In any system for Natural Language Processing having a dictionary, the question arises as to whiclh entries are included in it. In this paper, I address the subquestion as to whether a lexical unit having two senses should be considered ambiguous or vague with respect to them. The inadequacy of some common strategies to answer this question in Machine Translation (MT) systems is shown. From a semantic conjecture, tests are developed that are argued to give more consistent and theoretically well-founded results.
In any system for Natural Language Processing having a dictionary, the question arises which entries are included in it. In this paper, I will assume the environment of a mnltilingual MT system based on a linguistic analysis and transfer architecture, from which I will derive some argumentation.
The question which entries are included in the dictionary should be answered in two parts. First there is a mapping from graphic words to lexical units (lu's), then a mapping from lu's to readings, each of which is represented in an entry. The former mapping represents acertain level of analysis of the graphic word. It abstracts away from inflection and spelling variation, and, depending on the system's analysis component, may do so as well for productive
*I would like to thank my colleagues at the university and in Eurotra, especially Louis des Tombe and Henk Verkuyl, for their helpful comments. 162 derivation and compounding, and multi-word units. In this paper I will concentrate on the latter mapping, reading distinction, in a way that does not appeal to a particular choice on the relation between lu and graphic word.
A consistent approach to reading distinction is necessary, because inconsistencies in reading distinction in an MT system will complica.te transfer components between a pair of languages, and jeopardize extensibility of the system. A correct solution will save time in development and improve performance. The central question in this area can be formulated as in (1). (1) Given an lu X and two of its senses ocl
and 5'2, is X ambiguous or vague with
respect to S1 and $2 ? In (1) a sense of an lu is the meaning the 1ll has in a certain set of contexts. If the lu is vague, both senses are covered by the same reading. If it is ambiguous, S1 and $2 are examples of different readings of the lu, each reading being represented in a single entry.
Since every reading distinction creates lexical ambiguity that has to be solved, it seems attractive to use the features expressing the relevant information as criterion for answering (1): An lu is ambiguous between $1 and $2 iff there is a feature describing the dilference.
If we only take morphological and sylltactic features, many intuitively clear cases of ambiguity (e.g. bank as financial institution vs. as river verge) cannot be expressed. This will lead to problems in transfer or in generation. On the other hand, these features will cause unwanted distinctions as well, e.g. Prench fonctionnair'e (civil servant) with gender masculine or feminine, and kneel with past tense l~neegd or knelt. It makes no sense to try to disambiguate these.
The use of semantic features to define alnbiguity should be rejected for similar reasons. First, we have to determine a fixed set of features a priori, since otherwise no answers to questions of reading distinction evolve. This imposes an artificial upper bound on reading distinction. Moreover, the availability of a certain feature does not mean that it has to be assigned in all cases. We will certainly have a feature expressing the male/fema.le contra.st, but it is not desirable to create two readings of parent accordingly, leading to a translation of pa~r.nts into something meaning mother(s) a~zd/or Jhther~,:).
Alternatively, we couhl argue thal, since translation is the goal, it shouh\] be the criterium \[br reading distinction as well, answering (1): An lu having two senses is ambiguous iff there are diffe.rent translations for the two senses. Leaving aside the non.-trivial problem of determining whether there are different translations, we have to admit that there are cases of exceptional distinctions in one language, e.g. fle'uve vs. r'ivi&c in French, meani ng river ending in a, sea or in another river respectively. "l"hese distinctions will influence all other dictionaries in the system, in the sense that e.g. English 7+iver and Dutch r'ivier be-. come anabiguous, and there are two translat;ion rules between them. If we restrict our attention to a limited grou I) of languages, e.g. the languages in the system, the system becomes difficult to extend, siace adding a new language from outside this group will affect all existing dictionaries. Otherwise there is a conceptual problem, since it will ,rover be possible to decide that an lu is vague, unless we know al\] languages of the world. Instead, cases of exceptional distinctions, bilinqual ambiguities, are best ihandled in transfer between the two languages, because they really are translation pro blems..
Summarizing, taking the means (features) or the goal (translation) as a criterium for read-. ing distinction results in decisions that cause various practical problems and are intuitively incorrect. Furthermore these strategies detach the notion of reading from meaning, which is theoretically undesirable.
Taking only intuitions as our guide will link reading to meaning, but if even trained intuitions or lexicographers do not prevent inconsistencies, as can be seen in many published dictionaries, there is not much hope of reaching consistency, unless we manage to find some support for the intuitions.
A semantic method The tests I will propose here to decide on reading distinction are based on monolingual meaning, and yield a substantially greater degree of consistency than direct, ui~.aided intuitions. It is based on the following conjecture. (2) There is a set of processes P, that, given
a single occurrence of an lu, can stretch
the actual meaning of the 1,l in the con-
text to the boundaries of the reading
the lu has, but not beyond.
In order to be able to check the results, we will first consider some tests where specific processes in P are used, as a.pplied to some intuitiveh, clear cases of ambiguity (e.g. ba~dv" as financial institution vs. river verge) and vagueness (e.g. elephant as Indian elephant vs. African elephant). Then the scope of these tests will be expanded to other cases.
A well-known test evolving from (2) is based on conjunction. Lakoff (1970) proposed a. test where anaphoric so in the second clause of a conjunction refers back to an antecedent containing the lu :for which the question of reading distinction arises, ms in (3). (3) a. John went to a bank this morning,
and so did Mary.
b. John saw an elephant, and so did
Mary. The question to be asked in this cast is whether tile sentence is semantically normal when tile anaphor is interpreted in the other sense than its antecedent. Clearly, (3a) is strange in this interpretation, whereas (3b) is normal, confirming that bank but not elephant is ambiguous in the relevant way (cf. Cruse (1986) on the use of semantic normality judgements). Other anaphors, e.g. one, there can be used as well. The answers are more reliable in case of an antecedent contmning less lexical material outside the lu in question. In (3a), the antecedent of so is go to a bank, and ambiguity might be claimed to arise from the verb. Using one instead of so takes away this possibility, which is especially relevant in less clear cases.
Other processes use quantifiers. One, based on Wiggins (1971), uses universal quantification. It is exemplified in (4). (4) a. All banks in this town are safe.
b. All elephants in th.is zoo are old. The question to be asked here is whether all X can be interpreted as all $1 or all ,%., or only as all $1 and all $2. Whereas (4a) can mean either that there is no danger of flooding or that bank-robbers are effectively discouraged, and it is odd when used to mean both, (4b) can only be used to predicate over both African and Indian elephants in the zoo that they are old. A variant using negation in the same way is discussed by Kempson & Cormack (1981).
A slightly different test can be pertbrmed with a universal quantifier somewhat remote from the relevant lu, as in (5). (5) a. Every town has a bank.
b. Every zoo ha~s an elephant. The question to be asked here is whether the X (bank/elephant) has to be interpreted in the same sense for every Y (town/zoo). In a similar way numerals can be used as in (6), and coordination as in (7), requiring the same question. (6) a. This town has two banks.
b. This zoo has two elephants. (7) a. John and Mary went to a bank this
morning.
b. John and Mary saw an elephant this
morning.
Summarizing, there are three main classes of processes in P behaving as in (2). The first one refers to two elements from the extension of the lu, one of them by an anaphor, as in (3). The second one refers to the flfll extension of a reading at once, as in (4). The third one refers to a group of elements in the extension, exploiting distributivity, as in (5)-(7). Each class is associated with a difl'erent question the answer of which determines whether an analysis as ambiguity or as vagueness is correct. There are various realizations of test sentences for each class, some of which are subject to independently motivated constraints. In a natural way an intuitively appealing definition of reading evolves as in (8). (8) A reading of an lu is a coherent group of
senses, the boundaries of which ('aunot
be crossed by a single occurrence of the
lu without losing semantic normality.
In the previous section, semantic tests were shown to give correct answers in cases where we can check them. This proves that we should not immediately reject the tests. The reason we need them however, is that there are many cases wtmre unaided intuition is not sufficiently determinate, so that conflicts on the correct an,~tysls might arise.
A well-known problem area is the al~alysis of privative oppositions, where one of the senses is more general and includes the other one. Both dog and lion have senses animal belonging to a particular species of mammals and male specimen of that species. According to Kempson (1980) they are both vague with respect to these senses, but Zwicky ~: Sadock (1975) claim that dog but not lion is ambiguous. Applying various tests to them we get the following sentences. (9) a. John has a dog, and Mary ihas one
too.
b. The zoo has a lion, and the circus
has one too. (lo) a. All dogs of this breed are short-
sighted.
b. All lions in this wild reserve have
been killed by poachers. (11) a. This family has two dogs.
b. This zoo has two lions. The sentences (9) and (1.1) cannot lead to a. conclusion for independent reasons. Since an individual or a group in the more specific sense of the lu is also an individual or a group in the general sense, the general sense is always available to cover up the opposition. This is not the case when the flfll extensions are compared, however. Therefore from (10) we can in-deed conclude that dog is ambiguous and lion is not. Both (10a) and (10b) have the general interpretation, but only (10a) also has the more specific one (cf .... , b,tt not the bitches vs. *..., but not the lionesses).
Another problem that comes up is the construction of test sentences for other syntactic categories than nouns. Although the various processes are most easily demonstrated with nouns, nothing in the theory refers to nouns directly. VP-anaphors, e.g. so, can also be used for w~rbs. (12) a. John has been running 'all day, and
so has his washing machine.
b. John has been running all day, and
so has his dog.
(13) John followed Mary, and Bill (lid so
too. The sentences in (12) show the ambiguity of run between tile senses with a human and a machine subject, and the vagueness between senses with a biped and a quadruped subject. For transitive verbs, such as follow, having the sense understand and go after, the result of the test is more disputable, since (13) shows the ambiguity of follow Mary, and one could argue that it is due to ambiguity of Mary, e.g. between the senses thinking person and spatial object. Therefore, the use of a non-lexical anaphor, indicated by # in the examples, is to be preferred. (14) John followed Mary, and Bill # Kate.
It is rather difficult to construct a sentence with a quantifier over the verb comparable to (4) for nouns. Rather, a sentence such as (15) below displays the same distributivity effect as (5), that can also be achieved by coordination as in (16). (15) All boys followed Mary. (16) John and Bill followed Mary.
Tile test sentences for verbs can also be used for adjectives, if they are used predicatively. An example is (17), where black is shown to have different readings when used with a concrete object and with humour. (17) Her dress is black, and so is her hun~lour. For gradable adjectives, a comparison is a basis for constructing a test sentence. Although (17) can be used humoristica.lly, (118) below, illustrating the ambiguity of fair, ca:n hardly be interpreted. (18) Her hair is as fair as the salary she pa~ys
her employees. In general it seems that for gradable adjectives comparison provokes stronger judgements than anaphoric reference by so. In some cases, howew,,r, one of the senses cannot be used predicatively, and neither of the two processes can be used. An empty anaphor sometimes provides a solution, as in (19), where economic is shown to be ambiguous between the senses relati~g to the economy and not wasteful. (19) For Inany years, he produced economic
theories ~nd # cars. In sorne languages, there is a lexical anaphor that requires an adjective as its antecedent, e.g. dito in Dutch, as illustrated in (20). (20) Bij hun gouden bruiloft kregen
ze een dito horloge. (Litt.) 'At their golden wedding got
they a # watch'
Among the remaining problems is the comparison of two senses with big syntactic differences. All test sentences have to be syntactically correct, and syntax does not allow e.g. coordination of a noun and a verb in corresponding positions, tn such cases, tile semantic part of testing the senses is never a.rrived at.
In this paper, I developed tests to answer the question whether an lu with two senses is to be analyzed as ambiguous or vague with respect to them from the semantic conjecture (2). The tests allow for theoretically wellfounded and consistent decisions in many cases. In MT, they determine a proper balance on the cline between what can easily be disambiguated monolingually, and what is useful as a distinction in translation. As such they define the target for monolingual disambiguation, and the class of bilingual ambiguities, that should be treated in transfer. Since the MT environment has only been used in the argumentation, not in the solution proposed, theoretical wellfoundedness and consistency evolving from the tests presented here are equally valid in other environments where a monolingual dictionary is used. References (;ruse, D.A. (1986). Lexical Semantics, Cambridge University Press. Kempson, Ruth (1980). ~ Ambiguity and Word Meaning', in: Greenbaum, Sidney, Geoffrey Leech & Jan Svartvik, St'udics in English Linguistics, Longman, London / New York, p. 716. Kempson, Ruth & Annabel Cormack (1981). 'Ambiguity and Quantification', Linguistics and Philosophy ~, p. 259-309. Lakoff, George (1970). 'A Note on Vagueness and Ambiguity', Linguistic lnquiry i, p. 357359. Wiggins, David (1971). 'On sentence-sense, word-sense and difference of word-sense. Towards a philosophical theory of dictionaries.' In: Steinberg, Danny ,~ Leon Jakobovits (ed.). Semantics, Cambridge University Press, p. 1434. Zwicky, Arnold & Jerrold Sadock (1975). ~Ambiguity tests and how to fail them', in: Kimball, John (ed.). Syntax and semantics 1, Academic Press, p. 1-36. 166