1. Adam Pease and Christiane Fellbaum Presenter: 吳怡安
Formal ontology as interlingua: the SUMO and WordNet linking project and global WordNet
Adam Pease and Christiane Fellbaum
Presenter: 吳怡安

2. Outline
2.1 WordNet
2.1.1 Types and instances
2.1.2 Formal vs linguistic relations
2.1.3 Lexical vs conceptual ontologies
2.1.4 SUMO
2.2 Principles of construction of formal ontologies and lexicons
2.3 Mappings
2.4 Interpreting language
2.5 Global WordNet
2.5.1 The Interlingual Index
2.6 SUMO translation templates

3. 2.1 WordNet WordNet is a large lexical database for English
Basic unit: synset (cognitively equivalent synonyms)
Eg. {vacation, holiday}
Definition: leisure time away from work devoted to rest or pleasure
Sentence:"we get two weeks of vacation every summer”
synsets
Relations
Super-subordinate relation (also called hyperonymy, hyponymy)
Transitive: armchair-chair-furniture
Part-whole relation (Meronymy, holonym)
verbs towards the bottom of the trees express increasingly specific manners characterizing an event (communicate-talk-whisper)

4. 2.1.1 Types and instances
WordNet distinguishes among Types (common nouns) and Instances (specific persons, countries and geographic entities).
armchair is a type of chair
Barack Obama is an instance of a president
Instances are always leaf (terminal) nodes in their hierarchies.
However, other distinctions are lacking.
Brother and architect as Types of persons
Dwarf is a kind of person
One can refer to the same person as both a dwarf and an architect
Roles often refer to professions or functions associated with a person or temporary states (such as patient)

5. 2.1.2 Formal vs linguistic relations
One could extend the a few relations of WordNet to all the hundreds of relations that are found in a formally specified logical theory like SUMO.
Eg. part-of, beforeOrEqual, authors etc.
Relating informal linguistic notions with more formal ontological relations
Specifying complex relations that cannot be captured explicitly as simple links

6. 2.1.3 Lexical vs conceptual ontologies
Lexicon can be defined as the mappings of concepts onto words.
There are structural gaps where the geometry of the relations would require a word, yet where the language does not have one
the class of wheeled vehicles (like cars and motorbikes) vs vehicles that run on rails (trains, trams)
Crosslinguistic differences in lexicalization patterns
kinship relations
Existing words do not fully reflect the inventory of concepts that is available, so one can use the non-lexical ontology such as SUMO

7. 2.1.4 SUMO (1)
The Suggested Upper Merged Ontology (SUMO) is a formal ontology stated in a first-order logical language called SUO-KIF.
Upper ontology includes general notions in common-sense reality
time, spatial relations, physical objects, events and processes
Mid-Level Ontology (MILO): more specific
Domain ontologies cover over a dozen specific areas
world government, finance and economics, and biological viruses

8. 2.1.4 SUMO (2) 1,000 terms and 4,000 axioms (which includes 750 rules)
Term: named concept
Axiom: any statement in logic
Rule: a particular kind of axiom that has two parts: an antecedent and a consequent
(<=> (earlier ?INTERVAL1 ?INTERVAL2)
(before
(EndFn ?INTERVAL1)
(BeginFn ?INTERVAL2)))

9. 2.1.4 SUMO (3)
In a formal ontology, it is solely the axioms as mathematical statements that give the terms their meaning.
The meaning of the terms can be tested for consistency automatically with an automated theorem prover.
The names of terms in SUMO are just convenient labels, so there are no synonymy.

10. 2.2 Principles of construction of formal ontologies and lexicons
Lexicon must accurately reflect the inventory and use of words in a given language.
Neither eliminated nor ‘missing’ words
By contrast, an ontology is an engineered product.
Naming, categorization, creation of the terms are at will
Every lexicalized concept should be covered by a term.
Synonymy is not needed
The names of the terms could be replaced by arbitrary unique character strings and their meaning would still be the same.
In a formal ontology, meaning is determined only by the formal axioms.

11. 2.3 Mappings (1) Two phases:
First, mapping just SUMO itself to WordNet.
Three types of mappings were employed: rough equivalence, subsuming, and instance.
Equivalence: { artificial_satellite } to ArtificialSatellite
Subsuming: { elk } to HoofedMammal
Instance: { george_washington } to Human
Second, for each synset that occurred three or more times in SemCor, we also created a new concept in the MILO if one did not already exist in SUMO.

12. 2.3 Mappings (2)
Limitation: a single mapping from a lexical entity to a formal term does not fully capture the meaning of some lexical items
Eg. { Continue } can refer to many unrelated types of Process
referenced in the context of previous sentences
One would need a more complex relation structure to express the semantics of this lexical item

13. 2.4 Interpreting language
Deep semantic interpretation of language
Eg. ‘Brutus stabbed Caesar with a knife on Tuesday.’
(exists (?S ?K ?T)
(and
(instance ?S Poking)
(instance ?K Knife)
(instance ?T Tuesday)
(agent ?S Brutus)
(patient ?S Caesar)
(time ?S ?T)
(instrument ?S ?K)))
(=>
(instance ?X Knife)
(capability Cutting ?X instrument))

14. 2.5 Global WordNet
Vossen (1998) coordinated the effort to create eight European wordnets that follow a common design and are interlinked via an Interlingual Index (ILI)
Wordnets exist in over 40 languages spoken around the world at present.
Interconnected wordnets hold great potential for crosslinguistic applications.
Shedding light on commonalities and the differences in the ways languages map concepts onto words

15. 2.5.1 The Interlingual Index
there are language-specific ‘lexical gaps’
there are differences in the ways languages structure their words and concepts
The ILI initially consisted of all English WordNet synsets. Each international wordnet either links its synsets to the matching synsets in the ILI or adds a synset that is not yet in the ILI. Each synset in the language-specific wordnet has at least one equivalence relation to an entry in the ILI.
Due to the limitations of lexical ontologies, many wordnets have also been linked to SUMO.

16. 2.6 SUMO translation templates
In order to make SUMO more understandable, they created a system that performs rough natural-language paraphrasing of the formal axioms
(<=>
(earlier ?INTERVAL1 ?INTERVAL2)
(before
(EndFn ?INTERVAL1)
(BeginFn ?INTERVAL2)))
A time interval happens earlier than another time interval if and only if the end of the time interval happens before the beginning of the other time interval.

17. Thank you!