File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/metho/91/w91-0208_metho.xml
Size: 22,728 bytes
Last Modified: 2025-10-06 14:12:48
<?xml version="1.0" standalone="yes"?> <Paper uid="W91-0208"> <Title>Touche Ross Management Consultants</Title> <Section position="2" start_page="0" end_page="76" type="metho"> <SectionTitle> 1. Introduction: Lexical Implication Rules </SectionTitle> <Paragraph position="0"> It is a truism that people, in interpreting and producing language, make use of both linguistic and real-world knowledge. We contend that part of that linguistic knowledge is a knowledge of lexical semantics. Lexical semantics is discernible from real- world knowledge, because the inferences used are in certain cases formally distinct from commonsense inferences and from metaphors based on analogies. That is the position we attempt to substantiate in this paper.</Paragraph> <Paragraph position="1"> The focus of this paper is a series of arguments justifying the existence of Lexical Implication Rules (henceforth &quot;LIRs&quot;). Such rules in principle generate derived lexical entries from base lexical entries. We contend that they are bona fide rules in the formal grammar of languages involved; it is wrong to see them as mere reflections, quasi-systematic effects of applying the formal structures of a language in a real-world context.</Paragraph> <Paragraph position="2"> As evidence of this we adduce formal properties of the incidence of some of the rules. We postulate: * that application of the rules is sometimes blocked by the pre-existing topography of the lexicon (pre-emption); * that formal features of the words involved, apparently unrelated to the semantic effect of the rules, may inhibit LIR application; * that semantic features of the words input, apparently unrelated to the semantic effect of the rules themselves, may similarly inhibit LIR application; * that some rules are specific to particular languages, and indeed to specific dialects of a given language.</Paragraph> <Paragraph position="3"> All these facts seem incompatible with the hypothesis that LIRs are side-effects of the the interaction of language use with speakers' knowledge of the real world. They call for a structured theory of lexical semantics, and this paper is intended as a contribution to building such a theory.1 The position outlined here arises from current work on the analysis and categorization of a database of well over 100 instances of predictable lexical and semantic alternations. These instances of systematic word-sense relationship are hard to categorize fully, but the majority appear to have properties which show them not to be mere corollaries of speakers' free play with analogy and terse assumption of real-world situations. They have something in common with the set of Lexical Rules used in Lexical Functional Grammar (cf Bresnan (1982)), but are primarily conceptual rather than syntactic. While it is already clear that there is great variety in the theoretical status of the rules in this set, we shall argue that there is clear evidence that at least some of these rules, the LIRs, must be defined as applying to word-meanings; they are not derivable as generalizations over worldknowledge, or facts about the world or the society described, but must be stated in a full grammar of the language in which they occur.</Paragraph> <Paragraph position="4"> The significance of this is more than a theoretical nicety. It is important to computational linguistics to include these facts in its lexicon, rather than as a special kind of world knowledge, since the former will be in general more restricted and hence more manageable.</Paragraph> </Section> <Section position="3" start_page="76" end_page="76" type="metho"> <SectionTitle> 2. Some First Examples of LIRs; Contrast with some other Semantic Shifts </SectionTitle> <Paragraph position="0"> A few straightforward examples from English will give a first intuitive feel for the kind of lexical implications to be represented as LIRs.</Paragraph> </Section> <Section position="4" start_page="76" end_page="76" type="metho"> <SectionTitle> LIR Hunting Plural </SectionTitle> <Paragraph position="0"> NC (count noun) singular: animal -> NPL: in hunting context e.g. I can see an elephant. / They've gone out after elephant.</Paragraph> <Paragraph position="1"> also tiger, shark, shrimp, partridge, pigeon etc.</Paragraph> <Paragraph position="3"> using that vehicle e.g. He's got a new cycle. / Let's cycle into town. also motor, ferry, canoe, ship, jet etc.</Paragraph> </Section> <Section position="5" start_page="76" end_page="76" type="metho"> <SectionTitle> LIR Animal - Meat </SectionTitle> <Paragraph position="0"> NC: animal -> NU (uncountable or mass): meat from that animal e.g. Mary had a little lamb. / He won't touch lamb any more.</Paragraph> <Paragraph position="1"> also chicken, goose, monkey, dog, swordfish, shark etc.</Paragraph> <Paragraph position="3"> NU: energy source -> NMOD (noun modifier): powered by that source e.g. There's no gas left. / A gas oven.</Paragraph> <Paragraph position="4"> also steam, water, wind, coal, air etc.</Paragraph> </Section> <Section position="6" start_page="76" end_page="79" type="metho"> <SectionTitle> LIR Dance Verb </SectionTitle> <Paragraph position="0"> NC: dance -> VI: do that dance e.g. Can you do a waltz? / He's learning to waltz. also tango, quickstep, foxtrot, cancan etc.</Paragraph> <Paragraph position="1"> (5) of language as it is used; and (b) something impossible or imperceptible will not be lexicalized. (We characterize below the phenomenon of pragmatic exclusion.).</Paragraph> <Paragraph position="3"> NC: container -> \[NC+ful \]NC: amount it contains/its contents e.g. I don't use a thimble. / Add a mere thimbleful of brandy.</Paragraph> <Paragraph position="4"> also all purpose-built containers (trunk, tank, jug, hamper, bucket, basket etc.), and some other items that can be viewed as containers (e.g. spade, spoon, fork, hand, fist, belly,...) L1R Container - Amount/Contents (7) NC: container (purpose-built only) -> NC: amount it contains/its contents e.g. The glass broke. / Add a glass of wine. Don't drink the whole glass. also all purpose-built containers (trunk, tank, jug, hamper, bucket, basket etc.) LIR Material Adjective NU: material -> NMOD: made of that material e.g. It's best quality linen. / A linen sheet.</Paragraph> <Paragraph position="5"> also polyester, plastic, stone, diamond, granite, lead etc. (8) LIR -en Derived Adjective (defunct?) NU: material -> \[NU+-en\] ADJ : made of that material e.g. The ring is made of a gold. / A golden ring.</Paragraph> <Paragraph position="6"> also wood, leather, oak, lead, wool LIR Constitution - State NU: form of government -> NC: state with that form e.g. Dem~racy is at risk. / Democracies must take action. also oligarchy, monarchy, tyranny, dictatorship etc.</Paragraph> <Paragraph position="7"> (9) (10) LIR Nationality Plural ADJ: nationality -> \[the ADJ\]Np L * people of that nationality e.g. The French nation. / The French have objected to this. also Swiss, Dutch, Chinese, Spanish etc.</Paragraph> <Paragraph position="8"> (11) LIR Produce Young NC: animal young -> VI: produce young e.g. It's a foal. / The mare will foal shortly. also cub, pup, whelp, lamb etc.</Paragraph> <Paragraph position="9"> (12) LIR Feeling - Evoking (13) ADJ: (of subject) feeling an emotion -> ADJ: (of cause) evoking that emotion e.g. She's sad about it. / It was a sad day.</Paragraph> <Paragraph position="10"> also happy, angry, nostalgic, melancholy etc.</Paragraph> <Paragraph position="11"> food item -> NU: food substance Here's an egg. / He won't eat egg.</Paragraph> <Paragraph position="12"> potato, lettuce, banana, coconut, haggis, sausage, pie etc. (15) LIRs are distinguished in various ways from some comparable phenomena discussed in the literature: 1. They are more systematic than the Lexical Network framework used by Norvig to link the remoter senses of take in Norvig & Lakoff (1987); cf also Norvig (1989).</Paragraph> <Paragraph position="13"> 2. Each implication instance has a single clear base form. In this they are unlike the semantic alternations arising from the metaphorical concepts such as &quot;argument is war&quot; or &quot;communication is a conduit&quot; of Lakoff & Johnson (1980) and Sweetser (1990), which are often better represented as projections from one network of semantic relations on to an isomorphic one.</Paragraph> <Paragraph position="14"> 3. They do not depend on special background knowledge, unlike many of the innovative denominals of Clark & Clark (1979) (e.g.We all Wayned and Cagneyed ).2 4. They are not coerced by a governing head, unlike the instances of contextual determination (three martinis ago, before the jet engine) pointed out by Fillmore (p.c.). 3. Formal Definition of LIRs The form of an LIR may be given schematically as follows, using Cruse's terminology (cf Cruse 1986): A lexical unit LU1, consisting of a lexical component LC1 (consisting of phonological, orthographic, morphological and syntactic information) and a semantic component SC1, implies the existence of a lexical unit LU2, with lexical component LC1 or LC2 and semantic component SC2.</Paragraph> <Paragraph position="15"> More graphically: (i) LC l\[...a...\] : SC 1 \[...t~...\]--> LC 1 \[...a...\] : SC 2 \[...x...\] or (j) LC 1 \[...a...\] : SC 1 \[...~...\] --> LC 2 \[...0~...\] : SC 2 \[...'t...\] This means that within our definition of LIRs are included all defivational rules, since these always affect the sense of their input-word. Note that, on this theory, LIRs may alter the lexical component of an input word, but they must alter its semantic component. The formal change brought about by an LIR may be vacuous, but the semantic change can never be. LIRs, therefore, on our conception, are centrally concerned with semantics, but only incidentally with formal transformations. Hence, for example, the many lexical processes in English which involve transition from a count noun to a mass noun (e.g. Tree Wood; Animal - Fur; Animal - Meat; Food Item - Mass) will be viewed as distinct LIRs.</Paragraph> <Paragraph position="16"> The arrow in the LIR-schema makes it clear that LIRs are viewed as denvational processes, not as symmetrical relations. The claim is made that each implication instance has a single clear base form. 3 This is not an idle claim, but is in fact supported by the facts adduced to show the conditions under which LIRs are blocked (viz by Pre-emption, and by Formally- and Semantically-Based Exceptions, as exhibited below). Such constraints on the application of a rule block the existence of a potential output for a given input. Without clear directionality in the LIRs that are blocked, such constraints would be meaningless, or at least unspecific in their effect.</Paragraph> <Paragraph position="17"> 4. Our Database and its Usefulness in Lexicography. The R61e of Theory. Created (by Atkins) in an effort to systematize and accelerate the process of dictionary compiling, the database draws on the relationships discussed by Fillmore (e.g. 1969, 1978 and elsewhere) ; Apresjan's cases of regularpolysemy (1973); Leech's semantic transfer rules (1981); the verb alternations noted in Levin (forthcoming); the cases of systematic polysemy discussed in Miller (1978), Clark & Clark (1979), Aronoff (1980) and Lehrer (1990); items participating in the cross-language lexicalizafion patterns which Talmy discusses (1985); the nom/nal compounds analysed by Levi (1982); and Pustejovsky's generative lexicon (1991).</Paragraph> <Paragraph position="18"> The database is far from complete. It consists at present of a dBaselII Plus file, of which each of the 130 records holds information on one particular type of systematic meaning shift, and separate lists (as complete as possible) of the sets of words which participate in each type, and those which might be expected to, but do not.</Paragraph> <Paragraph position="19"> At present the lists of words which participate in the various identified meaning shifts may be used to remind a lexicographer that a particular word is expected to have certain specified usages, and to structure in advance certain aspects of the dictionary entry accordingly. In lexicography, when teams of a dozen or so people work on producing for one dictionary consistent descriptions of 100,000 or more words, such preparatory marking of items in the wordlist already makes the compiling faster and more consistent across members of the team. The computer can present them with an appropriate pro-forma entry to complete, rather than their having to compile each word from scratch.</Paragraph> <Paragraph position="20"> But besides these pragmatic aims, we are also concerned to clarify the theoretical status of the various types of meaning shift. Success here will allow rules to be formulated more accurately and comprehensively. In addition, by illuminating the semantic primitives which play a r61e in the rules, it should also lead towards standardization of the semantic component of lexical entries. We now turn to the evidence for LIRs as linguistic rules.</Paragraph> </Section> <Section position="7" start_page="79" end_page="81" type="metho"> <SectionTitle> 5. Formal Properties of LIRs - I: Pre-emption </SectionTitle> <Paragraph position="0"> Pre-emption is the prevention of application of a rule because the place for its output is, in some way, already taken. 4 We have noted two kinds of pre-emption blocking LIRs: semantic pre-emption, where the sense is already represented by a 4pre-emption is explicitly noted in this context by Clark & Clark (1979: p. 798). Cf also Gruber (1976: 295-6). These, however, only note the phenomenon of what we call semantic preemption. null is already there with a different meaning, viz &quot;to stagger, as from a shock&quot;. Hence it is not available to mean &quot;do the (Highland) reel&quot;: *the party reeled the night away. Although pre-emption of these kinds does appear to be a sufficient condition, and hence good evidence, for a semantic shift's being an LIR, it is not apparently a necessary one: sometimes it fails to work where LIRs do seem to be involved.</Paragraph> <Paragraph position="1"> For instance, -ful Derived Noun creates Contents and Amount senses, sometimes on the basis of quite unpromising nouns: belly -> a bellyful of rice & beans, cap -> a capful of disinfectant,fist -> a fistful of dollars, hand -> a handful of dust, needle -> a needleful of cotton, pocket -> a pocketful of rye, shovel -> a shovelful of spaghetti, spade -> a spadeful of earth, spoon -> a spoonful of sugar etc.</Paragraph> <Paragraph position="2"> However, the rule also applies to 'true' Container nouns (words whose primary sense is a hyponym of container): bottle... -> bottleful, bucketful, cupful glassful potful, sackful, tankful, thimbleful. In these applications, the rule's semantic effect seems to be indistinguishable from that of Container - Amount/Contents.</Paragraph> <Paragraph position="3"> Nonetheless, this latter is not pre-empted e.g. he drank the whole bottle(ful); the pail holds eight bottle(ful)s of milk. 5 Another case of failed pre-emption concerns the LIR Material Adjective. This rule which allows the names of materials to be used as adjectives to describe objects made of the material is unaffected by the application of the somewhat sporadic -en Derived Adjective rule.</Paragraph> <Paragraph position="4"> E.g., a gold ring vs. a golden ring a silk shirt vs. a silken shirt an oak chest vs. an oaken chest a leather bottle vs. a leathern bottle 5There must be an LIR involved here, and not just a vagueness in the sense of &quot;container nouns&quot;: the senses Amount and Contents are clearly distinct from the sense Container. To show this it is enough to point out that nouns like handful and bottleful have the senses Amount and Contents but not the sense Container.</Paragraph> <Paragraph position="5"> but a silver ring vs. *a silvern ring a hair shirt vs. *a hairen shirt a teak chest vs. *a teaken chest a rubber bottle vs. *a rubbern bottle 6 For the moment it is enough to point out that pre-emption is not always clear-cut and co-ercive. Where it is present, however, it indicates clearly the presence of an LIR.</Paragraph> <Paragraph position="6"> Pre-emption proper, however, should be distinguished from what we might call pragmatic exclusion. These are cases where a predicted application of a rule is blocked because of external fact about the world. Compounding with place-names can imply Source as well as DestinationT: e.g. When is the London train due in? (= &quot;train from London&quot;) vs. When is the London train due to leave? (=&quot;train for London&quot;). However, the Source option seems strangely to be excluded for roads. &quot;The London road&quot; means the road to London, not away from it. The paradox goes away when one considers the point of view of road users. Either they are in London, or they are not. If they are, then every road they might take is a road from London; talking about &quot;the London road&quot; in this sense is of little use to them. If they are somewhere else, then every road from London is co-extensive with a road to London: why bother to distinguish? Both sets of road-users can be kept happy if the convention is adopted that &quot;London road&quot; = &quot;road to London&quot;.</Paragraph> </Section> <Section position="8" start_page="81" end_page="83" type="metho"> <SectionTitle> 6. Formal Properties of LIRs - 2: Formally-Based Exceptions </SectionTitle> <Paragraph position="0"> Certain constraints on particular LIRs are quite formal. They have to do with non-semantic aspects of the lexical items involved, principally their phonological form, but also their morphological or syntactic sub-category. Such examples provide perhaps the clearest evidence that LIRs are not an effect of pragmatic attempts to use words tersely and interpret them constructively.</Paragraph> <Paragraph position="1"> Phonological Exceptions (i) Constitution - State provides quite generally for mass nouns designating types of government to act as count nouns designating states under such government.</Paragraph> <Paragraph position="2"> Hence there is democracy, tyranny, dictatorship, oligarchy, and there are democracies, tyrannies, dictatorships, oligarchies. 8 However, there is one large class of nouns, those ending in -ism, for which this LIR is blocked: there are no *fascisms, *communisms, *nazisms, *totalitarianisms, *marxisms, *leninisms, 6Evidently, there is vagueness in the precise definition of the two processes concerned. Some would say the adjectives in -(e)n are moving away from their original meaning &quot;made of&quot; to something vaguer such as &quot;redolent of&quot;; others that the outputs of Material Adjective are not really adjectives, but just compound uses of the original nouns. There are tests that bear on these points but the results are variable and inconclusive.</Paragraph> <Paragraph position="3"> 7Obviously, these possibilities are not exhaustive.</Paragraph> <Paragraph position="4"> 8Some types of r6gime have no corresponding term for a theory of government: e.g. republic, kingdom, dukedom, heptarchy. This is because not all r6gime-terms are derived from terms for political theories (contrast monarchy (which is) with kingdom (which is not)). In other cases, the possibility of a corresponding r6gime is only pragmatically excluded : politics, not grammar, precludes anarchies, bureaucracies and officialdoms.</Paragraph> <Paragraph position="5"> *stalinisms or *maoisms, although it is quite clear which states would have been candidates for such titles. 9 (ii) Nationality Plural allows for nationality adjectives, when preceded by the definite article, to also function as terms for a plural noun, referring to a given group of, or to all people of, that nationality. Strangely, however, the rule is subject to a phonological restriction: only terms ending in a sibilant consonant are eligible. So we may refer to the Swiss, the Spanish, the Dutch, the French, the Japanese, but not (as a collective plural) to *the Belgian, *the German, *the Coptic, *the Iraqi or *the Malay. 10</Paragraph> <Section position="1" start_page="82" end_page="82" type="sub_section"> <SectionTitle> Morphological Exception </SectionTitle> <Paragraph position="0"> Produce Young provides for the use of any name of a young mammal as a verb to denote the giving of birth to such an animal. Thus cows calve 11, mares foal, ewes lamb, vixens cub, dogs whelp, cats litter, sows farrow, seals pup, and does fawn. However, the rule excepts diminutives: kitten, piglet and puppy form no verbs. 12</Paragraph> </Section> <Section position="2" start_page="82" end_page="83" type="sub_section"> <SectionTitle> Syntactic Exception Feeling - Evoking applies to adjectives, converting an attribute of a person feeling </SectionTitle> <Paragraph position="0"> an emotion to that of any object causing the emotion. It appears to be systematically blocked from applying to any adjective which is only used preedicatively. Hence we have the following contrasts: a happy child *a glad child : the child is happy :: a happy occasion : the child is glad :: *a glad occasion : the occasion is happy : *the occasion is glad a fearful child *an afraid child : the child is fearful :: a fearful occasion : the child is afraid ::*an afraid occasion : the occasion is fearful :*the occasion is afraid a unhealthy child *an ill child : the child is unhealthy :: a unhealthy occasion : the occasion is unhealthy : the child is ill ::*an ill occasion :*the occasion is ill a rapt child : the child is rapt :: a rapt occasion *an agog child : the child is agog : :*an agog occasion : ?the occasion is rapt :*the occasion is agog 9An exception to the constraint is despotism, especially in the usage oriental despotism. 10Non-naturalized names are another, perhaps semantic, exception. The Hausa, the Wolof, the Konkani, the Tuareg strike this writer at least as acceptable collective plurals. Interestingly, too, the Maya seem OK, while the Inca and Chibcha are not. 11Th e verb calve instead of the expected *calf is a mild case of pre-emption in Modem English. The voicing-alternation, however, was once regular, and is merely evidence that the origins of this LIR go back at least to Middle English. 12Though, as the examples show, these could also be excluded by pre-emption.</Paragraph> </Section> </Section> class="xml-element"></Paper>