Formaliser les langues avec l’ordinateur : de INTEX à Nooj
| , ,Quatrieme partie. Perspectives
23. Semantic Web and linguistic methodologies
Texte intégral
Introduction
1Today’s systems of information retrieval are too general and imprecise to assure the exact recovery of information, in internet searches.
- 2 Look at the web site : www.apogeonline.com.
- 3 Let us consider, among others, W3C and linguistic laboratories which research new languages, codes (...)
2On one hand the incredible growth of the World Wide Web which reached 166 million web surfers in 2002 only in United States2, on the other hand, the big presence of Commercial Tools in the search engine market (Numerico 2003, 85-86), push different laboratories and centers of research to find new and better methodologies for the description, indexing and retrieval of web documents3.
3Beyond a “Surface Web” (Numerico 2003, 90-91), which is easily navigated by spiders and crawlers thanks to specific algorithms like Google’s PageRank, there is also a “Deep Web” which is still unexplored. Google’s algorithm, in fact, uses a social criteria to select information, because it considers the importance of a web-site or of a web-resource, to derive from the number of hyperlinks to it that exist. If we consider the big investments which characterize some commercial web sites, we can understand how the importance of a web resource is often connected to economic reasons (Numerico 2003, 90).
4AltaVista, another famous search engine, indexs HTML documents moving from classical and general metatag (title, author, date...), but it does not use linguistic methodologies useful to “put a word in the right linguistic context” (De Bueriis 2002, 29). Moreover, AltaVista uses traditional systems of research, like the boolean method (Leloup 1998, 160), which is not good enough to retrieve a complex linguistic resource like a digital text.
- 4 Interview with Umberto Eco, La Repubblica, september 2003.
5In a recent interview, Umberto Eco said that “looking for Grall in the web I had stopped at the 70th web-sites : just two were reliable, but one of these two was also banal”4. Others experts of the language ask simple and quick questions : ”how can we avoid getting lost in the big Ocean of digital words ?” (Elia 2001, 1).
I. The Semantic web and its languages
- 5 http://www.scientificamerican.com/article.cfm?articleID=00048144-10D2-1C70- 84A9809EC588EF21&catID= (...)
6In a famous article written in May 2001, Tim Berners Lee, James Hendler and Ora Lassila described the project of a new web based on Semantics5.
7In his new book titled “Weaving the web”, Lee underlines the logical and semantic insufficiency of present Search Engines, where it is not possible to distinguish rialable resources from unrielable ones (Lee 1999, Weaving the web, 155), and he proposes a new logical web, based on XML, RDF and ontologies (Lee 1999, 158-171). This is a project not very far from the dream of artificial intelligence, imagined by computational linguistics and experts of I.A. (Cordeschi 2001, 7-10).
8From Leibniz to Babbage, from Chomsky to Gross we can observe the permanence of the dream to build an Intelligence Engine, capable of recognizing words, concepts and meanings. In this way, this Engine could “replace the human mind in complex activities like reasoning or the translation from one language to others” (Elia 1996, 1).
9Before considering the linguistic limits and problems of Lee’s intelligence-Semantic Engine, let us describe Web Semantic’s languages and technologies.
10First of all, let us speak about XML.
1. Something about XML
11The eXtensible Markup Language was created in 1996 by W3C (World Wide Web Consortium) and it is a sublanguage of SGML (Standard Generalized Markup Language).
12XML is a markup language like HTML (Hypertext Markup Language), but it was designed to carry data (Tittel 2003, 1). If HTML was created to display data graphically, XML was created to describe data in a logical and semantic way. One of the biggest differences between these two markup languages is that XML tags are not predefined, because each user can define his own tags, as in this example, where we create a simple note-document :
< ?xml version = “1.0”encoding = “ISO-8859-1” ?>
<note>
<to> Lucy </to>
<from> Mike </from>
<heading> Reminder </heading>
<body> Don’t forget the conference this week ! </body> </note>
13XML documents use a Document Type Definition (DTD) or an XML Schema, to describe the data, where each user inserts personal tags useful to define the logic of the document.
14This is different from HMTL where there is just one tag – <META> – to recognize some meta-semantical aspects of the texts (the author, the general content, the date…), XML documents use a complete meta-definition which describe all of the tags of the texts and their occurencies.
2. RDF-RDF Schema
15After XML, another technology of the stack of W3C is RDF (Resource Description Framework). We can distinguish RDF and RDF Schema.
16RDF is a datamodel for resources and relations between them. It provides a simple semantics for this datamodel, and these data-models can be represented in an XML Syntax.
17RDF Schema is a vocabulary for describing properties and classes of RDF resources, with a semantics for generalization-hierarchies of such properties and classes.
18In general, we can define RDF technology as a “formal semantics”.
3. Ontologies
19The last essential element of the Semantic Web is the Ontologies (OWL). OWL is a taxonomical description of a concept or a word. After RDF, it adds more vocabulary for describing properties and classes (relations between classes, cardinality, equality, richer typing of properties, characteristics of properties). It is possible to distinguish three sublanguages designed for use by different communities of implementers and users : OWL LITE, OWL DL and OWL Full.
20The first one supports those users primarily needing a classification and simple contraints. The second one supports those users who want the maximum expressiveness while retaining computational completeness and decidability. The third sublanguage is for those users who want maximum expressiveness with no computational guarantees.
21OWL is composed of classes and individuals. Classes are categories of individuals that belong together because they share some properties. Every individual of the OWL world is a member of “OWL : Thing”. Classes can be organized hierarchically using the tag Subclassof.
22Here an example for the Liquer domain, where we use 3 general classes : “Distillation”, “Region”, “Alcholic and distilled thing” :
< owl : Class rdf : ID = “Distillation”/>
< owl : Class rdf : ID = “Region”/>
< owl : Class rdf : ID = “Alcoholic and sweet Thing”/>
23In addition to classes we have the individuals. If a class is a general category and a collection of properties, individuals are the members of those sets. To define the Liqueur Domain we first describe “Region” and “Lemon grove” individuals and then we define our first liqueur, “The limoncello” :
<Region rdf : ID = “Sorrento”/>
<locatedIN rdf : resource = “# Campania”/>
</Region>
<Lemon grove rdf : ID = “Sorrento lemon grove”/>
<Limoncello
rdf : ID = “Sorrento Limoncello”>
<Located IN rdf : resource = “#Sorrento Region”/>
<hasMaker rdf : “#Sorrento lemon grove”/>
</Limoncello>
II. OWL and DTD : versioning and problems
24The purpose of Web Semantic is to build an universal and valid Logic Engine for the description and retrieval of web resources. We can observe some theorical and methodological limits for the realization of this project.
25One problem is that DTD are differently realized from different kinds of users which invent their own tags implies that we are moving to a sort of “Babel’s XML tags”, like the same T.B. Lee observes in his book (Lee 2001, 142). In this sense, universal linguistic definitions could be useful to guarantee uniform textual descriptions ; we can for example imagine to replace general XML tags with linguistic ones, following Gross’s tables (Gross 1975).
26Another big problem is connected to Ontologies versioning and philosophy.
27Ontologies are like softwares, they will be maintained and will change over time. They can also be differently organized from different group of research. Moreover, in a linguistic perspective, with ontologies we risk needing to come back to first Wittengstein theories and to Katz ones (Putnam 1975), because here we are trying to fix the meaning of words in a unique and definitive way. But, is it really possible to speak about “Universal Semantic Descriptions” (Alinei 1974, 183-186) ?
28If it has been possible to find universal descriptors for phonology and morphology, linguistics found big difficulties at the moment they wanted to fix and describe the meaning of the words (Alinei 1974, 9-25).
29From one side the synchronic and diachronic change of languages (Saussure 1922, 172), from another side their ambiguity seem to suggest that a definitive semantic description of natural languages is not possible. Also the famous Chomsky’s exemple about three english words – Kill, Murder and Assassinate (1970) – uses a semantic description which is typical of English. In fact in Italian, for example, there is not a specific word for political killing. Moreover, if only we consider a semantic tool like Thesaurus, we can see that it is not capable to distinguish the double meaning occurences of the word “operation” : surgical or mathematical (Leloup 1998, 74).
30For all of these reasons we can say that “it is not possible to research universal semantics, but is possible to research dominant traits determined by social-historical evolution” (Alinei 1974, 186). If we do not want ontologies to become a sort of Thesaurus, not capable of resolving linguistic ambiguities and complexities because an artificial and monistic description of the concepts-words of a language contradicts the systemic and syntactic movement of that language, we need to put that word in the right lexico-grammatical context (De Bueriis 2002).
III. Some linguistic tools for the description of textual resources
31Linguistic tools created to describe and index digital texts enabled one to retrieve digital resources with a high level of pertinence, using lexico-grammatical categories and rules. Moving from the idea that semantics cannot be described in a scientific way, because of the transformation of lexical meaning in space and time (Alinei 1974, 183-188), and considering that it is the sentence’s context which defines the semantic interpretation of a word (Elia, Vietri in Burattini-Cordesci 2001, 204), these tools clarify semantic aspects and to resolve any ambiguities, thanks to lexico-grammatical descriptions.
32First of all we are referring to INTEX by Max Silberztein which is a powerful Tool, capable of eliminating linguistic ambiguities using local grammars and specific translators (Silberztein 2004, INTEX manual, 146-147). With the “linear tagging”, for example, “the ambigous forms are replaced by the separation of all of the corresponding lexical entries” (Silberztein 2004, 150-151). Moreover, in a semantic perspective, this tool is also capable of finding equivalent semantic structures, like “Reprises des activités, rachat d’activité, acquérir des magasins” (Poibeau 2003, 174-175).
33Others tools created by the University of Salerno use linguistic analysis to describe and retrieve digital texts. The tool, Cataloga, localizes the technical compound-nouns of a text, defines semantics domains from the occurrences of those words and finally decides the typology of text (economical, technological, medical…). The tool, Facilmente, replaces compound-words placed in a text with simpler structures.
IV. Building ontologies and DTD with Name Entities, Dictionaries, grammars and graphs.
34In his book “Extraction automatique d’information ; du texte brut au web sémantique”, Thierry Poibeau suggests using Name Entities for the organization of ontologies (Poibeau 2001, 87-116). This proposal is interesting not only because Name Entities are linguistic’s elements that are good indicators for the content of a text, but also because he finds a trait d’union between the actual tendency of Web Semantic and Linguistic methodologies.
35If we are really looking for a better Search Engine to reorganize web resources, Linguistics can and must propose approaches to acheive this.
36In this sense, technical compound words and Name Entities can be implemented for linguistic’s DTD and linguistic’s ontologies. Dictionaries, Grammars and Graphs can be used to describe and retrieve digital texts in a universal way, which is not possible with the artificial and atomistic systems proposed by Semantic Web.
37Complete ontologies of Name Entities should be produced for different languages, distinguishing, for example, between :
Nome di persona <N+NPR> Max
Nome di società <N+SOC> NooJ
Nome di società <N+Loc+City> Paris
38In fact, if it is true that technical compound-words like “scheda perforata”, “carta di credito”, “società per azioni” (Vietri 2001, 40-41) are good indicators of the content of a technical text, Name Entities are good indicators of the content of more generic ones. They are very common in all sort of texts and we can imagine using a tag-language to recognize them, like in the example from Poibeau’s book (2003, p. 98-99) :
With heavy rains flooding northern <LOCATION> France
</LOCATION>, a cartoon in L’Express last week showed
<PERSON> Mr. Chirac </Person>…
39A great deal of work is necessary to organize complete dictionaries and grammars for a new Semantic-Linguistic Engine capable to recognize equivalencies between words or sentences and, more generally, to retrieve the exact digital resource we need.
- 6 To clarify the concept of linguistic “ansia” look at De Bueriis : “Le parole come ordine del mondo” (...)
40But, at the same time, a linguistic lexico-grammatical Engine with semantic’s capacities is interesting, because it could try to solve two big “ansie” (anxieties)6. From one side, the of finding the information we are really looking for on the Web. From another side a bigger one : the problems of describing and understanding the mysteries and secrets connected to our natural languages.
Bibliographie
Bibliography
ALINEI, M. (1974), La struttura del lessico, il Mulino, Bologna.
BLOOMFIELD L. (1933), One word at a Time, The Hague- Paris, Mouton.
BURATTINI, E., CORDESCHI, R. (2001), Intelligenza artificiale, Roma, Carocci.
CALISE M., LOWI T. (2000), “Hyperpolitics : Hypertext, Concepts and Theory-making”, in International Political Science Review, July 2000, Number 3.
CHOMSKY N. (1970), “Remarks on Nominalization”, in R.A. JACOBS e P.S. ROSENBAUM, trad. it., Saggi Linguistici.
DE BUERIIS G. (2002), Le paole come ordine del mondo, Editoriale Scientifica, Napoli.
ELIA A. (1996), “Elogio dell’imperfezione. Lingue perfette, macchine e intelligenza artificiale”, in GAMBARA D., GENSINI S., PENNISI A. (eds.), Language Philosophies and the Language Sciences, Munster Nodus Publikationen.
ELIA A. (2001), Linguistica computazionale e multimedialità. Una semplice introduzione, Dipartimento di Scienze della comunicazione, Università di Salerno.
GROSS M. (1975), Méthodes en syntaxe, Paris, Hermann.
HARRIS Z.S. (1952), “Discourse Analisys”, in HARRIS (1970), Papers in Structural and Trasformational Linguistics, Dordrecht, Reidel.
HJELMSLEV L. (1943), Omkring sproteoriens grundlaeggelse, trad. it., I fondamenti della teoria del linguaggio, Torino, Einaudi, 1968.
LANDOW G. (1994), Hypertext 2.0. The convergence of Contemporary Critical Theory and Technology, trad. it., L’ipertesto. Tecnologie digitali e critica letteraria, Milano, Bruno Mondadori 1998.
LEE T. B. (2001), Weaving the web, L’architettura del nuovo Web, Milano, Feltrinelli Interzone.
LELOUP C. (2001), Moteurs d’indexation et de recherche, Eyrolles, Paris.
MERCER D. (2002), HTML, McGraw-Hill, Milano.
NUMERICO T., VESPIGNANI A. (2003), Informatica per le scienze umanistiche, il Mulino, Bologna.
POIBEAUS T. (2003), Extraction automatique d’information, du texte brut au web sémantique, Lavoisier, Paris.
PUTNAM H. (1975), Mind, Language and reality, Cambridge University Press, trad. it., Mente, linguaggio e realtà, Adelphi, Milano, 1987.
SAUSSURE (1922), Course de linguistique générale, Paris Editions Payot, trad. it., Corso di linguistica generale, Biblioteca Universale Laterza, 1997.
SILBERZTEIN M. (2004), INTEX manual.
TITTEL E. (2003), XML, The McGraw-Hill Companies, Milano.
VIETRI S. (2001), Navigare nei testi. Applicazioni in linguistica computazionale, Editoriale Scientifica, Napoli.
Notes
2 Look at the web site : www.apogeonline.com.
3 Let us consider, among others, W3C and linguistic laboratories which research new languages, codes and tag-systems to describe digital texts.
4 Interview with Umberto Eco, La Repubblica, september 2003.
5 http://www.scientificamerican.com/article.cfm?articleID=00048144-10D2-1C70- 84A9809EC588EF21&catID=2.
6 To clarify the concept of linguistic “ansia” look at De Bueriis : “Le parole come ordine del mondo”, Editoriale Scientifica, p. 171-174.
Le texte et les autres éléments (illustrations, fichiers annexes importés) sont sous Licence OpenEdition Books, sauf mention contraire.