Retrieving Ergative Verbs from a Lexical Data Base

RETRIEVING ERGATIVE VERBS FROM A LEXICAL DATA BASE THIERRY FONTENELLE and JOCELYNE VANANDROYE This paper refers to work carried out at the University of Liège on the Longman Dictionary of Contemporary English (hereafter LDOCE). The magnetic tape of this dictionary, first edition, was made available to us under contract with the publishers for research purposes in 1979. The tape, which was destined for typesetting, was first transformed into and exploited under STAIRS, a software retrieval package for IBM mainframes. In 1983, Archie Michiels and Jacques Jansen started working towards implementing LDOCE under dBase III+, a standard data base management system, which required much preparatory work. We will here simply point to major operations: a.the decompaction of grammatical codes. These grammar codes consist of a letter followed by a number; the letter describes the syntactic behavior of the lexical item (transitive , countable, etc.), and the number describes the environment in which this item can occur. As these codes occur in a compact form in the original version, the letter bearing on two or more numbers, they had to be rewritten so as to be automatically retrievable. b.the separation of examples from definition in the definition field and the separation of the newly created example field into its various constituents. 11 12Thierry Fontenelle and Jocelyne Vanandroye Let us take the following example: Dictionary entry tell 1 [Dl (io),5a,b;Tl {of)] to make (something ) known in words to (someone); express in words; speak: Did you tell Aunt Joan the news about Paul?/John told us he'd seen you in town./I can't tell you how pleased I am to be here tonight, . . . dBase entry: tell 1 to make (something) known in words to (someone); express in words; speak 1 Code: Dl ditransitive v/ [ NPn NPn] Right: - I to 2 Code: D5a ditransitive v/ [ NPn (that) S] 3 Code: D5b ditransitive v/ [ NPn SO/NOT] 4 Code: Tl monotransitive ?/ [ NPn] Right: - I of Did you tell Aunt Joan the news about Paul? John told us he'd seen you in town. I can't tell you how pleased I am to be here tonight c. the splitting up of definitions containing subdefinition markers (a, b) Dictionary entry: harden 4[often pass.] a (of water) to become HARD1 (8): Water hardens when lime gets into it b to cause (water) to become hard: Water is hardened by lime Retrieving Ergative Verbs from a Data Base13 dBase entry: harden 4 a (of water) to become HARD@1 (8) Water hardens when lime gets into it harden 4b to cause water to become hard Water is hardened by lime This paper concentrates on the use we made of LDOCE under dBase III+ in order to retrieve a computationally relevant subset of verbs, namely ergative verbs. I. Linguistic Motivations Following Lyons (352ff.) we define ergative verbs as verbs that may be used both transitively intransitively, and where the subject of the intransitive construction can become the object of the transitive construction. These verbs display the following alternation: John opened the door vs the door opened John cooked the rabbit vs the rabbit cooked In terms of semantic roles, the verbs involve an agent (generally an animate entity) and a patient (generally an entity that changes state). The alternation illustrated above is referred to as the causative/inchoative alternation (see Levin). This means that these verbs actually have a causative and a non-causative use and that the semantic relations between the verb and its arguments may be expressed in two different ways. The causative use of a verb (i.e., the transitive construction) implies that the agent is realized as the subject and the patient argument is expressed as the object . On the contrary, the non-causative (i.e., intransitive) 14Thierry Fontenelle and Jocelyne Vanandroye construction only involves a patient realized as the subject of the verb. Consequently, the class of ergative verbs is to be contrasted with purely causative verbs that only exhibit the transitive construction (e.g., to raise). Ideally, a dictionary organized in data base format should allow for the representation of such thematic relations. Semantic roles such as AGENT, PATIENT, THEME, INSTRUMENT , etc. should certainly be coded...