AceWiki: Collaborative Ontology Management in Controlled Natural Language

AceWiki: Collab orativ e On tology Managemen t in Con trolled Natural Language T obias Kuhn Departmen t of Informatics, Universit y of Zurich, Switzerland tkuhn@ifi.uzh.ch http://www.ifi.uzh.ch/cl/tkuhn Abstract. AceWiki is a protot yp e that sho ws ho w a semantic wiki using con trolled natural language — Attempto Controlled English (A CE) in our case — can mak e ontology managemen t easy for ev eryb ody . Sen- tences in A CE can automatically b e translated in to first-order logic, O WL, or SWRL. AceWiki integrates the OWL reasoner Pellet and en- sures that the ontology is alw ays consisten t. Previous results hav e shown that p eople with no bac kground in logic are able to add formal kno wledge to AceWiki without b eing instructed or trained in adv ance. 1 In tro duction Since ontologies are often defined within communities, semantic wikis could b e used for their collab orative creation and managemen t. Unfortunately , most of the existing semantic wikis do not supp ort expressiv e ontology languages in a general wa y . They do not allow the users to add complex axioms like “every landlo c k ed country b orders no sea”. F urthermore, the existing semantic wikis are often hard to understand for p eople who are not familiar with the technical terms of logic and on tologies. AceWiki 1 tries to solve b oth problems by using controlled natural language. Ordinary p eople who hav e no bac kground in logic should b e able to understand, mo dify , and extend the formal conten t of a wiki. Man y existing semantic wikis are classical wikis enriched with semantic an- notations. The goal is not to manage stand-alone on tologies, but rather to give some kind of formal bac kb one to the wiki articles. W e follo w a differen t approac h — similar e.g. to the myOntolo gy pro ject [7] — b y providing a wiki that is dedi- cated to building and maintaining ontologies. In con trast to myOn tology , we do not restrict ourselves to light weigh t (i.e. relatively inexpressive) ontologies. The use of controlled natural language allo ws us to express also complex axioms in a natural w ay . Figure 1 shows a screenshot of the AceWiki interface. In our usage scenario, a comm unity of domain exp erts uses AceWiki to cre- ate and maintain a formal knowledge base in a collaborative manner. There are t wo exemplary wiki instances — one about geography and the other ab out pro- tein interactions — that demonstrate ho w AceWiki could b e used to represent kno wledge of such communities. 1 See [6] and http://attempto.ifi.uzh.ch/acewiki Fig. 1. The w eb interface of the AceWiki prototype AceWiki has b een introduced in [6]. Since then, several new features ha ve b een added, for example the integration of a reasoner and the supp ort for num ber restrictions (“at most 3”, “exactly 5”, etc.). 2 A ttempto Con trolled English A ttempto Con trolled English (ACE) 2 is the controlled natural language that is used for AceWiki. ACE app ears completely natural since it is a subset of English. Restrictions of the syntax and the definition of a small set of in terpretation rules mak e it a formal language that is automatically translatable into first-order logic. A CE supp orts a wide range of natural language constructs: singular and plural noun phrases, active and passive v oice, relative phrases, anaphoric references, existen tial and univ ersal quantifiers, negation, mo dalit y , and more. In the past, A CE has successfully b een applied for different tasks in different research areas, for example as a query language for on tologies [1], as a kno wledge representation language for the biomedical domain [4], and as a rule language for a multi- seman tics rule engine [5]. F urthermore, ACE has b een used as a natural language front-end to O WL with a bidirectional mapping of ACE to OWL [3]. This mapping co vers all of OWL 2 except data prop erties and some very complex class descriptions. AceWiki relies on this work for translating ACE sentences into O WL, whic h allo ws us then to do reasoning with existing OWL reasoners. 2 See [2] and http://attempto.ifi.uzh.ch Fig. 2. The predictiv e editor of AceWiki 3 Design and Ev aluation The goal of AceWiki is to show that semantic wikis can b e more natural and at the same time more expressiv e than existing semantic wikis. Naturalness is ac hieved b y representing the formal statements in ACE. Since A CE is a subset of natural English, every English sp eak er can immediately read and understand the conten t of the wiki. In order to enable easy creation of A CE sentences, AceWiki provides a predictive editor that shows step-by-step the words that are syn tactically possible at a given p osition in the sen tence. Figure 2 shows a screenshot of the predictiv e editor of AceWiki. F urthermore, the AceWiki in terface do es not use technical terms like “ontological elemen t”, “prop ert y”, or “sub class” but uses instead terms like “word”, “transitive verb”, or “hierarch y” which should b e muc h more familiar to p eople with no background in logic. AceWiki mak es use of the high expressivit y of ACE that go es b ey ond OWL and SWRL. W e do not lik e the idea of cutting down the expressivity just for the sake of reasoning p erformance. Even if some statemen ts b ecome so complex that it is almost imp ossible to do reasoning with them, it is b etter to hav e them formalized than just left out. W e do not lose an ything, since w e are free to ignore those complex statemen ts for certain reasoning tasks. In our previous w ork [6], w e conducted a user experiment that prov ed that ordinary p eople with no bac kground in logic are able to deal with AceWiki. The participan ts — without being instructed how to interact with the in terface — w ere ask ed to add kno wledge to AceWiki. About 80% of the created sentences w ere correct and sensible. This is remark able since most of the sentences were quite complex: more than 60% of them con tained an implication or a negation or b oth. Using the predictive editor which the participants had never seen b efore, they needed on a verage only five minutes to create their first correct sen tence. 4 Reasoning in AceWiki W e ha v e started to in tegrate the OWL reasoner P ellet 3 in to AceWiki. Since A CE sen tences can b e b ey ond the expressivity of OWL, the reasoner cannot consider all sen tences. In order to make this clear to the users, each sentence is tagged as blue (inside of O WL) or red (outside of OWL): In this wa y , it is easy to explain to the users that only the blue statements are considered when the reasoner is used. W e plan to provide an in terface that allows skilled users to exp ort the formal con tent of the wiki and to use it within an external reasoner or rule-engine. Thus, even though the red statements cannot b e interpreted by the built-in reasoner they can still b e useful. Consistency chec king plays a crucial role because an y other reasoning task re- quires a consisten t ontology in order to return useful results. Most other semantic wikis do not ha ve this problem since their languages are simply not expressive enough to ev er run into inconsistency . In order to ensure that the on tology is alwa ys consistent, AceWiki chec ks ev ery new sentence — immediately after its creation — whether it is consis- ten t with the current ontology . Otherwise, the sentence is not included in the on tology: After the user created the last sentence of this example, AceWiki detected that it con tradicts the current ontology . The sentence is included in the wiki article but the red font indicates that it is not included in the ontology . The user can remo ve this sentence again, or keep it and try to reassert it later when the rest of the on tology has changed. F or this approach, it is very imp ortan t to p erform incremental reasoning whic h P ellet supp orts only partially at the moment. F or that reason, AceWiki do es not scale very well. W e exp ect that future reasoners will b e able to run m uch faster in such incremental scenarios. 3 http://pellet.owldl.com/ Not only asserted but also inferred kno wledge can be represented in A CE. A t the moment, AceWiki can show inferred class hierarchies and class memberships. F urthermore, we are working on a query feature for AceWiki. Questions will b e form ulated in ACE and ev aluated by the reasoner: Th us, ACE can b e used not only as an on tology- and rule-language, but also as a query-language. 5 Conclusions The AceWiki protot yp e sho ws how ontologies can b e managed in a natural w ay within a wiki. It demonstrates how semantic wikis using controlled natural language can b e expressive and easy to use at the same time. Our previous ev al- uation show ed that AceWiki is indeed easy to learn. W e explained how AceWiki ensures — in a very simple wa y — the consistency of the ontology whic h is the basis for other in tegrated reasoning services. References 1. Abraham Bernstein, Esther Kaufmann, Norb ert E. F uchs, June v on Bonin. T alking to the Seman tic W eb — A Con trolled English Query In terface for On tologies. Pr o c. 14th Workshop on Information T e chnolo gy and Systems , 2004 2. Norb ert E. F uc hs, Kaarel Kaljurand, Gerold Schneider. Attempto Controlled English Meets the Challenges of Knowledge Representation, Reasoning, Inter- op erabilit y and User Interfaces. Pr o c. 19th International FLAIRS Confer enc e (FLAIRS’2006) , 2006 3. Kaarel Kaljurand. Attempto Contr ol le d English as a Semantic Web L anguage . PhD thesis, F aculty of Mathematics and Computer Science, Universit y of T artu, 2007 4. T obias Kuhn, Lo ¨ ıc Roy er, Norb ert E. F uchs, Michael Schroeder. Improving T ext Mining with Con trolled Natural Language: A Case Study for Protein In teractions. Pr o c. Thir d International Workshop on Data Inte gr ation in the Life Sciences 2006 (DILS’06) , Springer, 2006 5. T obias Kuhn. AceRules: Executing Rules in Controlled Natural Language. Pr o c. First International Confer enc e on Web R e asoning and Rule Systems (RR2007) , Springer, 2007 6. T obias Kuhn. AceWiki: A Natural and Expressive Semantic Wiki. Pr o c. of Seman- tic Web User Interaction at CHI 2008: Exploring HCI Chal lenges , CEUR W ork- shop Pro ceedings, 2008 7. Katharina Siorpaes, Martin Hepp. myOn tology: The Marriage of On tology Engi- neering and Collective Intelligence. Pr o c. Bridging the Gap b etwe en Semantic Web and Web 2.0 , 2007