Language Technology and the Semantic Web Thierry Declerck & Paul Buitelaar (Saarland University & DFKI Gmb. H) T. Declerck, P. Buitelaar
We present collaborative research work on the combination of language technology (LT) and technologies for encoding (domain) knowledge in ontologies, supporting the emergence of the Semantic Web (SW), or maybe more appropriate: Semantic Webs MUMIS (dealing with multimedia content indexing and searching in the soccer domain, finished in December 2002) Much. More (dealing with cross-lingual information retrieval in the medical domain, finished in Mai 2003) Esperonto (developing a Semantic Annotation Service for upgrading the actual Web to the Semantic Web, Sept. 2002 - Mai 2005) T. Declerck, P. Buitelaar 2
Semantic Web Applications of LT Supporting accurate ontology-based semantic annotation of multilingual web documents (Knowledge Markup) Supporting Ontology Learning/Construction from linguistically/semantically annotated multilingual text (Knowledge Extraction) See also the Special Interest Group (SIG-5) Onto. Web-lt on Language Technology in Ontology Development and Use: http: //ontoweb-lt. dfki. de T. Declerck, P. Buitelaar 3
Knowledge Markup and Knowledge Extraction Text/Speech Mining Concepts, Relations, Events Linguistic and Semantic Annotations Linguistic Analysis Morpho-Syntactic Analysis and Tagging, Semantic Class Tagging, Term/NE Recognition, Grammatical Function Tagging, Dependency Structure Analysis T. Declerck, P. Buitelaar 4
Knowledge Markup and Knowledge Extraction (2) Text/Speech/Image-Video Text/Speech/Media Mining Concepts, Relations, Events Linguistic, Low-level Image and Semantic Annotations Linguistic and Media Analysis T. Declerck, P. Buitelaar 5
Integration of Language Technology and Domain Knowledge T. Declerck, P. Buitelaar 6
Linguistic Analysis Language technology tools are needed to support the upgrade of the actual web to the Semantic Web (SW) by providing an automatic analysis of the linguistic structure of textual documents. Free text documents undergoing linguistic analysis become available as semi-structured documents, from which meaningful units can be extracted automatically (information extraction) and organized through clustering or classification (text mining). Here we focus on the following linguistic analysis steps that underlie the extraction tasks: morphological analysis, part-of-speech tagging, chunking, dependency structure analysis, semantic tagging. T. Declerck, P. Buitelaar 7
Morphological Analysis Morphological analysis is concerned with the inflectional, derivational, and compounding processes in word formation in order to determine properties such as stem and inflectional information. Together with part-of-speech (Po. S) information this process delivers the morpho-syntactic properties of a word. While processing the German word Häusern (houses) the following morphological information should be analysed: [Po. S=N NUM=PL CASE=DAT GEN=NEUT STEM=HAUS] T. Declerck, P. Buitelaar 8
Part-of-Speech Tagging Part-of-Speech (Po. S) tagging is the process of determining the correct syntactic class (a part-of-speech, e. g. noun, verb, etc. ) for a particular word given its current context. The word “works” in the following sentences will be either a verb or a noun: He works [N, V] the whole day for nothing. His works [N, V] have all been sold abroad. Po. S tagging involves disambiguation between multiple part-ofspeech tags, next to guessing of the correct part-of-speech tag for unknown words on the basis of context information. T. Declerck, P. Buitelaar 9
Chunking Following Abney: chunks as the non-recursive parts of core phrases, such as nominal, prepositional, adjectival and adverbial phrases and verb groups. Chunk parsing is an important step towards making natural language processing robust, since the goal of chunk parsing is not to deliver a full analysis of sentences, but to extract just the linguistic fragments that can be surely identified. However, even if this strategy fails to produce an analysis for the whole sentence, the partial linguistic information gained so far will still be useful for many applications, such as information extraction and text mining. T. Declerck, P. Buitelaar 10
Named Entities detection Related to chunking is the recognition of so-called named entities (names of institutions and companies, date expressions, etc. ). The extraction of named entities is mostly based on a strategy that combines look up in gazetteers (lists of companies, cities, etc. ) with the definition of regular expression patterns. Named entity recognition can be included as part of the linguistic chunking procedure and the following sentence fragment: “…the secretary-general of the United Nations, Kofi Annan, …” will be annotated as a nominal phrase, including two named entities: United Nations with named entity class: organization, and Kofi Annan with named entity class: person T. Declerck, P. Buitelaar 11
Dependency Structure Analysis A dependency structure consists of two or more linguistic units that immediately dominate each other in a syntax tree. The detection of such structures is generally not provided by chunking but is building on the top of it. There are two main types of dependencies that are relevant for our purposes: On the one hand, the internal dependency structure of phrasal units or chunks and on the other hand the so-called grammatical functions (like subject and direct object). T. Declerck, P. Buitelaar 12
Internal Dependency Structure In linguistic analysis, for this we use the terms head, complements and modifiers, where the head is the dominating node in the syntax tree of a phrase (chunk), complements are necessary qualifiers thereof, and modifiers are optional qualifiers. Consider the following example: . “The shot by Christian Ziege goes over the goal. ” The prepositional phrase “by Christian Ziege” (containing the named entity Christian Ziege) depends on (and modifies) the head noun “shot”. T. Declerck, P. Buitelaar 13
Grammatical Functions Determine the role (function) of each of the linguistic chunks in the sentence and allow to identify the actors involved in certain events. So for example in the following sentence, the syntactic (and also the semantic) subject is the NP constituent “The shot by Christian Ziege”: “The shot by Christian Ziege goes over the goal. ” This nominal phrase depends on (and complements) the verb “goes”, whereas the Noun “shot” is the head of the NP (it this the shot going over the goal, and not Christian Ziege!) T. Declerck, P. Buitelaar 14
Semantic Tagging Automatic semantic annotation has developed within language technology in recent years in connection with more integrated tasks like information extraction, which require a certain level of semantic analysis. Semantic tagging consists in the annotation of each content word in a document with a semantic category. Semantic categories are assigned on the basis of a semantic resources like Word. Net for English or Euro. Word. Net, which links words between many European languages through a common inter-lingua of concepts. T. Declerck, P. Buitelaar 15
Semantic Resources Semantic resources are captured in dictionaries, thesauri, and semantic networks, all of which express, either implicitly or explicitly, an ontology of the world in general or of more specific domains, such as medicine. They can be roughly distinguished into the following three groups: Thesauri: Semantic resources that group together similar words or terms according to a standard set of relations, including broader term, narrower term, sibling, etc. (like Roget) Semantic Lexicons: Semantic resources that group together words (or more complex lexical items) according to lexical semantic relations like synonymy, hyponymy, meronymy, and antonymy (like Word. Net) Semantic Networks: Semantic resources that group together objects denoted by natural language expressions (terms) according to a set of relations that originate in the nature of the domain of application (like UMLS in the medical domain) T. Declerck, P. Buitelaar 16
The Me. SH Thesaurus Me. SH (Medical Subject Headings) is a thesaurus for indexing articles and books in the medical domain, which may then be used for searching Me. SH-indexed databases. Me. SH provides for each term a number of term variants that refer to the same concept. It currently includes a vocabulary of over 250, 000 terms. The following is a sample entry for the term gene library (MH is the term itself, ENTRY are term variants): MH ENTRY etc. T. Declerck, P. Buitelaar = = = Gene Library Bank, Gene Banks, Gene DNA Libraries Gene Bank 17
The Word. Net Semantic Lexicon Word. Net has primarily been designed as a computational account of the human capacity of linguistic categorization and covers an extensive set of semantic classes (called synsets). Synsets are collections of synonyms, grouping together lexical items according to meaning similarity. Synsets are actually not made up of lexical items, but rather of lexical meanings (i. e. senses) T. Declerck, P. Buitelaar 18
Word. Net: An example The word 'tree' has two meanings that roughly correspond to the classes of plants and that of diagrams, each with their own hierarchy of classes that are included in more general super-classes: 09396070 tree 0 09395329 woody_plant 0 ligneous_plant 0 09378438 vascular_plant 0 tracheophyte 0 00008864 plant 0 flora 0 plant_life 0 00002086 life_form 0 organism 0 being 0 living_thing 0 00001740 entity 0 something 0 10025462 tree 0 tree_diagram 0 09987563 plane_figure 0 two-dimensional_figure 0 09987377 figure 0 00015185 shape 0 form 0 00018604 attribute 0 00013018 abstraction 0 T. Declerck, P. Buitelaar 19
Cy. C: A Semantic Network CYC is a semantic network of over 1, 000 manually defined rules that cover a large part of common sense knowledge about the world. For example, CYC knows that trees are usually outdoors, or that people who died stop buying things. Each concept in this semantic network is defined as a constant, which can represent a collection (e. g. the set of all people), an individual object (e. g. a particular person), a word (e. g. the English word person), a quantifier (e. g. there exist), or a relation (e. g. a predicate, function, slot, attribute). The entry for the predicate #$mother: #$mother : (#$mother ANIM FEM) isa: #$Family. Relation. Slot #$Binary. Predicate This says that the predicate #$mother takes two arguments, the first of which must be an element of the collection #$Animal, and the second of which must be an element of the collection #$Female. Animal. T. Declerck, P. Buitelaar 20
Word Sense Disambiguation Words mostly have more than one interpretation, or sense. If natural language were completely unambiguous, there would be a one-to-one relationship between words and senses. In fact, things are much more complicated, because for most words not even a fixed number of senses can be given. Therefore, only in certain circumstances and depending on what we mean exactly with sense, can we give restricted solutions to the problem of Word Sense Disambiguation (WSD) T. Declerck, P. Buitelaar 21
A simplified Example of a Domain Ontology_1: Movies Title: String Date: mm/dd/yyyy Duration: minutes Type: (action, drama, . . ) Director: String Main Actors: Name_1: Role: Name_2: Role: Name_3: Role: … … T. Declerck, P. Buitelaar Instances Ontology_1: Movies Title: Lord of the Rings Date: Duration: Type: Director: Peter Jackson Main Actors: Name_1: Role: Name_2: Role: Name_3: Role: … … 22
Example of RDF Schema for the Movie Ontology <rdf: RDF xmlns: rdf='http: //www. w 3. org/1999/02/22 -rdf-syntax-ns#' xmlns: rdfs='http: //www. w 3. org/2000/01/rdf-schema#' xmlns: NS 0='http: //webode. dia. fi. upm. es/RDFS/Movie. Ontology#' > <rdf: Description rdf: about='http: //webode. dia. fi. upm. es/RDFS/Movie. Ontology#Special. Effects. Company. Acting'> <rdf: type rdf: resource='http: //www. w 3. org/2000/01/rdf-schema#Class'/> <rdfs: comment>Details of company that created special effects in this movie</rdfs: comment> <rdfs: sub. Class. Of rdf: resource='http: //webode. dia. fi. upm. es/RDFS/Movie. Ontology#Company. Acting'/> </rdf: Description> <rdf: Description rdf: about='http: //webode. dia. fi. upm. es/RDFS/Movie. Ontology#Police'> <rdf: type rdf: resource='http: //www. w 3. org/2000/01/rdf-schema#Class'/> <rdfs: comment>Films that deal solely with police activity</rdfs: comment> <rdfs: sub. Class. Of rdf: resource='http: //webode. dia. fi. upm. es/RDFS/Movie. Ontology#Crime'/> </rdf: Description> etc… T. Declerck, P. Buitelaar 23
Integration of Ontology and Semantic Lexicon Example of Semantic Lexicon is Word. Net (sometimes also referred to as a ‘Linguistic Ontology’) Ontologies are domain specific models, usually lacking linguistic information (Po. S, Morphology, Syntax etc. ) To be Integrated in One Resource or Kept/Accessed Separately? Standardization Þ Format Web-based Standards for Lexical Semantic Representation will Increase their Uptake (Easy Plug-and-Play, Remote Access, etc. ) Content Used Semantic Resources (Lexical) Þ will Widely lead to (Further) Semantic Standardization T. Declerck, P. Buitelaar 24
Defining a Linguistic Ontology for the Art World (Tentative) <rdf: RDF xmlns: rdf =”http: //www. w 3. org/1999/02/22 -rdf-syntax-ns#” xmlns: rdfs =”http: //www. w 3. org/2000/01/rdf-schema#” xmlns: xsd =”http: //www. daml. org/2000/10/XMLSchema#” xmlns: daml =”http: //www. daml. org/2001/03/daml+oil#” xmlns: art =”http: //www. art-world. org/art-world#” > <daml: Ontology rdf: about=”Concepts in the Art World”> <daml: imports rdf: resources=”http: //www. daml. org/2001/03/daml+oil#”> </daml: Ontology> T. Declerck, P. Buitelaar 25
Defining Art World Concepts (Classes, “Synsets”) (Tentative) <daml: Class rdf: ID="art-world. 01"> <rdfs: label>art-world. 01</rdfs: label> <rdfs: sub. Class. Of rdf: resource="http: //www. art-world. org/art-world. 00#"/> </daml: Class> <art-world. 01 rdf: ID="work"/> <art-world. 01 rdf: ID="painting"/> <daml: Class rdf: ID="art-world. 02"/> <art-world. 02 rdf: ID="beautiful"/> <art-world. 02 rdf: ID="colourful"/> <daml: Class rdf: ID="art-world. 03"/> <art-world. 03 rdf: ID="paper"/> <art-world. 03 rdf: ID="canvas"/> T. Declerck, P. Buitelaar 26
Defining Properties (“Selection Restrictions”) (Tentative) <daml: Object. Property rdf: ID="manner"> <rdfs: range rdf: resource="#art-world. 02"/> <rdfs: domain rdf: resource="#art-world. 01"/> </daml: Object. Property > <daml: Object. Property rdf: ID="medium"> <rdfs: range rdf: resource="#art-world. 03"/> <rdfs: domain rdf: resource="#art-world. 01"/> </daml: Object. Property > </rdf: RDF> T. Declerck, P. Buitelaar 27
(Semantic) Lexicons will be… …an Important Part of the Semantic Web …Represented Using Markup Languages (RDF) …Accessible in a Remote, Distributed Fashion …Central to Further Semantic Standardization T. Declerck, P. Buitelaar 28
Multilingual terminological lexicon, attached to a domain ontology (MUMIS) <lex-element id="ID" concept="Shot-on-goal"> <. . . lang="DE" type="main">Torschuss</term> <. . . lang="EN" type="main">shot on goal</term> <. . . lang="NL" type="main">schot op doel</term> <definition>ein Angriffsspieler kickt den Ball zu den gegnerischen Tor</definition> <. . . lang="DE" type="synonym">Distanzschuss</term> <. . . lang="DE" type="synonym">Nachschuss</term> <. . . lang="DE" type="synonym">Schuss</term> <. . . lang="DE" type="synonym">abzieh</term> </lex-element> T. Declerck, P. Buitelaar 29
Extension and Formalization of the multilingual terminological lexicon, including syncategorematic information. Supporting WSD. <lex-element id="ID" concept="Shot-on-goal"> <. . . lang = "DE" type = "main„ pos = „N“ mod = {„von concept = „Player“ | concept = „player“ gender = „gen“ | pos = „posspron“ } >Torschuss</term> <. . . lang="DE" type="synonym„ pos = „V“ comp = {„SUBJ“ concept = „Player“} >abzieh</term> <definition>URL: DFB home page/glossary</definition> </lex-element> T. Declerck, P. Buitelaar 30
Integrating. Syntacticand Domain Knowledge Including Syntactic Analysis for a more accurate tagging of domain specific semantic annotation T. Declerck, P. Buitelaar 31
Abstraction over Syntactic Annotation Ontology_3: Dependencies Head Comp Mod Spec Ontology_1: NP Head: N Mod: {Adj*, PP? , Gen. NP} Spec: {Det? Poss. Pron? } Type: {Ref. NP, Pro. NP, Date. NP, etc. } T. Declerck, P. Buitelaar Ontology_2: PP Head: Prep Type: {Loc. PP, Date. PP, etc. } Comp: NP Ontology_4: Grammatical Functions Subject, Object, Ind. Object NP Adjunct, PP Adjunct, etc. . 32
Merging of Syntactic and Domain Knowledge Example of a possible rule for conceptual annotation: If (Head of Subj_NP of Verb[type=soccer: : shot-on-goal] is a person) => { annotate head of NP with semantic class “soccer: : player”; …} Example of a rule for Instance Filling: If (term annotated with concept “soccer: : player”) => { try to find information about relations “Team”, “Age” etc. } (Template Filling in Information Extraction). T. Declerck, P. Buitelaar 33
NLP-based knowledge markup T. Declerck, P. Buitelaar 34
Much. More: DTD for Annotation umlsterms umlsterm id from to cui code pref tui msh xrceterms semrel id term 1 term 2 type ewnterms ewnterm sense gramrels document xrceterm gramrel id type chunk id from to type sentence chunks text T. Declerck, P. Buitelaar token offset id pos lemma 35
Much. More Linguistic. Annotation : (Lemmatization, POS, Basic Chunking) Balint syndrom is a combination of symptoms including simultanagnosia, a disorder of spatial and object-based attention, disturbed spatial perception and representation, and optic ataxia resulting from bilateral parieto-occipital lesions. <text> <token id="w 1" pos="NN">Balint</token> <token id="w 2" pos="NN">syndrom</token> <token id="w 3" pos="VBZ" lemma="be">is</token> <token id="w 4" pos="DT" lemma="a">a</token> <token id="w 5" pos="NN" lemma="combination">combination</token> <token id="w 6" pos="IN" lemma="of">of</token> <token id="w 7" pos="NNS" lemma="symptom">symptoms</token> . . . <token id="w 20" pos="JJ" lemma="spatial">spatial</token> <token id="w 21" pos="NN" lemma="perception">perception</token> <token id="w 22" pos="CC" lemma="and">and</token> <token id="w 23" pos="NN" lemma="representation">representation</token> . . . </text> <chunks> <chunk id="c 1" from="w 1" to="w 2" type="NP"/> <chunk id="c 7" from="w 20" to="w 23" type="NP"/> </chunks> > T. Declerck, P. Buitelaar 36
Much. More Semantic Annotation : (UMLS, Euro. Word. Net) Balint syndrom is a combination of symptoms including simultanagnosia, a disorder of spatial and object-based attention, disturbed spatial perception and representation, and optic ataxia resulting from bilateral parieto-occipital lesions. <umlsterm id="t 7" from="w 20" to="w 21"> <concept id="t 7. 1" cui="C 0037744" preferred="Space Perception" tui="T 041"> <msh code="F 2. 463. 593. 778"/> <msh code="F 2. 463. 593. 932. 869"/> </concept> </umlsterm> <umlsterm id="t 8" from="w 26" to="w 26"> <concept id="t 8. 1" cui="C 0029144" preferred="Optics" tui="T 090"> <msh code="H 1. 671. 606"/> </concept> </umlsterm> <semrel id="r 7" term 1="t 7. 1" term 2="t 8. 1" reltype="issue_in"/> <ewnterm id="e 2" from="w 21" to="w 21"> <sense offset="0487490"/> <sense offset="3955418"/> <sense offset="4002483"/> </ewnterm> T. Declerck, P. Buitelaar 37
MUMIS: DTD for Linguistic Annotation Subord-Clause Adv. P Document Paragraph Sentence AP NE NP PP T. Declerck, P. Buitelaar VG 38
MUMIS: DTD for Linguistic Annotation TYPE STRUK AP AP_AGR STRING W T. Declerck, P. Buitelaar AP_HEAD 39
MUMIS: DTD for Linguistic Annotation TYPE VG_SUBCAT_STEM VG_AGR STRING VG SENT_STRING KLAMMER STRUK W VG_STRG VG_HEAD. . . T. Declerck, P. Buitelaar VG 40
MUMIS: DTD for Linguistic Annotation W T. Declerck, P. Buitelaar STEM INFL POS TC TYPE STRING CLAUSE_PP_ADJUNKT CLAUSE_SUBJ SENT_STRING CLAUSE_PRED_SUBCAT CLAUSE_TYPE CLAUSE_VG_LIST CLAUSE_PP_LIST CLAUSE_NP_LIST CLAUSE_PRED_STRG CLAUSE_PRED_AGR. . . 41
MUMIS: Linguistic Annotation (Lemmatization … Dependency Structure) Industrie, Handel und Dienstleistungen werden in der ersten Liste aufgeführt, wobei die in Klammern gesetzten Zahlen auf die Mutterfirmen hinweisen. (Industry, trade and services are mentioned in the first list, in which numbers within brackets point to parent companies. ) <chunks> <chunk id="c 1" from="w 1" to="w 5" type="NP" head=”w 1, w 3, w 5”/> <chunk id="c 2" from="w 6" to="w 6" type="VG"/> <chunk id="c 3" from="w 7" to="w 10" type="PP" head=”w 7” complement=”w 8, w 9, w 10”/> <chunk id="c 4" from="w 11" to="w 1" type="VG"/> …. </chunks> <clauses> <clause id="cl 1" from="c 1" to="c 4" pred_struct="c 2 c 4" GF_Subj="c 1"/> <clause id="cl 2" from="c 6" to="c 9" pred_struct="c 9" GF_Subj="c 6"/> </clauses> T. Declerck, P. Buitelaar 42
MUMIS: Semantic Annotation (Events) 7. Ein Freistoss von Christian Ziege aus 25 Metern geht über das Tor. <chunks> <chunk id="c 1" from="w 1" to="w 5" type="NP" head=”w 2” pp modifier=”w 3 w 4 w 5”/> <chunk id="c 2" from="w 6" to="8" type="PP" head=”w 6” complement=”w 7 w 8”/> <chunk id="c 3" from="w 9" to="9" type="VG"/> <chunk id="c 4" from="w 10" to="w 12" type="PP" head=”w 10” complement=”w 11 w 12”/> </chunks> <clauses> <clause id="cls 1" from="c 1" to="c 4" pred_struct="c 3“ GF_Subj="c 1"/> </clauses> <events> <event id="e 1" clause=”cls 1” event-name=”free-kick”> <arguments> <argument id="arg 1" name="player” value=”w 4, w 5”/> <argument id="arg 2" name="location” value=” 25 -meter”/> <argument id="arg 3" name="time” value=” 07: 00”/> </arguments> </event> <event id="e 2" clause=”cls 1” event-name=”goal-scene-fail”> <arguments> <argument id="arg 1" name="player” value=”w 4, w 5”/> <argument id="arg 2" name="location” value=” 25 -meter”/> <argument id="arg 3" name="time” value=” 07: 00”/> </arguments> </event> </events> T. Declerck, P. Buitelaar 43
Conceptual Annotations for Multimedia Indexing and Retrieval: A multilingual cross-document and incremental IE approach (MUMIS) Technology development to automatically index (with formal annotations) lengthy multimedia recordings (off-line process): Find annotate relevant entities, relations and events Technology development to exploit indexed multimedia archives (on-line process): Search for interesting scenes and play them via Internet Test Domain: Soccer Games / UEFA Tournament 2000 T. Declerck, P. Buitelaar 44
Indexing by. . . Off-line Task Automatic Speech Recognition (Radio/TV Broadcasts) Automatically transforms the speech signals into texts (for 3 languages — Dutch, English and German) Natural Language Processing (Information Extraction) Analyse all available textual documents (newspapers, speech transcripts, tickers, formal texts. . . ), identify and extract interesting entities, relations and events Merging all the annotations produced so far Create a database with formal annotations Use video processing to adjust time marks T. Declerck, P. Buitelaar 45
Information Extraction (IE) is the task of identifying, collecting and normalizing relevant information for a specific application or user. The relevant information is typically represented in form of predefined “templates”, which are filled by means of Natural Language (NL) analysis. IE combines pattern matching mechanisms, (shallow) NLP and domain knowledge (terminology and ontology). T. Declerck, P. Buitelaar 46
Information Extraction (2) IE is generally subdivided in following tasks: - Named Entity task (NE) - Template Element task (TE) - Template Relation task (TR) - Scenario Template task (ST) - Co-reference task (CO) T. Declerck, P. Buitelaar 47
Subtask of IE Named Entity task (NE): Mark into the text each string that represents, a person, organization, or location name, or a date or time, or a currency or percentage figure. Template Element task (TE): Extract basic information related to organization, person, and artifact entities, drawing evidence from everywhere in the text. T. Declerck, P. Buitelaar 48
Subtask of IE (2) Template Relation task (TR): Extract relational information on employee_of, manufacture_of, location_of relations etc. (TR expresses domain-independent relationships). Scenario Template task (ST): Extract pre-specified event information and relate the event information to particular organization, person, or artifact entities (ST identifies domain and task specific entities and relations). Co-reference task (CO): Capture information on coreferring expressions, i. e. all mentions of a given entity, including those marked in NE and TE. T. Declerck, P. Buitelaar 49
IE applied to soccer Terms as descriptors for the NE task Team: Titelverteidiger Brasilien, den respektlosen Außenseiter Schottland Player: Superstar Ronaldo, von Bewacher Calderwood noch von Abwehrchef Hendry, von Jackson als drittem Stürmer, Torschütze Cesar, von Roberto Carlos (16. ), Referee: vom spanischen Schiedsrichter Garcia Aranda Trainer: Schottlands Trainer Brown, Kapitän Hendry seinen Keeper Leighton Location: im Stade de France von St. Denis (more fine-grained location detection would be: Stadion: im Stade de France and City: von St. Denis ) Attendance: Vor 80000 Zuschauern T. Declerck, P. Buitelaar 50
IE applied to Termssoccer (2) for NE Task Time: in der 73. Minute, nach gerade einmal 3: 50 Minuten, von Roberto Carlos (16. ), nach einer knappen halben Stunde, scheiterte Rivaldo (49. /52. ) jeweils nur knapp, das vor der Pause Versäumte versuchten die Brasilianer nach Wiederbeginn, . . . Date: am Mittwoch, der Turnierstart (? ), im WMEröffnungsspiel (? ) Score/Result: Brasilien besiegt Schottland 2: 1, einen 2: 1 (1: 1) -Sieg, der zwischenzeitliche Ausgleich, in der 4. Minute in Führung gebracht, köpfte zum 1: 0 ein T. Declerck, P. Buitelaar 51
IE applied to Relations for TR Task: soccer (3) Opponents: Brasilien besiegt Schottland, feierte der Top. Favorit. . . einen glücklichen 2: 1 (1: 1)-Sieg über den respektlosen Außenseiter Schottland, Player_of: hatte Cesar Sampaio den vierfachen Weltmeister. . . in Führung gebracht, Collins gelang. . . der zwischenzeitliche Ausgleich für die Schotten, der Keeper des FC Aberdeen, Brasiliens Keeper Taffarel Trainer_of: Schottlands Trainer Brown. . . T. Declerck, P. Buitelaar 52
IE applied to soccer (4) Events for ST task: Goal: in der 4. Minute in Führung gebracht, das schnellste Tor. . . markiert, Cesar Sampaio köpfte zum 1: 0 ein, Collins (38. ) verwandelte den Strafstoß, hätte Kapitän Hendry seinen Keeper Leighton um ein Haar zum zweiten Mal bezwungen, von dem der Ball ins Tor prallte Foul: als er den durchlaufenden Gallacher im Strafraum allzu energisch am Trikot zog Substitution: und mußte in der 59. Minute für Crespo Platz machen. . . T. Declerck, P. Buitelaar 53
NL Processing and Knowledge Markup of (German) soccer texts with the SCHUG system A multilingual ontological lexicon • • Formal Text 1 Formal Text 2 XML Soccer Annotation for Text 1 XML Soccer Annotation for Text 2 Merging of Annotations for Formal Texts Semi-Formal Text annotated with Soccer Information (XML) T. Declerck, P. Buitelaar 54
Multilingual Extension Spanish (Esperonto) Ontology: <lex-element id="ID" concept=„Second-half"> <. . . lang="DE" type="main">zweite Halbzeit</term> <. . . lang="EN" type="main">second half</term> <. . . lang=„ES" type="main">reanudacion</term> </lex-element> …. Processing with the SCHUG system: Example T. Declerck, P. Buitelaar 55
Conceptual Annotations for Multimedia Indexing and Retrieval: MUMIS Automatic Speech Recognition Formal Text Formal Text Formal Text Speech Text Formal Text Formal Text Formal Text Trans. Text ASR Signals Domain Modeling Ontology Formal Text Formal Text Formal Text Soccer Text cripts DM Texts Information Extraction Formal Text Formal Text Formal Text Legend EN Merging Merged Annotated formal text Formal Text Formal Text Formal Text Free Text DE FORMAL T. Declerck, P. Buitelaar Formal Text Formal Anno. Text IE tations Text User Interface NL Ontology Domain Lexicon Ontology Query Merging UI Annotations 56
The first user interface of MUMIS T. Declerck, P. Buitelaar 57
Esperonto: Partners Intelligent Software Components (Coord) : Semantic Web, Annotation Services. UPM : ontology development and evaluation. University of Innsbruck : Semantic Web languages. Saarland University : multilingual Annotation services, using Information Extraction UNILIV : Semantic indexation of Semantic Web content. Routing solutions. Visualization and navigation to make content presentation user -friendlier. Residencia de Estudiantes : Content provider. Cultural tour test case. Evaluation. CIDEM : Content provider. Fund finder test case. Evaluation. Bio. Vista : Content provider. Scientific Discovery test case. T. Declerck, P. Buitelaar 58
Aim Application Service Provision of Semantic Annotation, Aggregation, Indexing and Routing of Textual, Multimedia, and Multilingual Web Content The project aims at bridging the gap between the actual World Wide Web and the semantic Web by providing a service to "upgrade" existing content to semantic Web content. Ontologies play a key role in this effort, together with multilingual Natural Language Analysis of textual documents currently in the web as free or HTML encoded texts. T. Declerck, P. Buitelaar 59
Main Goals q To bridge the gap between the current web and the Semantic Web: Sem. ASP q Ontology-based annotation q Sources: q Static pages q Pages dinamically generated from DB q Textual and multimedia information q Web services q Added value knowledge-based services on top of the constructed semantic web q Routing based on P 2 P communication q Semantic aggregation q Meaning negotiation q Support Multilinguality on ontology construction, . . . T. Declerck, P. Buitelaar 60
Agent Visualization Service Provider Applications Multilingual NL Generation Semantic Web Router Portal Agent Router Semantic indices, Concept instances Certificate Multilinguality Ontology Repository Service Workbench Sem. ASP Maintenance Reengineering Mapping Tagger/ Wrapper Multilingual NL Understanding XML DAML OIL RDF(S) World Wide Web T. Declerck, P. Buitelaar Web Server Provider Dynamic Information Provider Static 61 Multimedia Data Information Provider
Ontology-based Annotation Annotate accurately document with concepts and terms described in various semantic resources: Euro. Word. Net, UMLS, Soccer ontology etc. Annotate documents with relations defined in the ontology T. Declerck, P. Buitelaar 62
Ontology construction from Text There are various methodologies under investigation for extracting/learning knowledge from text, and to encode it in an ontology (see Ontology Learning Overview - Onto. Web D 1. 5 http: //www. ontoweb. org). Many are based on Machine Learning techniques We discuss here the possibility of a rule-based approach for partial and shallow ontology construction from text, based on various levels of syntactic patterns annotated in the documents. T. Declerck, P. Buitelaar 63
Ontology construction from Text: A starting experiment: Medicine Document Set: 65 sample phrases that link symptoms with Rheumatoid Arthritis (RA). T. Declerck, P. Buitelaar 64
Ontology construction from Text: Apposition and Paranthesis (1) The effects of rheumatoid arthritis on bone include structural joint damage (erosions) and osteoporosis “ “ Linguistic Structure: [[The effects of rheumatoid arthritis] [on bone]] [include] [[structural joint damage ( erosions )] [ and] [osteoporosis]] => The Apposition (2 syntactic heads “joint” and “erosions” in one NP) including a parenthesis construction suggests a synonymy relation or a definition. Heuristic: Establishing Semantic Relations on the top of linguistic “head-modifiers” constructions T. Declerck, P. Buitelaar 65
Ontology construction from Text: Apposition with Paranthesis (2) “For symptoms of rheumatoid arthritis (pain, joint stiffness), the reference treatment is a nonsteroidal antiinflammatory drug (NSAID) such as diclofenac or ibuprofen. ” Linguistic Structure [For symptoms of rheumatoid arthritis ( pain , joint stiffness )] , [the reference treatment] [is] [a nonsteroidal antiinflammatory drug ( NSAID)] ÞSuggesting a semantic relation between („pain“ and „joint stiffness“) ÞClassify „pain“ and „joint stiffness“ as symptom of RA. The word „symptom“ is linguistically annotated as the head of the Compl-NP of the PP starting with „For“. T. Declerck, P. Buitelaar 66
Ontology construction from Text: Apposition with Paranthesis (3) But there is a need for constraining the hypothesis: “In patients with rheumatoid arthritis (RA)” => RA is abbreviation of rheumatoid arthritis And in the sentence: “Fourteen consecutive elbows have been treated for rheumatoid arthritis (9 elbows) and for post-traumatic osteoarthrosis (5 elbows) by total elbow replacement with the GSB III implant. “, the parenthesis (9 elbows) and (5 elbows) have no semantic relations to the preceding head nouns! T. Declerck, P. Buitelaar 67
Ontology construction from Text: Apposition with commas “Etoricoxib, a selective COX 2 inhibitor, has been shown to be as effective as non-selective non-steroidal anti-inflammatory drugs in the management of chronic pain in rheumatoid arthritis and osteoarthritis, …” Linguistic Structure: [Etoricoxib, a selective COX 2 inhibitor, ] [has been shown]… The same hypothesis as in the former examples: a semantic relation between “Etoricoxib” and “selective COX 2 inhibitor”. Probably a “isa” relation T. Declerck, P. Buitelaar 68
Ontology construction from Text: Compound Analysis „Joints destructions, „joint damage“, „joint disease“, „joint stiffness“ but „joint cartilage“. „Knee joints“ vs. „tender joints” What can happen to joins, where are joints located? . Use of synsets to detect relations? „Joint cartilage“ is not a disease. T. Declerck, P. Buitelaar 69
Ontology construction from Text: PP post-modification „inflammation of joints, synovial lining of joints” Here: use of synsets for grouping that what can happen to joints? T. Declerck, P. Buitelaar 70
Ontology construction from Text: Phrase Internal Coordination “The effects of rheumatoid arthritis on bone include structural joint damage (erosions) and structural joint damage “ Linguistic Structure: [[The effects of rheumatoid arthritis] [on bone]] [include] [[structural joint damage ( erosions )] [ and] [osteoporosis]] ÞRA causes structural joint damage AND structural joint damage (interpreting the head noun “effects” as a causation). ÞHypothesis: The two heads of an NP coordination are somehow related. T. Declerck, P. Buitelaar 71
Ontology construction from Text: Phrase Internal Coordination (2) “A study was conducted to determine the incidence of ulnar and peripheral neuropathy “ Linguistic Structure: … [The incidence of [[ulnar and peripheral] neuropathy]] Þ The AP “ulnar and peripheral” AP modifies the head noun “neuropathy”. The AP is a coordinated one, having two Adjectival heads. Þ Hypothesis: They correspond to two types of neuropathy T. Declerck, P. Buitelaar 72
Ontology construction from Text: Subject Verb Objetcs (Ind. Obj. etc. ) [Rheumatoid arthritis is an immunologically mediated inflammation of joints of unknown aetiology] and [often leads to disability] => RA leads to Disability (effect of ellipsis resolution: RA detected as the subject of the verb „leads“, even if not realised in text. Reference resolution very important for knowledge extraction) => Lexical semantic info: collects all objects of RA leads to … =>Suggest Causality (verb lead + to) T. Declerck, P. Buitelaar 73
Ontology construction from Text: Subject Verb Objects (Ind. Obj etc. ) “These changes constitute hallmarks of synovial cell activation and contribute to both chronic inflammation and hyperplasia” On line exercise! T. Declerck, P. Buitelaar 74
Future Work Still have to identify accurately the sub-set of linguistic tags, describing syntactic/semantic patterns that are relevant for ontology extraction (or even ontology mark-up). T. Declerck, P. Buitelaar 75
First Conclusions Construction of partial and shallow ontologies from (complex) syntactic patterns seems feasible. It might seem “expensive” in the sense that documents first should be (automatically) linguistically annotated. But Machine Learning methods also needs a lot of semi-automatically annotated data for training. A need to conduct a comparative evaluation taking into account as many parameters as possible. T. Declerck, P. Buitelaar 76
Practical Sessions (Adrian Raschip) • Exercise 1 : Semi-Automatic Terminological extension: Romanian and other languages. On the base of the TMX encoded MUMIS multilingual terminology • Exercise 2 : (Manual) linguistic annotation of English and Romanian Text on Soccer • Exercise 3 : Define a soccer ontology in Protégé • Exercise 4: Search for possible mapping rules between linguistic annotations and relations that might be relevant to be extracted T. Declerck, P. Buitelaar 77
Скачать презентацию Language Technology and the Semantic Web Thierry Declerck
b6fb7b734c260bed2514aff1bafd9154.ppt