ONTOMERGE Ontology translations by merging ontologies Paper Ontology

ONTOMERGE Ontology translations by merging ontologies Paper: Ontology Translation on the Semantic Web by Dejing Dou, Drew Mc. Dermott and Peishen Qi 2003 http: //cs-www. cs. yale. edu/homes/dvm/daml/ontology-translation. html presented by Laurentiu Vasiliu 5/19/2021 1

Introduction • Problem: ontology translation is required when: - translating datasets - generating ontology extensions - querying through different ontologies • Approach: ontology translation by ontology merging and automated reasoning. • Onto. Merge: online system that implements the previous approach http: //onto. cs. yale. edu: 4040/onto. Merge. html 5/19/2021 2

Introduction • Onto. Merge: implement ontology translation with inputs and outputs in DAML+OIL or other web languages. • Approach focus: formal inference from facts expressed in one ontology to facts expressed in another ontology. • The merge of 2 ontologies: taking the union of the terms and the axioms that define them. • XML namespaces are used to avoid name clashes. • Bridging axioms are added, to relate the concepts between the 2 ontologies. • Bridging axioms = description of mappings between ontologies. 5/19/2021 3

Introduction • Devising and maintaining a merged ontology must involve the contribution of human experts. • Inference mechanism: a theorem prover optimised for ontology translation task. • Inference used for: - dataset translation - ontology extension generation - query handling through ontologies 5/19/2021 4

Ontology translation problems • Ontologies differences: syntactic & semantic • Syntactic & semantic translation needed. • Problem 1: agents: web agents try to exchange datasets but they use different ontologies to describe them. • Ontology translation for datasets: translation of a dataset from one ontology to another. • Dataset: set of facts expressed in a particular ontology. 5/19/2021 5

Ontology translation problems • Problem 2: ontologies generation: translation required when generating ontologies extensions (sub-ontologies) • Generating ontologies: given O 1 and O 2 – two related ontologies and an extension O 1 s of O 1 construct the corresponding extension of O 2 s. • Ontology experts are developing subontologies manually • Tedious work: tools needed 5/19/2021 6

Ontology translation problems • Problem 3: Query: knowledge to be used to answer query may be in multiple knowledge bases • These knowledge bases may use different ontologies than the querying agent. • Without ontology translation, querying is very difficult 5/19/2021 7

Ontology mapping vs. translation ontology • They are different. • Ontology mapping: find correspondence (mapping) between concepts of 2 ontologies. • Mapping are expressed by mapping rules – how the concepts correspond. • Ontology translation: needs to know the mapping first; then can use the mapping rules. • Mappings can be generated by experts or generated automated. • In this paper’ implementation the mapping are performed manual, by experts. • Automation is an active area of research. 5/19/2021 8

Previous work on ontology translations for datasets • Two strategies used: • 1: Translate a dataset in any source ontology in one big centralised ontology; from it can be translated into a dataset in any target ontology – [Ontolingua] • It can’t really work unless a global ontology can cover all the existing ontologies + agreement by all ontology experts to write translators between their own ontology and the global ontology • Daily maintenance of all ontologies consistent with the One Truth Theory is very difficult 5/19/2021 9

Previous work on ontology translations for datasets • 2: perform ontology translation directly from a dataset in a source ontology to a dataset in another target ontology. [Ontomorph] • For practical purposes is very useful • It relies on special properties of the dataset to be translated. • Does not address the question of producing a general-purpose translator. • Previous work on ontology translation for query handling is closely related to database mediators. 5/19/2021 10

New approach: separate syntactic and semantic translation • Break ontology translation into 3 parts: - syntactic translation from the web language to internal representation - semantic translation by inference using the internal notation - syntactic translation from the internal representation to the target web language 5/19/2021 11

New approach: separate syntactic and semantic translation • Syntactic issues are managed in the 1 st and 3 with a (extendable) translator: PDDAML for translating between the internal representation and DAML+OIL • Internal representation language: Web-PDDL • Web-PDDL – first order language with LISP like syntax • It extends Planning Domain Definition Language (PDDL) with XML namespaces and more flexible notations for axioms. 5/19/2021 12

New approach: Ontology merging and automated reasoning • Translating datasets: given a set of facts in one vocabulary (source), infer the largest possible set of consequences in another (the target); this can be broke in 2 phases: • Step 1: Inference: working in a merged ontology that combines all the symbols and axioms from both the source and target, draw inferences from source facts. • Merged ontologies: contain symbols, facts + bridging axioms 5/19/2021 13

New approach: Ontology merging and automated reasoning • Step 2: Projection: retain conclusions that are expressed only in the target vocabulary. • For the near future the merged ontologies has to be constructed by human experts. • For very large ontologies automated tools may give human suggestions to human experts • The merged ontology is an ontology itself that can be further merged with other ontologies. • Skolem terms are used: when translation requires talking about an object that cannot be identified with an existing object. 5/19/2021 14

New approach: Ontology merging and automated reasoning • Term generation functions are introduced • Allow finer control over term generation than skolemisation • For translation is used theorem proving • Concern: in general theorem provers can run for a long time and conclude nothing usefull • However, the inference needed to be made are focused on: 5/19/2021 15

New approach: Ontology merging and automated reasoning - Forward chaining from source to target - Backward chaining from queries in one ontology to datasets in another - Introduction of skolem terms and termgenerating functions. - Use of equalities to substitute existing constant terms for skolem terms • Here, theorem prover called Onto. Engine is specialised on these sorts of inference 5/19/2021 16

New approach: Ontology merging and automated reasoning • Onto. Engine uses chaining through implications with specified directions instead of full fledged resolutions • It is not complete in the logical sense • Trade: completeness vs. efficiency • Even a logically complete theorem prover would fail in general to achieve translation completeness because the source ontology and target ontology might not overlap. 5/19/2021 17

Onto. Merge architecture for translating datasets 5/19/2021 18

Conclusions • Ontology translation is one of the most difficult problems. • Ontology translation is required when translating datasets, generating ontologies extensions or querying through different ontologies. • Ontology translation can be thought in terms of ontology merging. • If all ontologies, datasets and queries can be expressed in terms of the same internal representation, semantic translation can be implemented by automatic reasoning. • The required reasoning can be thought as simple typed, first order inference. 5/19/2021 19