Richly interlinked, machine-understandable data constitute the basis for the Semantic Web and by extension the Social Semantic Desktop[2]. Manual semantic annotation is a complex and arduous task both time-consuming and costly often requiring specialist annotators. (Semi)-automatic annotation tools attempt to ease this process by detecting instances of classes within text and relationships between classes, however their usage often requires knowledge of Natural Language Processing(NLP) and/or formal ontological descriptions. This challenges researchers to develop user-friendly annotation environments within the knowledge acquisition process. Controlled Natural Languages (CNL)s o er an incentive to the novice user to annotate, while simultaneously authoring, his/her respective documents in a user-friendly manner,yet shielding him/her from the underlying complex knowledge representation formalisms. CNLs have already been successfully applied within the context of ontology authoring, yet very little research has focused on CNLs for semantic annotation. We describe a user friendly semantic annotator, based on Controlled Language for Information Extraction (CLIE) tools, which permits non-expert users to semi-automatically both author and annotate meeting minutes and status reports using controlled natural language.
\Controlled Natural Languages are subsets of natural language whose grammars
and dictionaries have been restricted in order to reduce or eliminate both
ambiguity and complexity.3" The use of CNLs for ontology authoring and population
is by no means a new concept and it has already evolved into quite an active
research area[
creation and semantic annotation, for instance the CLIE technology permits
ontology creation and population by mapping both concept de nitions and
instances of concepts to a ontological representation using CLOnE - Controlled
Language for Ontology Editing[
CNLs cannot o er a panacea for semi-automatic annotation since it is unrealistic to expect users to annotate every textual resource using CNL, however there are certain use-cases where CNLs can o er an attractive alternative as a means for semi-automatic semantic annotation, particularly in contexts, where controlled vocabulary or terminology is implicit such as health care patient records or business vocabulary. Our use case focuses on administrative tasks such taking minutes during a project team meeting and weekly status reports. Very often such note taking tasks can be repetitive and boring. In our scenario the user is a member of a research group which in turn is part of an integrated EU research project. Based on pre-de ned templates, the user simultaneously authors and annotates his/her meeting minutes or status reports in CNL, using a semantic note taking tool - SemNotes6, which is an application available for NepomukKDE7 - the KDE instance of the Social Semantic Desktop. The metadata is available for immediate use after creation for querying and aggregation, whereby 4 http://www.projecthalo.com/ 5 http://www.vulcan.com 6 http://smile.deri.ie/projects/semn 7 http://nepomuk.kde.org/ the retrieved RDF triples can be passed to a Natural Language Generator to produce tailored textual reports and summaries. 4
In our scenario, the CNL annotator is realised within a Semantic Note. The CNL is anchored to existing semi-structured data such as a AgendaTitle, Scribe or ActionItem based on prede ned meeting minutes or status report templates. The annotator is based on CLIE. The CNL itself is very similar to the CLOnE language with signi cant modi cations. The annotator architecture contains a standard GATE pipeline8(see Figure 1) which includes the following language processing resources: The GATE English tokeniser, the Hepple POS tagger, a morphological analyser, a gazetteer list component for recognising useful keyphrases, such as structured elements from the templates and reserved CNL phrases. Any sentences for example, preceded by a Comment: element are considered candidates for controlled language parsing. Any remaining tokens from the CNL sentence which are not recognised as reserved CNL key-phrases are used as names to generate links to ontological objects(See Figure 2). This is followed by a standard Named Entity(NE) transducer in order to recognise useful NEs, a preprocessing JAPE9 nite state transducer(FST) for identifying quoted strings, chunking Noun Phrases(NPs) and additional preprocessing. A second gazetteer list look up is applied which identi es trigger phrases associated with NEs which intersect with quoted and unquoted NP annotation spans. Additional feature values are then added to the NP chunks to indicate the appropriate class to link an NP chunk as an instance to. The last FST parses the CNL from the text and generates the metadata. The current tool is bootstrapped via the Nepomuk Core Ontologies10 and currently the application creates/populates a meeting minutes/status report ontology which references the users Personal Information Model Ontology(PIMO) 11, via the GATE Ontology API. We intend to modify the code to write directly to Nepomuk KDE RDF store.
Incentive for the user to annotate his/her respective documents plays an
important role for the realisation of both the Semantic Web and Social Semantic
Desktop. We have described a Semantic Annotator which allows non expert users
to simultaneously create content within, and add relational metadata to, notes
on the Semantic Desktop, using CNL. Furthermore, our annotator has already
been implemented and wrapped as a plugin for a semantic note taking tool.
Finally, we intend to complete the integration with Nepomuk KDE and
evaluate the user-friendliness of our annotator based on the previously successful
empirical methods employed in CLOnE [