Many are the schemas developed so far for the different kinds of annotation required in the field of Corpus Annotation. Besides, with the appearance of the Semantic Web (Berners-Lee et al., 1999) many other schemas have been devised (most of them based on ontologies (Gruber 1993; Borst 1997)) for web page annotation. Thus far, on the one hand, Corpus Linguistics researchers are trying to cover as many levels and aspects of annotation as possible to describe language phenomena -from a linguistic point of view (Wilson & Thomas, 1997; Schmidt, 1988); on the other hand, researchers in the Semantic Web area are focusing on achieving a sound model of semantic annotation for web pages, that is able to capture as much knowledge from these pages as possible, so that computers can process them in a much smarter way (Benjamins et al., 1999, Motta et al., 1999, Luke et al., 2000, Staab et al., 2000). However, there is an emerging road in-between, nowadays, that seeks to merge and sum up both kinds of annotations, combining them in order to bear a new, unified, multilingual, flexible, extensible and fully semantic model of annotation, useful for both communities (Aguado et al., 2003a). Moreover, as shown by the ISO - TC37SC4 (2003) "there is an increasing need for new standardization as well as urgent recognition of existing de facto standards and their transformation into International Standards". In fact, one of the main aims of this committee is "to develop standards and related documents to maximize the applicability of language resources". The OntoTag model for Semantic Web Annotation (Aguado et al., 2003b), whose Linguistic Ontologies we present here, is being developed following this in-between road aforementioned, as well as a number of guidelines hitherto published (EAGLES 1996a, 1996b; CES 1999; MILE 2003; GDA 2002), in order to achieve the goal of standardisation sought within the ISO - TC37SC4 committee.

The rest of this paper is organised as follows: firstly, the PLAN-H-SemWeb and ContentWeb projects are presented; then, in the following sections, the different ontologies developed within them for the model OntoTag are described and, finally, in the conclusion, we summarise the advantages of OntoTag focusing on those derived from its use of ontologies for annotation.

OntoTag's linguistic ontologies, which we present here, are being developed within a couple of Spanish government funded projects, namely, ContentWeb (up to the semantic level) and PLAN-H-SemWeb (further linguistic levels).

The main goal of these ongoing projects is to implement a multilevel, multifunctional hybrid annotation platform that helps to integrate the linguistic annotation levels into the scope of the Semantic Web, by incorporating also the semantic richness of ontology-based models. PLAN-H-SemWeb aims also at widening and complementing the preliminary results obtained in the ContentWeb project, both in the levels concerned and in the depth and detail of the other levels also included in ContentWeb.

On the one hand, so far, within ContentWeb, we have finished an exhaustive comparative analysis of the different alternatives within Corpus Linguistics and the existing standards and recommendations in this field (CES, 1999; EAGLES, 1996a, 1996b; Schmidt, 1988). This study, as well as our experience and knowledge, has been used to define the most appropriate annotation at each linguistic level, especially the lemmatic one, the morphosyntactic one and the syntactic one; a comparison of the results is being completed and will be presented shortly. On the other hand, the semantic and pragmatic levels are much less developed and currently there are neither established guidelines nor tools that perform fully annotated documents at all these levels. This has not allowed us to make complete practical comparisons such as the ones done for the other levels within PLAN-H-SemWeb. Nevertheless, some important steps have been taken in order to characterise the textual and non-textual meaningful units in this corpus-based study on cinema web pages. First, the main entities mapping concepts in the ontology have been identified and analysed following Pustejovsky's work (1995). Then, the semantic classes and the linguistic realizations of the evaluative expressions (adjectives (Bouillon & Viegas, 1999) and

NPs) that qualify those entities that have been extracted, as well as the lexical functions they perform (Melcuk, 1988, 1996). Second, the non-textual meaningful units are also worth to be considered when annotating web pages as different non-textual, metalinguistic elements are present and they incorporate invaluable information. For that reason, we have implemented software that pre-processes web pages to make them format-compliant for existing linguistic annotation tools (basically by removing tags, images, sound files, etc.). In order to 'reconstruct' the original contents of the pages, we are also developing an appropriate storage component for the non-textual, metalinguistic elements included in the pages.

We think that the kind of work described above can be useful in different areas of both CL and AI. Researchers from CL will have a platform to compare and integrate different linguistic annotation models. This platform will also allow a wider and more integrated reuse of linguistic resources. People from AI can apply it to the Semantic Web for different purposes such as Data Mining, Machine

Translation, and Information Retrieval and Extraction. In particular, we plan to use it in the Ontology Learning field. We intend to construct a (theoretical) base that presents a set of heuristics of use in our platform and that could be integrated into an ontology development tool, specifically in WebODE (2003), as part of a new annotation service (ODETag) that helps ontology engineers to carry out their work.

One of the main components of the OntoTag model is its set of linguistic ontologies, devised to represent the structure and relationships between the elements of language at different linguistic levels. The kind of elements and relationships considered in them are the ones usually included in existing annotation schemas and also those already discussed in the literature but not implemented yet (Wilson & Thomas, 1997; Schmidt, 1988) as well as some others, specified by our research team.

Figure 1 : OntoTag's Syntactic Unit Sub-Ontology.

First of all, a Linguistic Level Ontology (LLO) has been implemented both to capture the stratification of natural language analysis and generation and to simplify the study of the other elements. Then, following the EAGLES guidelines for morpho-syntactic annotation of corpora (EAGLES 1996a), but obviously broadening its scope, three different ontologies have been implemented to represent the category-attribute-value formalism at all levels of annotation (morpho-syntactic, syntactic, semantic, discourse and pragmatic): a Linguistic Unit Ontology (LUO), a Linguistic Attribute Ontology (LAO), and a Linguistic Value Ontology (LVO).

The Linguistic Unit Ontology (LUO - see Figure 1 for a partial view) includes all the units (categories) identified at the different levels of annotation considered in the LLO, and incorporates an adaptation of the SIMPLE (2000) ontologies at the semantic level; the Linguistic Attribute Ontology (LAO) includes the various attributes associated to the units in the LUO; and the Linguistic Value Ontology (LVO) accounts for the possible values of the attributes in the LAO.

Complementing these four ontologies, a fifth one, the Linguistic Pattern Ontology (LPO) has been designed for the representation of the patterns that these units follow when combined in an utterance. Finally, the OntoTag Integration Ontology (OIO) establishes the main relationships between documents (annotated and non-annotated), units, attributes and values both in the linguistic and in the ontological areas of annotation.

The application of these six ontologies in the OntoTag annotation model is twofold: first, as discussed above, they identify the different elements (mostly linguistic, but also ontological) that can be annotated in the Semantic Web field; second, once the ontology has been populated (instantiated) by the annotations obtained with OntoTag, the ontologies will also act as a repository or database of these annotations.

To conclude, we could say that, due to the extensibility and flexibility capabilities of the Linguistic Ontologies presented here, the OntoTag model of annotation inherits these properties as well. This model can also be considered as domain independent in the sense that the source ontologies can be replaced and, still, meaningful annotations would be obtained. As the design of the different Linguistic Ontologies follows (and broadens) the EAGLES guidelines, which are of a multilingual nature, OntoTag becomes also applicable to the annotation of the languages studied in these guidelines. The consensual nature of the ontologies and the sources used in their construction (EAGLES 1996a, 1996b; CES 1999; MILE 2003; GDA 2002; Dubuc & Lauriston 1997; Faber & Tercedor 2000; Melcuk 1996, 1988; Pustejovsky 1995) enables them (and the annotations obtained with them) to be considered standardised.

This research has partly been supported by the ministry of Science and Technology grant (Reference TIC2001-2745 CONTENTWEB project) and by the UPM

grant (Reference 14286 PLAN-H-SEMWEB project) References

XML/RDF(S)/OWL Semantic Web Page Annotation in ContentWeb" in Proceedings of the 2nd Workshop on NLP and XML (NLPXML-2003) - Language Technology and the Semantic Web, pp. 25-32. 10th Conference of the European Chapter of the Association for Computational Linguistics. EACL'03. Budapest, Hungary.

• Aguado de Cea, G., Alvarez de Mon, I., Pareja-Lora, A. 2003a. "Primeras aproximaciones a la anotacion lingüistico-ontolögica de documentos de Web Semântica: OntoTag" in Revista Iberoamericana de Inteligencia Artificial, Vol 1, pp 37-49.

• Alvarez-de- Mon Rego, I. 2003. La cohesiön del texto cientifico-técnico: un estudio contrastivo inglés-espanol. Universidad Complutense de Madrid.

• Benjamins, V.R., Fensel, D., Decker, S., Gomez-Pérez, A. 1999. (KA): Building Ontologies for the Internet: a Mid Term Report. IJHCS, International Journal of Human Computer Studies, 51, pp. 687-712.

• Berners-Lee, T., Fischetti, M. 1999. Weaving the Web: The Original Design and Ultimate Destiny of the World Wide Web by its Inventor. Harper. San Francisco.

• Borst, W. N. 1997. Construction of Engineering Ontologies. PhD thesis, University of Twente, Enschede.

• Bouillon, P. & Viegas, E. 1999. The Description of Adjectives for Natural Language Processing : Theoretical and Applied Perspectives. TALN 1999.

• CES. 1999. Corpus Encoding Standard. http://www.cs.vassar.edu/CES/

• Dubuc, R. and Lauriston, A. 1997. "Terms and Contexts" in Wright, S.E. and G. Budin, Handbook of Terminology Management Vol 1, Amsterdam/Philadelphia: John Benjamins, pp. 80-87.

• EAGLES. 1996a. EAGLES: Recommendations for the Morphosyntactic Annotation of Corpora. EAGLES Document EAG--TCWG—MAC/R.

• EAGLES. 1996b. EAGLES: Recommendations for the Syntactic Annotation of Corpora. EAGLES Document

EAG--TCWG—SASG/1.8.