The second year review considered Deliverable 4.2 and Deliverable 4.3 insufficient and they were not approved by the reviewers in their current status. The objectives of WP4 are, as stated in the DoW? :
(i) research and development of two-way grammar-ontology interoperability bridging the gap between natural language and formal knowledge; (ii) infrastructure for knowledge modeling, semantic indexing and retrieval; (iii) modeling and alignment of structured data sources; (iv) alignment of ontologies with the grammar derived models.
D4.2 should contain a report on the Data Models, Alignment Methodology, Tools and Documentation. More specifically, it should contain information about the aligned semantic models and instance bases. While D4.2. contains information about Reason-able views and the key principles constituting these views are stated in the document, it does not state how these key principles have been implemented in the MOLTO-project. D4.2 does not comply with the key principle stating “Clean up, post-process and enrich the datasets if necessary, and do this in a clearly documented and automated manner.” D4.2 should contain exactly all details about the automation process of multiple ontologies. so that this knowledge and technique can be re-used to integrate new ontologies with the existing ones.
D4.3. should clear out the issue of the two-way interoperability between ontologies and GF grammars. This is still unclear, although objective (i) of WP4 is clear that this is a research-intensive part of MOLTO. Based on the WP4 presentation given in the review, this process requires the manual writing of mapping rules (NL Query -> GF, GF-> SPARQL query), which means limited potential for further re-use. The partners must clear the degree of automation that can be performed. What is required for porting this to a new application? Concrete steps should be provided making clear what can be automated and what cannot with the provided infrastructure. Details about mapping rule induction etc. should be provided.
As for the ontology/grammar mappings, here is what we have concretely got so far:
The examples show that the owl to GF mapping need not be difficult in any given case. What seems open is how to generalize these examples for the general case of generating a mapping for a new domain. In particular, we want a solution that allows the reuse of ontology to GF mappings to create more complex grammars from existing parts. The modularity of both OWL and GF suggest ways of approaching this goal.
One approach to a more general solution is to use the term ontologies developed in TermFactory? to also store parts of mappings needed for GF verbalization. In a TermFactory? term ontology, a term is a pair of a general language expression and a special language concept. In this approach, an ontology concept would map to an abstract grammar term. Individual language expressions and terms associated with the concept map to concrete grammar terms. A term or expression would inherit GF grammar properties from classes to which it belongs (say, exp:Noun). Grammatical properties common to all uses of a given general language expression would be stored as properties of the expression. GF terms or grammatical properties that are specific to a domain GF grammar would stored as properties of a domain specific term.
Instead of having to define a new grammar and create concept to grammar associations from scratch, a grammar would be compiled from appropriate choices of resource from the term ontology plus a language and/or domain specific syntactic base. To extend a vocabulary, we add a new term (expression, concept) instance, typed in the appropriate categories, and add to it any further GF properties that are relevant to its correct linearization. The concrete expression associated to a compositional abstract grammar term need not be specified in the ontology, if it can be compositionally derived from the GF abstract syntax associated to the concept and other resources in the ontology. The above does not claim to do more than propose a way to decompose the ontology to grammar mapping into reusable parts.
If the approach seems useful, UHEL is prepared to invest effort to building a test case using the museum case as a starting point.