Title: Chinese Characters andTop Ontology in EuroWordNetPaper by: Shun Sylvia Wong &Karel PalaPresentation By: Patrick BakerIntroduction WordNet, Cyc, HowNet, and EuroWordNet eachuse a hierarchical structure of languageindependent concepts to reflect the importantsemantic differences between concepts EuroWordNet uses a hierarchy called TopOntology (TO) This paper compares EuroWordNet’s TO withthe natural organization found in thepictographic based Chinese languageTop Ontology? Ontologies are artificial constructs builtwith the primary purpose to serve as thelexical databases for knowledgerepresentation systems Top Ontology distinguishes betweenthree types of entities This paper focuses on the third typeThe Three Entity Types of TO: There are three types of entities distinguished atthe first level of TO:1. 1st Order – any concrete entity publicly perceivable bythe senses and located at any point in time, in a three-dimensional space (persons, animals, discrete objects)2. 2nd Order – any Static Situation (property, relation) orDynamic Situation, which cannot be grasped, heard,seen, felt as an independent thing (events, processes,states-of-affair)3. 3rd Order – unobservable propositions which existindependently of time and space. They can be true orfalse rather than real (ideas, thoughts, theories, plans,reasons)The Chinese Language Chinese script originated from picture-writing Only a couple hundred characters in thelanguage are actual pictograms According to the etymological dictionarywritten by Xu Shen around 100 A.D.,Chinese characters can be divided into sixgroupsSix Groups of Chinese Characters1. Pictographs (≈4%): represent real-life objects bydrawings2. Ideographs (≈1%): represent positional and numeralconcepts by indication3. Logical Aggregates (≈13%): form a new meaning bycombining the meanings of two or more characters4. Phonetic Complexes (≈82%): form a character bycombining the meaning of one character and anothercharacter which links through a shared sound5. Associative Transformations (a small portion): extendthe meaning of a character by adding more parts to theexisting one6. Borrowings (a small portion): to borrow the writtenform of a character with the same soundThe Chinese Language The average educated Chinese personknows only about 6000 of the 50,000characters in the Chinese language Since many of the characters arecombinations of simpler characters,knowing the meaning of one or more ofthe constituent characters allowsdeduction of the overall meaningThe Chinese Language Because Chinese characters can not beordered alphabetically in a dictionary, theyare ordered by Section Heads or ChineseRadicals There are 213 Chinese Radicals In most cases, a character is groupedunder a certain Chinese Radical if itsconcept relates to the concept representedby the radical in some wayThe Chinese Language and 3rd OrderEntities The concepts in the 3rd Order Entity list areabstract and difficult to grasp; most arerepresented by use in the form of a sentence(e.g. “John thought the movie was good”) Wong & Pala (2001) have shown that no directcorrespondence can be found between ChineseRadicals and the concepts in the 3rd Order list In most cases, the Chinese counterparts ofthese concepts are represented by morecomplicated lists of charactersThe Chinese Language and 3rd OrderEntities For each of the basic concepts in the 3rd Orderlist, the authors located their Chinesecounterparts Each concept created a list of Chinesecharacters representing synonyms, hyperonyms,and/or meanings that collectively defined thescope of the concept The meanings of the component radicals ofeach character in the list were then examinedThe Chinese Language and 3rd OrderEntities The authors found that certain radicals (withspecific meanings) were found associated withone or two 3rd Order concepts This association is called Sense Transfer e.g. the characters (logic/reason/theory),(opinion/theory/discussion), and (theory/toexplain/to say) appear more often under theory e.g. the characters (to think/to consider) and (to think/to contemplate) appear more oftenunder idea/thoughtSense Transfer and Other Languages Sense transfer exists in most languages, thoughnot necessarily to the extent as pictograph basedlanguages English examples: care-free, side-light, un-think-able Czech example: u_-i-t-el (a root denoting theconcept ‘teach’ + a verb-making affix + an infinitiveaffix + an agentives suffix = teacher) The inadequacy of existing ontologies to show thissense transfer property means there exists no wayto derive the meaning for a new word even if itscomponents already exist in the ontologyThe Chinese Way to Represent Concepts Wong & Pala (2001) have observed thatChinese seems to organize concepts in acontextual manner, with each Chineseradical serving as the characterizing basicconcept in the respective concept Through observation, the authorsdetermined that many of the characterssubsumed in the radicals can be classifiedalong five main linesThe Chinese Way to Represent Concepts The five conceptual lines are:1. As an object2. As a property3. As a typical event (situation, process)4. It’s component5. As a consequence• e.g. the character (fire) ‘as an object’is part of (stove) and (charcoal),and ‘as a typical event’ is part of (toburn) and (to cremate)Lexical/Conceptual Organization The Chinese way of organizing concepts (evenabstract ones) from simpler, more concreteconcepts/entities provides an alternative to theorganization provided by existing ontologies Such an organization would form a semanticnetwork as opposed to the tree structure foundin such ontologies Such a semantic network is richer, morecomplete, and more transparent, as eachconcept is derived not from verbalized concepts,but a semantic context of discrete entitiesConclusion By comparing EuroWordNet’s TO to theintrinsic structure provided by the naturallanguage Chinese, it can be seen that:Humans more naturally think of concepts asbeing composed of more concrete entities, asopposed to derived from abstract conceptsThe more natural way to represent suchconcepts is a semantic graph, not the treestructure found in most existing