described in the dictionary, enable the recognition and analysis of expressions such as de (um) modo rápido, de (uma) forma/maneira rápida (in a fast/quick way) (which could be considered as relations between an adjective and an adverb, but which were not considered), and also inflected forms such as dar uns passeios (go for some walks), etc. http://www.linguateca.pt/Repositorio/ Port4NooJ/relacoes_semanticas_explicitas/. Port4NooJ contains approximately 600 derivational rules, most of them transforming verbs into predicate nouns (587). 119 of these rules are productive, covering nominalizations. 486 rules correspond to verb relations between verbs and autonomous predicate nouns. Rules were only superficially evaluated.
- This paper presented semantic relations, namely domain-independent semantico-syntactic and ontological relations, suitable for paraphrasing and cross-language tasks, including machine translation. We have demonstrated that given the appropriate linguistic resources, the generation of semantic relations can become very systematic. Any grammar to generate monolingual semantic relations can be reused to generate cross-language relations, rules can be standardizes and often re-used across close languages, etc. Even though the methodology adopted was applied to the OpenLogos resources, it is compliant with the exploitation of other lexical resources with semantic relations, for any language besides English and Portuguese, studied in this research. Future work would gather and combine open source available semantic resources, enhance properties on the existing resources, and enlarge the linguistic phenomena coverage.
- Semantic relations, namely domain-independent semantico-syntactic and ontological relations, are highly suitable for paraphrasing and cross-language tasks, including machine translation. We have demonstrated that given the appropriate linguistic resources, the generation of semantic relations can become very systematic. Any grammar to generate monolingual semantic relations can be reused to generate cross-language relations, rules can be standardizes and often re-used across close languages, etc. Even though the methodology adopted was applied to the OpenLogos resources, it is compliant with the exploitation of other lexical resources with semantic relations, for any language besides English and Portuguese, studied in this research. Future work would gather and combine open source available semantic resources, enhance properties on the existing resources, and enlarge the linguistic phenomena coverage.