A Taxonomic Classification of WordNet Types

Abed Alhakim Freihat Fausto Giunchiglia Biswanath Dutta Qatar Computing Research Institute University of Trento Indian Statistical Institute (ISI) Doha, Qatar Trento, Italy Bangalore, India [email protected] [email protected] [email protected]

Abstract been given, so far, to the principles or rules used to identify polysemy types. In fact, none WordNet represents polysemous terms by of these approaches can explain how to identify capturing the different meanings of these the polysemy types of the discovered polysemy terms at the lexical level, but without giv- structural patterns or how to differentiate for ing emphasis on the polysemy types such example, between homonymy and metaphoric terms belong to. The state of the art pol- structural patterns. Although Apersejan’s seman- ysemy approaches identify several poly- tic similarity criterion (Apresjan, 1974) can be semy types in WordNet but they do not ex- used to account for regularity in polysemy, it plain how to classify and organize them. can not predict the polysemy type of the regular In this paper, we present a novel approach polysemy types in WordNet. Our hypotheses in for classifying the polysemy types which this paper is that identifying and differentiating exploits taxonomic principles which in between the polysemy types of the regular pol- turn, allow us to discover a set of poly- ysemy structural patterns requires understanding semy structural patterns. the hierarchical structure of WordNet and, thus, the criteria related to the taxonomic principles that 1 Introduction the hierarchical structure of WordNet comply with Polysemy in WordNet (Miller, 1995) corresponds or violates. In this paper, we show how to use two to various kinds of linguistic phenomena and taxonomic principles as criteria for identifying can be grouped into various polysemy types the polysemy types in WordNet. Based on these (Falkum, 2011). Although WordNet was inspired principles, we introduce a semi automatic method by psycholinguistic and semantic principles for discovering and identifying three polysemy (Miller et al., 1990), its conceptual puts types in WordNet. greater emphasis on the lexical level rather than on the semantic one (Dolan, 1994). Lexicalizing The paper is organized as follows. In Sec- polysemous terms without any further information tion two, we discuss the problem. In Section about their polysemy type affects the usability of three, we introduce the formal definitions we WordNet as a knowledge resource for semantic use. In Section four, we discuss the taxonomic applications (Mandala et al., 1999). principles that we use to discover three of the In general, the state of the art approaches suggests polysemy types in WordNet. In Section five, we different solutions to the polysemy problem. The give an overview of our approach. In Section six, most prosperous among these approaches are the we show how to use the taxonomic principles to regular/systematic polysemy approaches such as identify metaphoric structural patterns. In Section (Buitelaar, 1998) (Barque and Chaumartin, 2009) seven, we demonstrate how to determine special- (Veale, 2004) (Peters, 2004). These approaches ization polysemy structural patterns. In Section propose the semantic regularity as a basis for eight, we describe how to discover homonymy classification of the polysemy classes and offer structural patterns. In Section nine, we explain different solutions that commensurate the nature how to handle false positives in the structural of the discovered polysemy types. patterns. In Section ten, we present the results of Despite the diversity and depth of the state of our approach. In Section eleven, we conclude the the art solutions, no or very little attention has paper and depict our future work. 2 Problem Statement approaches do not reveal the principles or the cri- teria used to classify polysemous terms into poly- WordNet is a machine readable online lexical semy types. database for the English language. Based on psy- In this paper, we explain how to use the exclusive- cholinguistic principles, WordNet has been devel- ness property and the collectively exhaustiveness oping since 1985, by linguists and psycholinguists property (Bailey, 1994) (Marradi, 1990) for iden- as a conceptual dictionary rather than an alpha- tifying the following polysemy types. betic one (Miller et al., 1990). Since that time, several versions of WordNet have been developed. 1 Metaphoric polysemy: Refers to the poly- In this paper, we are concerned with WordNet 2.1. semy instances in which a term has literal WordNet 2.1. contains 147,257 , 117,597 and figurative meanings (Evans and Zinken, synsets and 207,019 -sense pairs. The num- 2006). In the following example, the first ber of polysemous words in WordNet is 27,006, meaning of the term fox is the literal mean- where 15776 are nouns. ing and the second meaning is the figurative. In this paper, we deal with polysemous nouns at #1 fox: alert carnivorous mammal. the concept level only. We do not consider pol- #2 dodger, fox, slyboots: a shifty ysemy at the instance level. After removing the deceptive person. polysemous nouns that refer to proper names, the remaining polysemous nouns are 14530 nouns. 2 Specialization polysemy: A type of related WordNet does not differentiate between the types polysemy which denotes a hierarchical rela- of the polysemous terms and it does not contain tion between the meanings of a polysemous any information in terms of polysemy relations term. In the case of abstract meanings, we that can be conducted to determine the polysemy say that a meaning A is a more general type between the synsets of a polysemous term. meaning of a meaning B. We may also use The researchers who attached the polysemy prob- the taxonomic notations type and subtype lem in WordNet gave different descriptions for the instead of more general meaning and more polysemy types in WordNet. For example, poly- specific meaning respectively. For example, semy reduction approaches (Edmonds and Agirre, we say that the first meaning of turtledove 2008) (Mihalcea R., 2001) (Gonzalo J., 2000) is a subtype of the second meaning. differentiate between contrastive polysemy and #1 australian turtledove, turtledove: complementary polysemy. Regular polysemy ap- small Australian dove. proaches such as (Barque and Chaumartin, 2009) #2 turtledove: any of several Old (Veale, 2004) (Peters, 2004) (Freihat et al., 2013) World wild doves. (Lohk et al., 2014) give more refined classification of the polysemy types into , metaphoric, 3 Homonymy: Refers to the contrastive specialization polysemy, and homonymy. In one polysemy instances, where meanings are not of our recent papers, compound noun polysemy is related. Consider for example the following introduced as a new polysemy type beside the for- polysemy instance of the term bank. mer four polysemy types in WordNet (Freihat et #1 depository financial institution, al., 2015). bank: a financial institution. So far, no polysemy reduction approaches have #2 bank: sloping land (especially introduced a mechanism for classifying the pol- the slope beside a body of water). ysemy types into contrastive and complemen- tary. Instead, these approaches adopt seman- 3 Approach Notations tic and probabilistic rules to discover redundant We begin with the basic notations. is the and/or very fine grained senses. On the other basic lexical unit in WordNet that refers to the base hand, the regular polysemy approaches embrace a form of a word or a collocation. Based on this defi- clear definition for classifying polysemous terms nition, we define a natural language term or simply into regular and non regular polysemy (Apresjan, a term as a lemma that belongs to a grammatical 1974). Although, the definition of regular poly- category; i.e., noun, verb, adjective or adverb. semy in these approaches is useful to distinguish between regular and non regular polysemy, these Definition 1 (Term). A term T is a quadruple xLemma, Caty, where a) entity P S is the single root of the hierarchy; a) Lemma is the term lemma; b) E „ S ¢ S; b) Cat is the grammatical category of the term. c) ps1, s2q P E if s1 s2; d) For any synset s  entity, there exists at least 1 1 Synset is the fundamental structure in Word- one synset s such that s s. Net that we define as follow. Definition 2 (WordNet synset). In this definition, point (a) defines the single root of the hierarchy and point (d) defines the x A synset S is defined as Cat, Terms, Gloss, connectivity property in the hierarchy. y Relations , where We move now to the semantics of WordNet. We a) Cat is the grammatical category of the synset; define the subset of the semantics of WordNet b) Terms is an ordered list of synonymous terms hierarchy that is relevant for our approach. A full that have the same grammatical category Cat; definition of the WordNet semantics is described c) Gloss is a text that describes the synset; in approaches such as (Alvarez, 2000) (Rudolph, d) Relations is a set of semantic relations that hold 2011) (Breaux et al., 2009). between synsets. We define the semantics of WordNet using an Interpretation I  x∆I , fy, where ∆I is an non Now, we move to the hierarchical structure empty set (the domain of interpretation) and f is of WordNet. WordNet uses the relation direct an interpretation function. hypernym to organize the hierarchical relations Definition 5 (Semantics of WordNet Hierarchy). between the synsets. This relation denotes the superordinate relationship between synsets. For Let WH  xS, Ey be wordNet hierarchy. We I example, the relation direct hypernym holds define an Interpretation of WH, I  x∆ , fy as between vehicle and wheeled vehicle where follows: I I vehicle is hypernym of wheeled vehicle. The a) entity  ∆ ; I direct hypernym relation is transitive. In the b) K  H; I I following, we generalize the direct hypernym c) @s P S: s € ∆ ; p [ qI  I X I relation to reflect the transitivity property, where d) s1 s2 s1 s2; p \ qI  I Y I we use the notion hypernym instead of a direct e) s1 s2 s1 s2; „ I „ I hypernym. f) s1 s2 if s1 s2. Definition 3 (hypernym relation). 1 1 In points a) and b), we define the empty and For two synsets s and s , s is a hypernym of s , if I 1 universal concepts. Point c) states that ∆ is the following holds: s is a direct hypernym of s , 2 closed under the interpretation function f. In and or there exists a synsets s such that s is a direct 2 2 1 d) and e), we define the conjunction and disjunc- hypernym of s and s is a hypernym of s . tion operations. In f), we define the subsumption For example, vehicle is a hypernym of car, relation. because vehicle is direct hypernym of wheeled vehicle and wheeled vehicle is a direct hyper- We present now the polysemy notations. A nym of car. term is polysemous if it is found in the terms of We use the following symbols to denote direct more than one synset. A synset is polysemous if hypernym/hypernym relations: 1 1 it contains at least one polysemous term. In the a) s s if s is a direct hypernym of s 1 1 following, we define polysemous terms. s ¦ s s s c) if is a hypernym of Definition 6 (polysemous term). A term t = xLemma, Cat, T-Ranky is polysemous Using the direct hypernym relation, wordNet 1 1 if there is a term t and two synsets s and s , organizes noun-synsets in a hierarchy that we 1 s  s such that define as follows. 1 1 a) t P s.Terms and t P s .Terms; 1 Definition 4 (wordNet hierarchy). b) t.Lemma = t .Lemma; 1 Let S  ts1, s2, ..., snu be the set of noun-synsets c) t.Cat = t .Cat. in WordNet. WordNet hierarchy is defined as a connected and rooted digraph xS, Ey, where In the following, we define polysemous synsets. Definition 7 (polysemous synset). p2 the pattern hyponyms. For exam- A synset s is polysemous if any of its terms is a ple, the structural pattern of the poly- r tbazaar, bazaru, s , s s polysemous term. semy instance 1 2 is xmercantile establishment, marketplace, shopy It is possible for two polysemous synsets to share more than one term. Two polysemous synsets and as shown in Figure 1, where mercantile their shared terms constitute a polysemy instance. establishment is the pattern root and In the following, we define polysemy instances. marketplace and shop are the pattern hy- ponyms. A special structural pattern is the Definition 8 (polysemy instance).

A polysemy instance is a triple rtT u, s1, s2s, where s1, s2 are two polysemous synsets that have the terms {T} in common.

For example, the term bazaar belongs to the following polysemy instances: rtbazaar, bazaru, #1, #2s, rtbazaaru, #1, #3s, and rtbazaaru, #2, #3s. #1 bazaar, bazar: a shop where a variety of goods are sold. Figure 1: Example of a structural pattern #2 bazaar, bazar: a street of small shops. common parent structural pattern as illustrated in bazaar, fair: a sale of miscellany; #3 Figure 2. A strcutural pattern P  xr, p1, p2y of a often for charity . polysemy instance I = r tT u, s1, s2s is a common We move now to the last part of our definitions. parent structural pattern if p1  s1 or p2  s2. We exploit the structural properties in WordNet hierarchy to identify the polysemy types of the polysemy instances in WordNet. According to the connectivity property of WordNet hierarchy in definition 4, any two synsets in wordNet have at least one common subsumer that we define as follows. Definition 9 (common subsumer).

Let s1, s2, and s be synsets in wordNet. The Figure 2: Common parent structural pattern synset s is a common subsumer of s1 and s2 if ¦ ¦ s s1 and s s2. 4 Taxonomic principles in WordNet The WordNet hierarchy is a DAG (directed WordNet hierarchy represents a classification hi- acyclic graph). This implies that it is possible erarchy where synsets are the nodes. Classifica- for two synsets to have more than one common tion hierarchies should fulfill among other require- subsumer. We define the least common subsumer ments the exclusiveness property and the exhaus- as the subsumer with the least height. tiveness property. In the following, we define structural patterns. We begin with the exclusiveness property. Definition 10 (structural pattern). Definition 11 (Exclusiveness property).

A structural pattern of polysemy instance I = Two synsets s1, s2 P S fulfill the exclusiveness r t u s  x y I [ I  KI T , s1, s2 is a triple P r, p1, p2 , where property if s1 s2 . For example, abstract a) r is the least common subsumer of s1 and s2; entity and physical entity fulfill the exclu- b) r p1 and r p2; siveness property. On the other hand expert and ¦ ¦ c) p1 s1 and p2 s2. scientist do not fulfill this property because expertI [ scientistI  KI . We call r the pattern root and p1, The exclusiveness property means that any two sibling nodes ni, nj in the hierarchy are dis- and female to a concept. This is because person I ‚ I I ‚ I joint, i.e., ni nj and nj ni . Analyz- is a direct hypernym of the concept organism and ing the structural patterns in WordNet shows that the role causal agent. the exclusiveness property is not always guar- anteed in WordNet. For example, the pattern 5 Approach Overview xperson, expert, scientisty shown in Figure 3 We exclude the structural patterns whose pattern does not fulfill this property because forcing this root resides in the first and second level in Word- property would result in preventing a scientist to Net hierarchy. Accordingly, any structural pattern be an expert or an expert to be a scientist. We whose root belongs to the synsets {entity, abstract entity, abstraction, physical entity, physical object } was automatically excluded. Our hypothesis is that the pattern hyponyms in these structural patterns in general fulfill the exclusiveness and the exhaustiveness property. These patterns are subject to our current research in discovering metonymy structural patterns. On the other hand, exclusiveness and exhaustiveness property are not guaranteed for all structural patterns whose roots reside in the third level and beyond. The input of the algorithm is the taxonomic structure of WordNet, starting from level 3, after removing lexical redundancy in com- Figure 3: An example of exclusiveness property pound nouns (Freihat et al., 2015). The output violation consists of three lists that contain specialization polysemy, metaphoric polysemy and homonymy are concerned with the cases, where the synsets instances. The first step of our algorithm is s1 and s2 are not disjoint and each of them sub- automatic, while the other two are manual. sumes a synset of the same polysemous term such as the term statistician in Figure 3. The fact that S1. Structural pattern discovery: The input of the two synsets of the polysemous terms are not this step is the current structure of WordNet disjoint implies that the polysemy type of these after removing lexical redundancy. The al- two synsets can not be homonymy, metonymy, or gorithm returns structural patterns associated metaphoric. This can be explained as follow. The with their corresponding polysemy instances. polysemy type homonymy implies that the two synsets are unrelated and that the disjointness be- S2. Structural pattern classification: In this tween the two synsets indicates a relation between step, we manually classify the structural pat- the two synsets. Metonymy on the other hand terns returned in the previous step. The out- means that one synset is a part of the other synset. put consists of four lists of patterns associ- Now, we explain the exhaustiveness property. ated with their polysemy instances. These four lists are: Definition 12 (Collective Exhaustiveness). Specialization polysemy patterns: This list Two synsets s1, s2 P S are collectively exhaustive contains the patterns whose corresponding if it is possible to find a synset s such that instances are specialization polysemy candi- I  I \ I s s1 s2 and s1, s2 fulfill the exclusiveness dates. property. Metaphoric patterns: This list contains the For example, abstract entity and patterns whose corresponding instances are physical entity fulfill the collectively ex- metaphoric candidates. haustiveness property because entityI  patterns: This list contains the abstract entityI \ physical entityI . On patterns whose corresponding instances are the other hand worker and female in the pattern homonymy candidates. xperon, worker, femaley do not fulfill this Singleton patterns: The patterns in this group property because worker corresponds to a role are those patterns that have one polysemy in- stance only and thus cannot be considered to organism, they are not collectively exhaustive as be regular. explained. The polysemy instances that belong to this pattern are 326 instances. Consider for S3 Identifying false positives: In this step, we example the following instance. manually process the polysemy instances in #1 snake, serpent, ophidian: limbless the four lists from the previous step. Our scaly elongate reptile. task is to decide the polysemy type for the #2 snake, snake in the grass: a instances in the singleton patterns list and re- deceitful or treacherous person. move false positives form the other three lists. Another example is the pattern x y 6 Metaphoric Structural Patterns attribute, property, trait . Although, both synsets share the same hypernym attribute, Identifying metaphoric patterns is based on the they are not collectively exhaustive be- distinction between the literal meaning and the cause traitI is a special case of propertyI figurative meaning. Our idea is that it is not (traitI  propertyI [ personI ). The polysemy possible for a literal and the figurative meaning instances that belong to this pattern are 111 to be collectively exhaustive. Violating the instances. Consider for example the following exhaustiveness property in a structural pattern instance. xr, p1, p2y may be a result of the following: #1 softness:the property of giving little a) p1 and p2 belong to different types and can not resistance to pressure and being easily be subsumed by the pattern root r, or cut or molded. b) p1 € p2 or p2 € p1. #2 gentleness, softness, mildness: For example female and worker can not acting in a manner that is gentle and be subsumed by person in the pattern mild and even-tempered. xperson, female, workery as shown in Fig- ure 4. On the other hand, it is correct that 7 Specialization Polysemy Structural Patterns We use the exclusiveness property and the pattern root in a structural pattern to discover specializa- tion polysemy candidates indirectly. The relation between the synsets in specialization polysemy is hierarchical. The hierarchical relation between the synsets in a specialization polysemy instance indi- cates that the exclusiveness property does not hold between synsets and thus between the structural pattern hyponyms. Figure 4: Example of a metaphoric polysemy in- We define specialization polysemy patterns as fol- stance lows. Definition 14 (specialization polysemy structural person and animal are organisms in the structural pattern). xorganism, animal, persony but it is clear that p  xr, p , p y personI € animalI A pattern 1 2 is a specialization polysemy pattern if a) and b) hold In the following, we define metaphoric patterns a) p and p do not fulfill the exclusiveness structural pattern as follows. 1 2 property. Definition 13 (Metaphoric structural pattern). b) p1 and p2 fulfill the exhaustiveness property. A pattern p  xr, p1, p2y is metaphoric if p1 and In the following we give examples for identified p2 do not fulfill the collectively exhaustiveness specialization polysemy patterns. All instances property. that belong to the common parent structural In the following we give examples for iden- patterns are classified as specialization polysemy tified metaphoric patterns. The pattern instances. The polysemy instances that belong xorganism, animal, persony is metaphoric. to this pattern are 2879 instances. Consider for Although both synsets share the same hypernym example the following instance. #1 capital, working capital: assets flowers. available for use in the production of #2 red fox, Vulpes fulva: New World fox; further assets. often considered the same species as the #2 capital: wealth in the form of Old World fox. money or property owned by a person or Another example is the pattern business and human resources of economic xvertebrate, bird, mammaly. The poly- value. semy instances that belong to this pattern are 13 Another example is the pattern instances. Consider for example the following. xact, action, activityy. The polysemy in- #3 griffon, wire-haired pointing griffon: stances that belong to this pattern are 406 breed of medium-sized long-headed dogs. instances. Consider for example the following. #4 griffon vulture, griffon, Gyps fulvus: #1 employment, work: the occupation for large vulture of southern Europe and which you are paid. northern Africa. #2 employment, engagement: the act of giving someone a job. 9 False Positives Identification Another example, is the pattern xanimal, invertebrate, larvay. The poly- In this section, we describe the third step of our semy instances that belong to this pattern are 17 approach. Our task here is to process the four lists instances. Consider for example the following. returned at the end of the pattern classification #1 ailanthus silkworm, Samia cynthia: and remove false positives. These lists are the large green silkworm of the cynthia moth. metaphoric polysemy list, the specialization #2 cynthia moth, Samia cynthia, Samia polysemy list, the homonymy list, and a list of walkeri: large Asiatic moth introduced non regular (singleton patterns) list. This task can into the United States; larvae feed on only be performed manually due to the implicit the ailanthus. and missing information in synset glosses. Our procedure for determining the polysemy class of 8 Homonymy Structural Patterns a polysemy instance is based on the three defini- tions in the previous section, where we process We define homonymy patterns as follows. the polysemy instances instance by instance to Definition 15 (Homonymy structural pattern). determine the the relation between the synsets of A pattern p  xr, p1, p2y is homonymy pattern if the polysemy instances. the following condition hold. If a polysemy instance does not belong to the a) p1 and p2 fulfill the exclusiveness property; polysemy type it was assigned to (false positive b) p1 and p2 fulfill the exhaustiveness property; instance), we assign it to its corresponding poly- c) There is no relation between p1 and p2. semy type. In the following we give examples for iden- In the following, we give examples for false tified homonymy patterns. The pattern positives. The common parent structural pattern xorganism, person, planty. The polysemy which was automatically assigned to the spe- instances that belong to this pattern are 40 cialization polysemy type (step 1 in Section 5) instances. Consider for example the following contains 180 false positive polysemy instances, 98 instance. of them were identified as homonymy instances. #1 spinster, old maid: an elderly One example is: unmarried woman. #1 cardholder: a person who holds a #2 zinnia, old maid, old maid flower: credit card or debit card. any of various plants of the genus #2 cardholder: a player who holds a card Zinnia. or cards in a card game. Another example is the pattern Metaphoric false positives (82 instances) were xorganism, animal, planty. The polysemy also identified in the common parent class. Con- instances that belong to this pattern are 41 in- sider for example the following instance. stances. Consider for example the following. #1 game plan: (figurative) a carefully #1 red fox, Celosia argentea: weedy thought out strategy for achieving an annual with spikes of silver-white objective in war. #2 game plan: (sports) a plan for agreement of the evaluators with our approach was achieving an objective in some sport. on 96.5% of the instances. In the following Table Another example is the pattern 3, a refers to our approach, e1, e2 refer to evalua- xorganism, animal, persony which was as- tor1 and evaluator 2 respectively. signed to the metaphoric polysemy type contains 326 polysemy instances, 74 of them were identi- e1  a _ e2  a 3665 (96.5%) fied as such as the following instance. a  e1 3621 (95.3%) #2 Minnesotan, Gopher: a native or a  e2 3600 (94.8%) resident of Minnesota. Table 3: Evaluation of the polysemy classification #3 ground squirrel, gopher, spermophile: any of various terrestrial burrowing rodents of Old and New Worlds. 11 Conclusion and future Work In this paper, we have presented how to use two 10 Results and Evaluation taxonomic principles for classifying the polysemy The number of polysemy instances computed by types in WordNet. We have demonstrated the use- the polysemy instances discovery algorithm is fulness of our approach on classifying three pol- 41306. We excluded 28318 instances because the ysemy types, namely, specialization, metaphoric pattern roots of these instances reside in the first and homonymy. In this approach, we were and the second level of the hierarchy as per the able to discover all specialization polysemy struc- approach discussed in Section 5.The remaining tural patterns and subsets of the metaphoric and number of polysemy instances is 12988. These metonymy structural patterns. We aim to continue instances are divided in two groups as follow. our work to study the metonymy patterns in the 12988 of these instances belong to 1028 regular upper level of WordNet hierarchy, where we gen- type compatible patterns and 1569 instances be- eralize our structural pattern definition as follows. long to single tone patterns. The classification of Definition 16 (generalized structural pattern). the pasterns and the result of the false positive re- A structural pattern of polysemy instance I = moving is shown in the following tables. r tT u, s1, s2s is a triple P  xr, p1, p2y, where a) r is the least common subsumer of s and s ; #Type #patterns #instances 1 2 b) r ¦ p and r ¦ p ; Specialization 823 9902 1 2 c) p ¦ s and p ¦ s . Metaphoric 134 1697 1 1 2 2 Our hypothesis is that in case of metonymy struc- Homonymy 71 1389 tural patterns: the nodes p and p fulfill the ex- Total 1028 12988 1 2 clusiveness and the exhaustiveness properties and Table 1: Classification of the regular structural there is a part of relation between p1 and p2. The patterns conditions for metaphoric and homonymy struc- tural patterns obtained by adapting the new struc- In Table 2, we show the results removing false pos- tural definition remain the same as explained in itive instances, where we see that the average false this paper. positives is about 17%. Acknowledgment

#Poly Type #Instances #False Positives The research leading to these results has re- Specialization 9902 1740 ceived partially funding from the European Com- Metaphoric 1697 175 munity’s Seventh Framework Program under Homonymy 1389 295 grant agreement n. 600854, Smart Society Total 12988 2210 (http://www.smart-society-project.eu/).

Table 2: False Positives in Pattern Classification References To evaluate our approach, 3797 polysemy in- Jordi Alvarez. 2000. Integrating the ontology stances were evaluated by two evaluators. The into a description logic system. JU. Apresjan. 1974. Regular polysemy. Linguistics, Rila Mandala, Takenobu Tokunaga, and Hozumi pages 5–32. Tanaka. 1999. Complementing wordnet with ro- get’s and corpus-based thesauri for information re- Kenneth D. Bailey. 1994. Typologies and Taxonomies: trieval. In EACL, pages 94–101. The Association An Introduction to Classification Technique. Sage for Computer Linguistics. Publications, Thousand Oaks, CA, 1994///. Alberto Marradi. 1990. Classification, typology, tax- Lucie Barque and Franois-Rgis Chaumartin. 2009. onomy. Quality & Quantity: International Journal Regular polysemy in wordnet. JLCL, 24(2):5–18. of Methodology, 24(2):129–157. MoldovanD. I. Mihalcea R. 2001. Ez.wordnet: Prin- Travis D. Breaux, Annie I. Anton, and Jon Doyle. ciples for automatic generation of a coarse grained 2009. Semantic parameterization: A process for wordnet. FLAIRS Conference, pages 454–458. modeling domain descriptions. ACM Transactions on Software Engineering Methodology, 18(2). G. A. Miller, R. Beckwith, C. Fellbaum, D. Gross, and K. J. Miller. 1990. Introduction to WordNet: an on- Paul Buitelaar. 1998. Corelex: Systematic polysemy line lexical database. International Journal of Lexi- and underspecification. PhD thesis,Brandeis Uni- cography, 3(4):235–244. versity, Department of Computer Science. George A. Miller. 1995. Wordnet: A lexical database W. B. Dolan. 1994. ambiguation: clus- for english. Commun. ACM, 38(11):39–41, Novem- tering related senses. In Proceedings of COLING94, ber. pages 712–716. Wim Peters. 2004. Detection and characterization Philip Edmonds and Eneko Agirre. 2008. Word sense of figurative language use in wordnet. PhD thesis, disambiguation. Scholarpedia, 3. Natural Language Processing Group, Department of Computer Science, University of Sheffield. Vyvyan Evans and Jrg Zinken. 2006. Figurative lan- Sebastian Rudolph. 2011. Foundations of description guage in a modern theory of meaning construction: logics. In Arenas C. Polleres A., dAmato´ C., editor, A lexical concepts and cognitive models approach. Reasoning Web. Semantic Technologies for the Web of Data - 7th International Summer School 2011, Ingrid Lossius Falkum. 2011. The semantics and prag- volume 6848 of LNCS, pages 76–136. Springer. matics of polysemy: A relevance-theoretic account. PhD thesis, University College London. Tony Veale. 2004. Pathways to creativity in lexical on- tologies. In In Proceedings of the 2nd Global Word- Abed Alhkaim Freihat, Fausto Giunchiglia, and Net Conference. Biswanath Dutta. 2013. Regular polysemy in word- net and pattern based approach. International Jour- nal On Advances in Intelligent Systems, 6(3&4), jan.

Abed Alhkaim Freihat, Biswanath Dutta, and Fausto Giunchiglia. 2015. Compound noun polysemy and sense enumeration in wordnet. In Proceedings of the 7th International Conference on Information, Pro- cess, and Knowledge Management (eKNOW), pages 166–171.

Verdejo F. Gonzalo J., Chugur I. 2000. Sense clus- ters for information retrieval: Evidence from semcor and the eurowordnet interlingual index. ACL-2000 Workshop on Word Senses and Multi-linguality, As- sociation for Computational Linguistics, pages 10– 18.

Ahti Lohk, Kaarel Allik, Heili Orav, and Leo Vhandu. 2014. Dense components in the structure of word- net. In Nicoletta Calzolari (Conference Chair), Khalid Choukri, Thierry Declerck, Hrafn Lofts- son, Bente Maegaard, Joseph Mariani, Asuncion Moreno, Jan Odijk, and Stelios Piperidis, editors, Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC’14), Reykjavik, Iceland, may. European Language Re- sources Association (ELRA).