A Controlled Natural Language Interface for Semantic MediaWiki Jie Bao Rensselaer Polytechnic Institute Paul R. Smart, Nigel R. Shadbolt University of Southampton Dave Braines, Gareth Jones IBM UK ACITA 2009, Sep 23, 2009 University of Maryland

Motivation The timely collection of facts obtained at the tactical (and even the interpersonal level) may help to create better operational decisions at all levels. The goal is to create networks of soldiers who are capable of collecting information within their sphere of influence and who can share this information with other members of the net in a timely fashion. ---Defense Science Board Summer Study on 21st Century Strategic Technology Vectors, Volume II, Critical Capabilities and Enabling Technologies. Office of the Under Secretary of Defense for Acquisition, Technology, and Logistics, Washington, DC,

Motivation 3

Key Requirements Collaboration Support –Can be used by a variety of different types of contributors –Supports cross-platform access Ease of Use –Requires least training –Provides simple user interface –Content ease to understand Meaningful Structure –Enables some automated processing of data (e.g., automated data propagation) –Enables discovery of hidden knowledge from explicitly known information. 4

Enabling Technologies 5 Semantic Wiki Controlled Natural Language Every A is B … +

Semantic Wiki Semantic Wiki Multi-user content creation and editing Browser-based, Cross-platform easy to use Support semantic annotations for automated processing and inference

Semantic MediaWiki (SMW) It is the most popular semantic wiki system extending MediaWiki (the Wikipedias platform) 7 Mediawiki: What you edit what you see

Semantic MediaWiki 8 SMW: What you edit (Modeling Script) what you see typed link (property) To author knowledge

Semantic MediaWiki 9 SMW: What you edit (Querying Script) what you see To retrieve knowledge

Challenges However, SMW is limited by Expressivity constraints No support for the full range of RDF/OWL E.g., we cant specify that capital of property should always link a City to a Country (property domain and range) Usability Thinking in the subject-property-object fashion may not be ease (as one may think!) Based on our own experiments with entry-level users. Knowledge engineers may find it difficult to understand others contributions. Users may prefer different languages styles or even different natural languages.

Challenges However, SMW is limited by Expressivity constraints No support for the full range of RDF/OWL E.g., we cant specify that capital of property should always link a City to a Country (property domain and range) Usability Thinking in the subject-property-object fashion may not be ease (as you may think!) Based on our own experiments with entry-level users. Knowledge engineers may find it difficult to understand others contributions. Users may prefer different languages styles or even different natural languages. By extending SMW with an OWL Meta-model (introduced in details later) By allowing modeling/representing knowledge using controlled natural language (CNL)

Controlled Natural Language CNL Every A is B… Production of knowledge without using a formal logic. Potential production and comprehension benefits Multiple OWL-compliant CNLs are available: e.g. Rabbit, Sydney OWL Syntax, ACE-OWL

Controlled Natural Language Example (In Rabbit syntax): The capital of" relationship can only have a City" as a subject. The "capital of" relationship can only have a Country" as an object. 13 Instead of saying (In OWL Abstract Syntax) ObjectProperty(ex:capital_of domain(ex:City) range(ex:Country)) Or (In Description Logics) capital_of.T City capital_of -.T Country

Solution Summary Collaboration Support –Natively collaborative for a wiki –Web-based interface accessible from any device that has a browser. Ease of Use –Simplify knowledge input using semi-automatically generated forms –Multiple OWL-compliant CNLs: e.g. Rabbit (English, Chinese), ACE-OWL Meaningful Structure –Meta-model do the translation between OWL and CNL, and between OWL and SMW scripts –Query and inference supported. 14

Architecture Prototype system:

Class(Rabbit partial intersectionOf(animal restriction(eat someValuesFrom(FreshVegetable))) OWL: Rabbit eats some fresh vegetable SMW OWL Meta Model Us wiki templates to create OWL meta-model extensions for SMW Form-based editing interface associated with templates 132

OWL Class Templates 17 Similarly, we have templates for properties and individuals

CNL Generation {{#vardefine:label|{{CNL.getLabel|{{{1}}} }} }} {{#vardefine:super | {{#ask: [[:{{{1|{{FULLPAGENAME}}}}}]] |?Category= |mainlabel=-|format=list|link=none }} }} {{#if: {{#var:super}} |{{#arraymap:{{#var:super}}|,|xxx| Every [[:{{{1}}}{{!}}{{#var:label}}]] is a kind of [[:xxx|{{CNL.getLabel|xxx}}]] }}|}} Use SMW queries to retrieve knowledge statements in the OWL meta model Construct sentences according to the syntax of the target CNL

Multiple CNL Support Accommodating new CNLs requires relatively minor changes to the wiki script

I/O Support Allows importing of an RDF/OWL ontology Exporting in three CNL syntaxes Ontology in the meta-model can be exported in RDF. Some limitations apply –Limitation supports for blank nodes –Datatypes –Ontology management (e.g. imports) 20

Related work AceWiki Differences: underlying wiki system editing interface light-weight extensions support for multiple CNLs customization of target CNLs 21 Tobias Kuhn (2008)

Related work ROO 22 Vania Dimitrova et al Differences: Protégé-based vs. wiki- based Rabbit only vs. multiple CNLs Formal modeling only vs. formal+informal modeling

Summary We have developed an OWL meta-model extension to SMW to support the representation of OWL ontologies We have provided a light-weight form-based interface to support ontology editing We have provided an RDF import mechanism to support the import of existing ontologies We have developed multiple CNL verbalizers to support the serialization of semantic wiki content to CNLs Future work: –Improved expressvity –Enable users to create/customize CNL output –Implement wiki-based CNL editing capability 23