Title: Sorting annotations to trace interactions.

Abstract:

Within a distributed workgroup, collective sensemaking (Weick) and awareness (Dourish and Belotti) are supported by exchanges through annotations to argue around documents and a tool is required to structure and to retrieve this information. Thus, we propose to support collaborative annotation activity by a tool allowing comments' anchoring to documents and comments' retrieval. Hereby, we focus on the use of NLP techniques to structure and to retrieve annotations. Indexation is time-consuming, so, we propose to support this task by using robust NLP techniques to identify indexing terms from the annotation's anchoring context, and some ontologies (domain-specific and argumentation) to index the annotation. The user finally selects the appropriate indexing terms.
NLP methods are used to build initial ontologies from reference corpora and to identify terms for indexing user's annotations. We chose Topic Maps (TM, Biezunski) to represent ontologies due to their advantages: portability, user-centered browsing and maintenance (user might add new terms), shared concept definition (available on URL), and faceted data representation (domain, arguing and annotator's roles). TM is built up by automatically extracted collocations (Rousselot) structured in a concept hierarchy. When indexing a new annotation, NLP methods identify possible candidates (domain and argument terms). These candidates are matched to the ontology concepts, based on a Latent Semantic Indexing algorithm (Berry) applied on annotation's context.
As a first case study, we use a corpus of e-mails of a collaborative project on mechanical aeronautic engineering to build an initial TM. A first version of this collaborative annotation tool is available implementing basics commenting and indexing annotations' features.