08ma-raw [⌂ EvenYrbk]

08ma, PDF
Categorial ambiguity 1

Lajos Marosán

Categorial ambiguity: alternative analyses

Introduction There are some words in English as well as in other languages which can be — and, in some grammars, are — analysed as belonging to more than one word category or subcategory of one word class. For instance, the English words this, all, yesterday, damp are such. In a traditional analysis it is suggested that the phoneme sequence this represents actually three different syntactic words: (a) a demonstrative determiner, as in This book was expensive; (b) a demonstrative pronoun: This was expensive; © adverb: The book was this thick. The word yesterday may also be multiply ambiguous as far as its category membership is concerned between a proper noun: Yesterday was the 15th of February, adjective: I’m her yesterday man and adverb: It was snowing heavily yesterday readings. The number of words showing similar characteristics could be multiplied. What problems do such examples raise? First, a simple question that occurs is this: how many words does the phonological form this represent? Second, what grammatical property can this variety of distribution be ascribed to while many words of the vocabulary are restricted to appearance in one particular sentence position? Thirdly, how does meaning influence the distribution of these elements if at all? In the following pages I will be discussing these issues. How many syntactic words, then, does the phonological word this stand for? How shall we account for this phenomenon? We might offer the following analyses, which I will dub as the ‘three-word solution’ and the ‘one-word solution’. In the following section I will be discussing the former of the two.

The three-word solution On this view, there are three individual lexical items whose phonological form is the same but belong to three different word classes or word categories: this is classified as a determiner when it is followed by a noun, which, in turn, may be preceded by one or more modifiers, typically, adjectives; in other words, when the word this is followed by a nominal structure and it is a dependent element in a phrase, — a noun phrase — which, in turn, may appear as various clause The Even Yearbook 8 (2008),Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 1218-8808, http://seas3.elte.hu/delg/publications/even, © 2008, Lajos Marosán

Categorial ambiguity 2

elements; (ie. this expensive book can function as subject, object, predicative complement or prepositional complement). In contrast, the word this is seen as a representative of the pronoun class whenever it functions as subject, object, predicative or prepositional complement on its own; that is, as a pronoun it occurs as an independent clause element, or as a dependent in a prepositional phrase. Further, our word — this — is easily identified as an adverb whenever it combines with an adjective or an adverb. To sum up what has been said so far: these this-words illustrate homonymy. This threefold analysis of one phonological word assumes a formal approach to the establishment of the parts of speech, one that is exclusively based on syntactic distribution. More particularly, a word class is set up on the basis of the combinability potential of its members with words which are thought to belong to other word classes, irrespective of their putative morphological and semantic properties. In our case, a determiner is a morphologically simple word that combines with a noun to form noun phrases, while a pronoun, which is not a dependent element in this sense, is ready to combine with verbs as subject, predicative complement or object, and prepositions. On the other hand, an analysis along these lines (which presupposes a view that connects syntactic positions to word categories) may be related to a further, mostly implicit, presupposition: a clause is actually a series of functionally definable slots. It would not, therefore, be impossible to relate each syntactic slot to a separate functional/positional category: word class or a form class or whatever we would like to call such a broad category. More specifically, if we take this idea seriously, at least, for a few lines, we will establish that the subject-this could be one class along with all the elements — words, phrases and clauses — that are capable to substitute it in that position, while the object-this, the prenominal-this and the post prepositional-this will be members of different functional/positional classes together with items with which they are in paradigmatic relationship. Such an analysis would bring together elements which grammarians insist are distinct: if each syntactic slot cries out for a category of its own which can fill that position adequately, constructions occurring in identical positions, that is, constructions entering into paradigmatic relations should be lumped together as members of the same word or form category — an unwanted result. For further reference let me call such an unwieldy eclectic distributional category broad category. For instance, [This book]/[This]/[That he didn’t appear]/[After eight] was a surprise to us all &c.; the items before was a surprise to us all have the same distribution, are in paradigmatic contrast,

Categorial ambiguity 3

therefore, a view that sees a one-to-one correspondence between syntactic slots and categories are forced to classify these into the same broad category.1 However, grammatical tradition and the related grammatical intuition are very much against this move. A categorisation along these lines would be unwieldy even if it was very much in the spirit — or the letter? — of the original Greek technical term ta meré tou logou. The Greek expression is usually translated as parts of speech where instead of the word speech, the word clause would have been a much better choice. In other words, the Greek term means elements of the clause, which is obviously ambiguous between what is now recognised as categorial and functional analysis. The advantages of the three word solution are fairly obvious: it accounts for all appearances of the word in distinct syntactic and lexical terms, and also helps to establish a parallel with words that are restricted to the same syntactic position, and, therefore, are classified in the same way. More particularly, the determiner this patterns with, for instance, the articles (a, the), possessive determiners (the my-series), and some quantifiers (every), all of which are restricted to prenominal position; while the pronoun this shows a similar distribution to, for example, full noun phrases (eg. the dog), personal and other pronouns (you, we; anybody; which). Again, the adverb this enters into paradigmatic contrast with very, too. From a practical — educational and lexicographic — point of view the different labelling of the word facilitates sentence construction and explication of grammatical structures. The disadvantages are of a more abstract nature. On the one hand, the threeword solution encodes some important grammatical information, as well as it fits our (or at least, my) intuition that a word that is ready to combine with distinct word categories cannot be the same word in all its appearances. On the other hand, this solution is unable to account for another intuition that these distinct syntactic appearances of this are related by their meaning and are, therefore, occurrences of one and the same word. The second criticism, which closely corresponds to the first, raises a theoretical issue: by suggesting that one phonological word form actually represents three unrelated lexical items, we make one word into three, therefore, first, this solution is unable to lead to a general statement about this and words of similarly flexible distribution and second, violates the principle of economy, which is widely adopted in various fields of research.

1 The idea comes from Kratochvíl (1967), who attempts to establish word classes in Modern Standard Chinese, and his first move is to examine possible paradigmatic contrasts in a simple clause. Similarly, Fries (1952) bases his word classification on syntactic distribution. The Even Yearbook 8 (2008),Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 1218-8808, http://seas3.elte.hu/delg/publications/even, © 2008, Lajos Marosán

Categorial ambiguity 4

The two-word solution An alternative to the three-word solution would be to say that there is only one phonological and syntactic word, which is associated with one particular meaning, and which is capable of occurring in three distinct syntactic positions. This suggestion is motivated by the common phonological form that appears in all these positions and, importantly, by the tacit acceptance of the fact that in the different syntactic positions these words retain the same meaning. A further, more abstract but nonetheless important, motivation comes from an effort to keep the number of lexical items and classes at a minimum. In other words, the one-word solution would cater for the problems that the three-word solution created: it promises to be the intuitively correct solution from both a syntactic and a semantic point of view. However, the acceptance of the one-word solution requires that we assign the word this to a word class which naturally allows for the three syntactic positions enumerated above. ‘Naturally’ here means that the distribution of the word should not be in conflict with that of the other members of the chosen word class. We would want to see some regularity in grammar hoping to find one-to-one and predictable correspondences between lexical categories and syntactic positions/functions. More specifically, if we proposed that this was a, say, pronoun, we would find it more easily explainable and intuitively more natural that it appears together with a noun, on the one hand, and can take up a clausal function on its own, on the other, since a pronoun is a nominal category; it can substitute for a noun phrase; while its distribution as an adverb would be rather exceptional and much in need of explanation. In other words, out of the three appearances of the word two could be explained with reference to intuition and tradition; while in the opposite case, that is, when we chose the adverb as the appropriate word class for this, we should be able to give adequate explanation or exemplification for the determiner and pronominal occurrences of other adverbs. I really wonder whether we would find adverbs which can also appear in the same positions as determiners or pronouns. Now, if we opt for the one-word solution, the problem will be that of lexical representation: how to formally express the decision that there is one phonological word capable of three distinct syntactic occurrences, that is, how to reconcile syntactic distribution and category membership. Let us explicate and illustrate the above speculations. Thus, we might choose to assign the word this to the pronoun class, considering its occurrences in [This book] was expensive and [This] was expensive. Such a decision follows fairly naturally from our grammatical education or more rudely, from our grammatical prejudices since both determiners and pronouns are associated with the nominal category. (For instance, in the Hungarian tradition articles are The Even Yearbook 8 (2008),Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 1218-8808, http://seas3.elte.hu/delg/publications/even, © 2008, Lajos Marosán

Categorial ambiguity 5

analysed as a type of pronoun.) In this case, however, we would find it extremely difficult, if not completely impossible, to analyse the preadjectival occurrence of this (The book I bought was [this] thick) as a pronoun.

Emonds’ 1987 proposal In this section I will be discussing Emonds’ 1987 article on how to analyse word classes in a generative grammar. In this article Emonds advocates the one-word solution, which is based on distribution only, ignoring semantic considerations. His suggestion involves the manipulation of subcategorisation information within the framework of a generative model, which, among other things, purports to keep the number of word categories as low as possible. Emonds builds his theory on the traditional assumption that words naturally fall into major and minor word classes; however, Emonds organizes major and minor classes into a coherent and clever word class system so that each major class has a corresponding minor class that serves as its specifier. For instance, adjectives combine with a restricted set of degree adverbs to form adjective phrases, while the adequate specifier category of the nouns are determiners, which together make up noun phrases. Emonds tries hard to get rid of the traditional ‘undisciplined’ adverb category though his attempts are far more motivated by linguistic considerations than philosophical expectations, that is, by the protean nature of the adverb class than by Ockham’s warning. As is widely known, grammars usually put in the adverb category widely and wildly different items; even the earliest grammars established a category which contained all the words which could not be positively defined, and called this ‘pandektes’, meaning ‘all-receiving’. While it is fairly easy to see and appreciate the difference between slowly, which modifies the verb as in He drove slowly on the slippery road, and fortunately, which expresses the speaker’s attitude to whatever is conveyed by the rest of the sentence as in Fortunately, it wasn’t raining, we are left with a load of words which are neither verb nor sentence adverbs, such as moreover, only, as &c.. To decrease the number of problematic elements which are traditionally recognised as adverbs, Emonds, then, assigns some of them to the adjective class — the ones which clearly modify verbs; others are classed with the prepositions, such as there, then; see further details below. Again some of them are seen as proper nouns: yesterday. His discussion of the question is not exhaustive in that he does not concentrate exclusively on assigning the members of the traditional adverb class into other categories; he tentatively reclassifies some of the category but leaves others alone. Also, the traditional pronoun category is amalgamated into the determiner class. Our problem of the multiple occurrence of this could be examined from this The Even Yearbook 8 (2008),Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 1218-8808, http://seas3.elte.hu/delg/publications/even, © 2008, Lajos Marosán

Categorial ambiguity 6

perspective. More particularly, following Emonds, we could suggest a solution by assigning this to one class only, to the determiner class (ie. the specifier category of the major noun class), and allowing it to have three subcategorisation frames: 1) [_N] (this book); 2) [_0] (this); and finally, 3) [_A] (this thick). Emonds’ solution relies on the observation that complementation is not always obligatory, and crucially, on the assumption that minor classes may also take complements; for example, some verbs and adjectives can optionally take complements, respectively, such as smokeV (some sea salmon), eatV (a ham sandwich); sorryA (about last night), allergicA to the serum. The observation and the ensuing analysis can be extended to less plausible cases: inP (the house), hisD (house) and to data which would — from a traditional point of view — never be associated with complementation. In Emonds’ system these items obligatorily occur without complementation; eg. heD: [NP_], slowlyA: [AP_]. That is, in is a preposition which optionally takes an NP (or some other) complement; in this way it is always a P unlike in some grammars where it is analysed as a preposition when complemented with an NP but classified as an adverb without an NP. Also, the word his is a determiner, which is either followed by a noun (a D in the traditional sense), or stands alone as an NP in a clause — a pronoun in the traditional analysis. In contrast to this, the word he — traditionally a personal pronoun — is a determiner which never takes a complement, and slowly is an adjective which cannot take a complement either. To sum up, in Emonds’ system there are lexical items in all — major and minor — classes which take complements (a) optionally ([PPin (NP)], [VPeat (NP)], [NPthis (N)]); (b) obligatorily ([VPkick NP], [PPto NP], [NPthe N]; and which (obligatorily) take © no complement ([NPwe], [APquickly], [Psoon]).2

Some remarks on terminology As was pointed out in the previous paragraph, Emonds relies on the notion of complement; this term, however, has a looser application in his article than in other grammars where the ability of a lexical item to take complements is restricted to the major classes; for example, in English and in Hungarian a large number of verbs take complements, which are obligatory in most of the cases; for instance, the most unproblematic complements verbs can take are objects; adjectives can take PP or clausal complements &c. More particularly, it seems 2 On this view, then, items such as slow/slowly and my/mine should be treated as one lexeme, respectively, where the syntactic word forms are suggestive of distribution, unlike, for instance, eat which has only one syntactic form for both the transitive and intransitive appearance. The Even Yearbook 8 (2008),Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 1218-8808, http://seas3.elte.hu/delg/publications/even, © 2008, Lajos Marosán

Categorial ambiguity 7

that all Emonds’ word categories — major as well as minor — can take complements, and this indirectly suggests a grammar which does not recognise dependence or subordination relations: the traditional notion of complement assumes that the element that takes the complement governs, or is superordinate to, the item that functions as its complement, which, in turn, is dependent on, or subordinated to, the governor (or head). For instance, in Emonds a determiner takes a noun as complement, and this grammatical characteristic is formally expressed as an instruction: [_N], that is, insert any member of the category D into an environment where it is (obligatorily) followed by a noun. On the other hand, a noun is defined as a word that takes a determiner as complement: [D_]. In other words, these major and minor word class pairs are each other’s complements: which is, then, subordinated to the other? This may look a secondary issue from the perspective of Emonds’ word class system; however, subordination has been a crucial concept in whichever grammar we examine, therefore, doing away with this concept, though indirectly, requires some explanation. As it must have emerged from the above account, Emonds defines a class with reference to another class; in this way he avoids the conflicts grammarians create for themselves by manipulating semantic characteristics word classes are alleged to have. By associating some semantic content to each of the parts of speech or word categories, traditional as well as some modern grammarians attempted to account for the intuition that language has some relationship to the non-linguistic world we perceive and think of. They may have been right but, unfortunately, they opted for a hypothesis that created controversial analyses. Below we will have a chance to say some words about this issue. Following a structuralist approach, which explicitly gives up on relating meaning to structure by associating categories to meaning types, Emonds’ is a self-containing system, in which the elements of analysis are supposed to explain one another, which, thus, leaves no room for the explication of how it is possible for a self-sustaining organization to reach out of its bounds.

Analysis of Emonds’ proposal In this section we will be looking at some specific issues of Emonds’ solution. The following comments are in order. First, we might wonder how important the syntactic labels are that words are marked with. More particularly, if subcategorisation, that is, distributional information is all that is needed to accommodate lexical items in clausal structure, category labels can only serve mnemonic purposes, that is, they do not have the informative weight they have in traditional grammar (cf. Fries (1952)). It appears that it does not really matter The Even Yearbook 8 (2008),Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 1218-8808, http://seas3.elte.hu/delg/publications/even, © 2008, Lajos Marosán

Categorial ambiguity 8

whether an element is labelled as a noun or an adjective if they have overlapping distribution. What emerges is that categories are not associated with a characteristic distribution, only individual lexical items are. In this way Emonds avoids one of the tasks that traditional grammatical analysis sets itself as an aim: the task of harmonizing category and distribution, that is, class and syntactic function, as we saw in a section above. The distributional approach that Emonds advocates shows a tension between individual words and word categories. If whatever is important about words from a syntactic point of view is encoded in subcategorisation frames, which gives information about the combinability potential of individual words, while the category membership of a particular word is not at all important; what is is in what environment(s) a particular word appears, and this environment is defined in terms of category labels. Such a representation enables the grammarian to encode grammatical information relevant for one word without explicit reference to category membership of that particular word, therefore, he does not have to decide on this — sometimes controversial — issue at all. On the other hand, the frames are expressed in terms of word category, that is, the tension is created by the method that a word which is neutral from a categorial point of view is defined in terms of categories. To illustrate. The observation that the definite article combines with nouns is encoded by the frame: the [_N]. The representation does not tell us, and it is not particularly important to know, what category the definite article itself is; what is essential is what words it is ready to combine with, and this information is encoded in terms of category and, naturally, relative position. Secondly, an issue that regularly harasses grammatical analysis is meaning. In both traditional and (some) current grammars the age-old view is advertised that members belonging to a lexical class have the same meaning type — a contention Emonds completely gives up. Whether or not this is the right thing to do I do not know; in any case, Emonds’ idea to consider this as one word is intuitively feasible even if he does not involve meaning in the discussion.

The Hungarian analysis In this section I am discussing the relevant Hungarian analysis — the strategy that is adopted in academic Hungarian grammars. In Hungarian grammars it is customary to distinguish between words of ‘pure word category meaning’ (cf. (1) below) and those of ‘impure word category meaning’ (cf. (2)); I will only concentrate on the relevant aspects of the Hungarian classification: (1) tiszta szófaji jelentés (pure word category meaning) The Even Yearbook 8 (2008),Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 1218-8808, http://seas3.elte.hu/delg/publications/even, © 2008, Lajos Marosán

Categorial ambiguity 9

(2) nem tiszta szófaji jelentés (impure word category meaning) a) keresztezıdı szófajúság (cross category words) b) kettıs szófajúság (double category words) c) átmeneti szófajúság (transitory category words) As I perceive, this belongs to subgroup (2b), which I translate as double category words. This solution divides the problem of homonymous words between lexical representation and syntax: in the lexical representation a double category word bears two category labels, which signals its ability to appear in the syntactic positions associated with those particular word categories. In actual occurrence, however, the word shows the characteristic syntactic and/or morphological as well as semantic features of either one or the other class. For instance, there are numerous Hungarian words labelled as noun/adjective (for example, colour, nationality and material names as well as numbers) since these can occur in two grammatical contexts. Naturally, in one particular context these words are either adjectives or nouns. Eg. Fölvett egy [kékA] blúzt ‘She put on a blue blouse’ Kedvelem [a kéketN] ‘I like the blue (colour)’ [ReggelAdv] nézz ki az ablakon! ‘Look out through the window in the morning’ Egy szép [reggelenN] kinéztem az ablakon ‘I looked through the window on a fine morning’ What view of word classes does this analysis suggest? At first sight, it looks as if it was a syntax-based system. It is not, but we will learn the details of this parts of speech analysis from a wider choice of examples: distribution, morphological properties as well as semantic characteristics are equally taken into consideration; that is, this is what Schachter (1985) calls an eclectic system. Why, for instance, the occurrence of kék in the first example was analysed as an adjective comes from its distribution (viz. its prenominal position) and its meaning: it characterises the noun — adjectives are said to express some characteristic or attribute; while the second appearance of the same word is a noun: nouns combine with case endings (so do adjectives), and the accusative word form kéket refers to an entity (viz. the colour itself) rather than to some characteristic: on this The Even Yearbook 8 (2008),Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 1218-8808, http://seas3.elte.hu/delg/publications/even, © 2008, Lajos Marosán

Categorial ambiguity 10

view of the parts of speech, nouns are attributed ‘entity’— meaning. In other words, it is believed that in a particular context the word does not only acquire the necessary trappings characteristic of the syntactic position it appears in but also its meaning is adjusted; for instance, a double-membership nationality or a colour word will have the ‘entity’ meaning in nominal while the ‘attribute, characteristic’ meaning in the attributive position. Another feature of the analysis is that not all the three criteria can be put to work at the same time: when we identified the word as an adjective, we omitted the morphological criterion as irrelevant (since both adjectives and nouns can combine with the same case suffixes) whereas in the second case, the distributional criterion did not help. Without going into details of the question, I must mention a further interesting feature of this strategy: it distinguishes between lexemes and actual word forms. Category membership characterises the lexemes, not individual word forms but in some cases words belonging to a particular category may occasionally occur as another category. For instance, in [A hangoskodóktólN] elvették az üzenıfüzetet ‘The loud ones were asked for their message booklets’ the word hangoskod-ó+k+tól is a plural (-k-) participial adjective (-ó-) (plus case ending -tól ‘from’), which turns into an occasional noun; this reclassification, (which seems quite regular to me) however, only concerns this actual word form, not the lexeme HANGOSKODÓ. In our case this is always a word form and not a lexeme since it has one phonological form only. (Unless we insist that the phonologically unrelated these is the ‘same’ word.) The reader will remember that in standard introductory textbooks a lexeme is usually defined as an abstract entity comprising all distinct morpho-syntactic forms of a particular word; for instance, dog, dogs represent the lexeme DOG while the Hungarian lexeme KUTYA subsumes the following syntactic words: kutya, kutyák, kutyákat, kutyá(n)knak, kutyaként, kutyául, kutyával, kutyá(k)ból &c. So, a Hungarian grammarian would insist that this is one word with three distinct category labels since the same form may appear in three distinct syntactic environments; morphological analysis is naturally out of the question but the analyst could argue that we have, at least, two different meanings: one which points out and, therefore, refers to an entity, the other which points out a contextually defined degree of a characteristic expressed by the adjective it modifies. It seems that even the most innocent grammatical analysis is marred by the burden of meaning. How is it possible that a double category word shows entity meaning once and the very same word has a different meaning: the meaning of attribute in some other context? We can only answer this question if we assume what Hungarian and The Even Yearbook 8 (2008),Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 1218-8808, http://seas3.elte.hu/delg/publications/even, © 2008, Lajos Marosán

Categorial ambiguity 11

some other grammarians, too, take for granted: word categories — at least, the major ones — are associated with some generic category meaning. From this perspective the question above is absolutely legitimate. We might suggest an answer that polysemy is at work here, which is supposed to be a lexical property of words; a phenomenon which has little to do with syntax: a word has slightly different meanings in different contexts but these meanings are related to each other. I do not think that this type of alternation in meaning is explainable by polysemy. Instead, I would propose that it is syntactic position that imposes the type of generic meaning on words or phrases that some grammarians believe in and which can be activated with a generous amount of introspection; further this generic category meaning is identified as ‘entity, action’ or ‘attribute’ meaning (depending on the category). More specifically, on this view, it transpires that the meaning of, say, kék (‘blue’) as an abstract lexical item is neutral as far as the above mentioned generic meanings are concerned. However, in a particular context where it appears with nominal trappings (that is, in full morphological panoply as well as the characteristic syntactic position), it acquires the ‘entity’ meaning while in a context in which the same word turns up as an adjective, the ‘attribute’ meaning is — so the speak — activated. In other words, I think it is a particular syntactic slot that can be associated with a particular meaning not the actual word form, never mind what the actual lexical meaning the word has. The details of this proposal, however, require elaboration.

Meaning and syntactic position In this section I would like to examine a question that every now and then appears in traditional grammars, and which I think is related to the above discussion. The question is whether the meaning of individual lexical items corresponds to the syntactic positions they can appear in, and if the answer is yes, in what way. In other words, on this view it is the meaning that determines the class membership of a word and consequently, its syntactic potential. Morphologically simple, that is, structurally unanalysable words, such as this, have no explicit markers that would assign them to some word category and, consequently, to some syntactic position(s). Thus, if such words are formally unmarked, in virtue of what do they have the distribution they have? The distribution of words which have explicit category markers are (or could be) explained by these markers themselves. Some grammarians have thought that the syntactic potentials of unmarked or morphologically simple words should reside elsewhere, in some not formally obvious feature. Therefore, the idea that the syntactic behaviour of such morphologically unmarked words is governed by their meaning may naturally The Even Yearbook 8 (2008),Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 1218-8808, http://seas3.elte.hu/delg/publications/even, © 2008, Lajos Marosán

Categorial ambiguity 12

occur. More specifically, if we, then, start out from the hypothesis that in all the three occurrences of this the word has the same meaning, then, it would not be unreasonable to ask whether the meaning of the word has any role in its distribution. In other words, is it not the case that this appears in these three syntactic positions because its meaning is required for and compatible with, the meaning of other words it can combine with? This last question, however, points up a conflict. There are actually two aspects to consider. On the one hand, we mentioned that the three syntactic occurrences of this are made possible by the fact that the meaning of the word is compatible with the words it combines with. On the other hand, one cannot help wondering how it is possible that three syntactic slots are compatible with one meaning, in other words, how it is possible that the same meaning can be linked to three distinct clausal functions. These questions are only relevant from the point of view that meaning determines syntactic function. In the following paragraphs I will be discussing the problems just raised. First, I will be looking at the issue from a wider perspective examining what Kuryłowicz had to say in 1935, and second, I will discuss a more modest version of the same idea.

Syntactic function derives from meaning Let us see the first problem, then. Above I mentioned that the idea that syntactic function of a word derives from its meaning is a topic which often, though mostly superficially, appears in traditional grammars. I would like to summarize Kuryłowicz’s position which is the best elaborated of the texts I am familiar with. Though this article was written several decades ago, the same views appear time and again in more current literature; see, for instance, Telegdi (1961), Hadrovics (1969), Lengyel (2000). Kuryłowicz (1935) discusses quite a few topics in his article; I am concentrating on what is relevant from the point of view of this discussion. Kuryłowicz insists that linguistic elements — words and phrases — get inserted into syntactic positions due to their meanings: he claims the reason that, for instance, a noun or a noun phrase is able to appear as subject of a clause is closely related to its meaning, which he thinks can be grasped with the help of the traditional Aristotelian logical categories. Thus, a noun, since it expresses ‘entity’ meaning, is ready to function as subject. Basically the same argument is presented for verb forms occurring as predicates and adjectives functioning as modifiers: verbs express ‘activity’, therefore, they occur in predicative function whereas the ‘characteristic’ or ‘attribute’ meaning of adjectives makes them adequate modifiers. However, Kuryłowicz fails to explain two important aspects of the problem. On the one hand, he does not comment on what logical or objectThe Even Yearbook 8 (2008),Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 1218-8808, http://seas3.elte.hu/delg/publications/even, © 2008, Lajos Marosán

Categorial ambiguity 13

ive relationship, if any, holds between the meaning-function pairs, in other words, what necessitates the noun-subject, verb-predicate, adjective-modifier &c. correspondence; or, to put it differently, why the category N and the syntactic function subject pair off, the verb and the predicate and so on. Second, he cannot convincingly accommodate to his theory the observation that the same category may appear in various functions. For instance, a noun, or a noun phrase may occur either as subject, object, predicative complement or prenominal modifier. Kuryłowicz perceives the conflict and makes an attempt to save the theory by proposing that each category has a primary and a secondary syntactic function. Thus, the primary function of a nominal element is to function as subject while its secondary syntactic function is to appear as object. (Note that this article was written in French and most of the examples are also in French, where — as in English — the syntactic form of the subject and object are the same.) As we all know, a research article is also a type of narrative with a logic and internal structure of its own; still, this narrative cohesion does not save Kuryłowicz’s theory. Kuryłowicz addressed a question which merits discussion since I think there are quite a few grammarians entertaining the intuition that the purpose of grammatical structure is to express meaning, and therefore, it is not unreasonable to suppose a directional relationship between the two. However, Kuryłowicz’s attempt, in this form, is obviously unsuccessful: he cannot convince his readers that syntactic function follows from semantic content. A more — so to speak — domesticated variant of this theory, which is doomed to failure in the same way, and which is advocated, for example, by Martinet (1960), claims that only some elements of the vocabulary perform their syntactic functions due to their meaning. In the following paragraphs I will present Martinet’s position. He puts forward the interesting suggestion that the meaning of some words such as mardi, hier ‘(on) Tuesday, yesterday’ and mange, donne ‘eat, give’ also signifies the function these words perform; in particular, both mardi and hier occur as adverbials in sentences, while the two verbs function as predicates by virtue of their meaning. In other words, Martinet tries to argue for a one-to-one correspondence between meaning and function in some cases. In the next passage we turn our attention to how much this tame version of the theory is feasible.

Word categories The establishment of word categories or parts of speech can be examined from two different aspects; we can consider it as a discovery procedure but we might also speculate how much word classes, which are the result of the grammarian’s work, relate to the speakers’ putative internal grammar. We expect the discovery The Even Yearbook 8 (2008),Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 1218-8808, http://seas3.elte.hu/delg/publications/even, © 2008, Lajos Marosán

Categorial ambiguity 14

procedures we apply to reveal various lexical categories. The point of setting up word categories is — among other things — for the linguist to be able to make general statements about the correspondences between words and their syntactic occurrences. It is fairly obvious that explicit morphological as well as distributional differences between word forms, that is, formal features, lead us to the establishment of word categories. Where we do not find such explicit features, we might think — at least, some traditional grammarians did so, and not necessarily unreasonably — that morphologically unmarked words are inserted into syntactic structures on the basis of their meanings; meaning — it transpires from these texts — should be taken in a pretheoretical, intuitive sense. This view, which creates a causal relation between meaning and syntactic function, however, confuses two things: (1) discovery procedures and the results that follow from them, and (2) speakers’ putative mental processes. I personally do not know what happens, what physiological processes occur in our minds when we speak or manipulate language in one way or another, therefore, in what follows I can only rely on my own private speculations. I should think that the association of some meaning to a word form in our minds is just as mysterious as the association of a word to a category on the basis of its formal properties. In other words, just because grammarians are able to find explicit material in words which enables them to assign the words to one or the other class does not mean that the mind recognizes and manipulates lexical items in the same manner. The corollary is that, unless we have positive evidence to the contrary, there is no reason to suppose that speakers insert words into linguistic structures on two different principles: one formal and one semantic. Explicit formal features help the grammarians to establish word classes but we have no evidence that speakers’ knowledge also operates in very much the same way. In fact, we cannot know it for sure whether there are grammatical categories in the native, mental grammar along the lines grammarians would like to see it. In other words, even though we think that our classification of words is based on formal differences, and where we do not find such distinctive characteristics, we turn our attention to meaning, we have no reason to think that our discovery procedures exactly match our mental processes. To sum up the paragraph: in the lines above I exposed the weaknesses of the suggestion proposed by several grammarians that the distribution of some items in the vocabulary of a language can be accounted for with reference to meaning, arguing that this idea is incoherent. These critical remarks, however, need not discourage us from further examination of the suggestion, especially, since, though not very elegant, it would not be impossible to conceive of a grammar in which the syntactic behaviour of words were accounted for by different principles.

Categorial ambiguity 15

Meaning and word category I feel the question raised in the previous paragraph can be looked at from two aspects. One aspect of the question concerns meaning itself, the other the relation between meaning and word category. First, we should determine the dimension of meaning. It seems that the formation of phrases, that is, grammatically and semantically compatible combination of words, is partly determined by the meaning which the constituent words possess. We might argue that a phrase is a well-formed structure because its constituent words are not only syntactically but also semantically compatible: it would be possible to create phrases in which the constituents morphologically and syntactically are compatible, still, we would find them unacceptable due to semantic reasons. On the basis of these observations and speculations, we may conclude that low-level semantics, that is, individual word meanings, does influence syntax. If we believe that this is so, we should try and define or at least characterise how this works: how meaning determines the combinability potential of words. However, it should be emphasized in this context that — if our speculations are adequate — meaning only partly determines the word combinations that we create; the other aspect of acceptability involves grammaticality, which, in turn, means that only syntactically compatible words are able to combine. To put it differently, meaning always occurs in a particular syntactic shape, that is, there is no abstract, linguistically unformed, categorially untouched meaning: a meaning is either expressed as a noun, or an adjective, or an adverb &c.. In other words, we have no reason to state that meaning logically precedes syntactic categories. The question is whether there is a higher level semantics in words which determines the syntactic potential of a particular lexical item has been answered. As mentioned above, some grammarians insist that the meaning of a lexical item determines its distribution; we have seen two versions of this contention: the pervasive view and the partial view. Meaning in this context corresponds to the two different views. On the one hand, meaning is a simple pretheoretical expression: a meaning is what words are associated with, and it is to convey meaning that a language has words for. This naive view of meaning emerges from Martinet’s work, who advocates ‘the partial view’, suggesting that the syntactic potential of only some words is determined by their meaning. The other position, characteristic of the ‘pervasive view’, is more controversial and is associated with what could be called word class meaning. These authors (eg. Kuryłowicz (1935), Telegdi (1961), Hadrovics (1969), Lengyel (2000)) claim that word class membership imposes some meaning on all the words of the vocabulary of a language, which, then, has a decisive role in distribution. The word class meanings are usually identified with Aristotle’s categories, such as ‘entity, The Even Yearbook 8 (2008),Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 1218-8808, http://seas3.elte.hu/delg/publications/even, © 2008, Lajos Marosán

Categorial ambiguity 16

activity, attribute, relation, place &c.’, because it is hoped that in this way a parallel can be drawn between the categories of being and categories of language; this hope is ultimately based on the conviction that human language expresses the non-linguistic reality, therefore, the linguistic and the metaphysical structure of the world must share important features, otherwise language would not be capable to convey information about the world. In contrast to this, the more digestible variant — the partial view — of this ‘meaning-distribution view’ restricts the applicability of this idea to morphologically unanalysable lexical items, claiming that, for instance, some morphologically unmarked adverbs appear in clausal functions they do due to their meaning, and it seems that meaning in this case is simply meaning — not word class meaning. We can assume, hypothesize, be convinced and believe that some morphologically simple words have the syntactic behaviour they have due to their meaning, but, unfortunately, we cannot marshal any semantic or syntactic evidence in favour of this view: we have to decide by fiat. The other, the pervasive view can be defended by suggesting that it is part of a philosophical speculation about language and the world, and if we give up this opinion, word class semantics goes down with it. However, there are less sophisticated arguments against the view that associates word classes and certain meanings. As pointed out above, a parallel is drawn between the category of a word and its high-level semantic properties; more specifically, it is assumed on this view that a noun has nominal while a verb verbal whereas an adjective adjectival meaning. Nominal, verbal and adjectival meanings can be expressed in another way with reference to classical logical categories: nouns denote entities, verbs refer to activities while adjectives identify characteristics or attributes. The problem is that if one claims that, say, a noun appears as subject, it may do so, then, for two reasons: 1. it belongs to the category noun, and nouns may occur eg. as subjects; 2. it has a characteristic meaning type which assigns it to subject position in the clause. We could ask, however, that if we can relate syntactic distribution to category, why would we need the other — semantic — characteristic? If syntactic and semantic characteristics always go together and correspond to each other in a one-to-one fashion — at least, this is what transpires from the relevant texts — how do we know it for sure that it is meaning that is decisive? As far as low-level semantics is concerned, our word this can happily combine with a wide range of nouns and adjectives, and is ready to occur as subject, object or prepositional complement on its own; that is, its meaning (taken in a pretheoretical pedestrian sense) appears compatible with nouns, adjectives, on the one hand, and main clausal elements, on the other. Following Martinet, I would like to say that its meaning determines the syntactic functions it is ready to perform. Let us cast a quick glance at syntactic functions. Subject and object The Even Yearbook 8 (2008),Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 1218-8808, http://seas3.elte.hu/delg/publications/even, © 2008, Lajos Marosán

Categorial ambiguity 17

are clausal functions which we feel we are completely familiar with. Prepositional complement, however, sounds a bit empty, devoid of some real content but, at least, the concept of complement is a familiar one; but the situation gets much worse and more hopeless when we would like to identify and explicate the function that involves the word this appearing with a noun, that is, when it is a determiner; then, we are not better off than when we would like to explicate what meaning is. Does being a determiner involve a function, for instance? Or does this term only stand for a category or for both a function and a category? If it is a function, what is its content? That is, we might accept that the meaning of a word determines its function but if we are not able to define the meaning, neither delimit the function, what can we do, then? Let us try to explicate the problem of the meaning of this in Kuryłowicz’s classical logical terms. Kuryłowicz’s theory, the reader will remember, claims that word classes have characteristic meanings; note that this view is theoretically independent of what terms these meanings are expressed in. The reader will also remember that in § 6. we discussed the Hungarian tradition and saw that in those texts meaning is expressed in terms of the classical Aristotelian logic and word classes are each attributed a meaning type. As already pointed out above, these logical meaning types helped the Hungarian grammarian to establish that kék is an adjective in Fölvett egy [kékA] blúzt (‘She put on a blue blouse’) while the same phonological word is a noun in Kedvelem a [kéketN] (‘I like the blue (colour)’). So, manipulating these logical meaning types is not easy either since it requires that we name first, what category this belongs to, and second, what meaning type this represents and then assign it to, perhaps, a primary, secondary and, finally, tertiary syntactic function(s). That is, what seems to be the case is that once we know what syntactic category a word is a member of we are able to assign a meaning type to it. Let us see, now, how the idea works. There is a claim going back to Medieval grammarians (see Bursill-Hall (1970)) that some derivationally related words have the same meaning. On this view, then, the Hungarian words, say, futV, futóPart, futásN have the same meaning. We can argue that in the following clauses A fiú futott és fellökte az öreg nénit ‘The boy ran and knocked over the old lady’ A futó fiú fellökte az öreg nénit ‘The running boy knocked over the old lady’ A fiú futás közben fellökte az öreg nénit ‘(During) Running the boy knocked over the old lady’ The Even Yearbook 8 (2008),Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 1218-8808, http://seas3.elte.hu/delg/publications/even, © 2008, Lajos Marosán

Categorial ambiguity 18

the highlighted words can be subsumed under the same logical category: activity. Unfortunately, there is another, equally acceptable but diagonally opposite, intuition which suggests that these are three different words: futV, futóPart, futásN, each of which has a meaning of its own though these meanings are similar in one way or another. Telegdi would say, explicating the intuition that these items are very similar, that though their meaning is different, they refer to the same concept; unfortunately, he leaves the question open what relationship there is between concepts and meanings, and the notion of concept is not as self-evident as it looks. The analysis on the basis of the Medieval claim relies on the (implicit) assumption that meaning and word category are — at least, in some cases — two separable entities, while the second suggestion — the one about word class meaning — is more realistic in the sense that it claims that meaning is given to us, or filtered through, embodied in, syntactic categories, that is, we have no chance to see meanings independent of the syntactic categories which words belong to. (This is the idea which, I think, motivates Telegdi to introduce concepts into the discussion.) How does one decide on the number of meanings, then? There is an old and venerable tradition which is applied for identifying the sameness or difference of meaning: the manipulation of truth conditions. On this view two propositions, represented by clauses, are synonymous if the same conditions make them true. Whether by ‘synonymous’ we would want to mean identity or only similarity of meaning or both is again another question. That is, the following two sentences are synonymous on this view; A csákó kiperkálta a mangát a verdáér és elhúzott a gádzsival a rákba (informal) A férfi kifizette az autót és örökre eltávozott a fiatal hölgy társaságában (formal) ‘The man paid for the car and left for good in the company of the young lady’ We might insist that the same conditions make these two sentences true. This analysis, however, depends on the view that a situation is independent of the way we describe it — a view that has little to do with linguistic or semantic analysis. The same method is usually used to find out whether two words are synonymous; in our case, however, the conditions that would render true the expressions: fut, futás, futó would turn out to be truth conditions for fut, rohan, száguld, &c., and the syntactically related forms of these words; that is, while with reference to truth conditions we identify what semanticists call cognitive synonyms, a finegrained identification requires another method; what this method is I do not know. What follows from the above speculations? First, we have no logical or conceptual category that this would fit because the categories are not fine-grained enough. Second, the truth conditional method is, again, not refined enough for us

Categorial ambiguity 19

to be able to adjudicate whether the three syntactic occurrences of the word represent the same meaning.

Summary To wind up the discussion we have to ask what we have achieved. Let us take stock of what we have done. The purpose of this paper was to examine what coherent ways there are to account for the multiple syntactic occurrence of words such as this, yesterday &c., and comment on the possible solutions. I omitted the third logical possibility: the two-word solution in the hope that the reader will be able to make up for the omission after the extensive discussion of the three-word solution. What emerges is that it is a question of strategy whether one opts for either solution: both have merits and a philosophy of grammar which can only be appreciated from inside the system. It has also become clear that the semantic criterion, that is, the examination of meaning, comes out as the losing party. I suppose the reason is that the scholarly study of a problem presupposes and involves the manipulation of technical terms which pin down as well as determine the phenomena it examines. It seems that the phenomenon of meaning cannot be pinned down since each meaning may be taken as a self-referring technical term which cannot be reduced to or subsumed under, some cover terms. In other words, it is not clear whether we should operate with a general definition of meaning, and then decide how meaning and syntactic relation are related or whether individual, concrete meanings of words are relevant in this search. More specifically, we have seen that identifying meanings by some technical term is a problem in general; also, the twofold exploitation of the logical categories is incoherent since the same terms are used to describe generic meanings, on the one hand, abstracted from the actual word class, and word class meaning, on the other. That is, the words fut, futó, futás are identified as expressing activity when we try to contemplate their generic meaning, ignoring their syntactic class; the same words, however, will be assigned different meanings if we look at their category: the verb fut has activity meaning, the noun futás entity meaning while the participle may be both depending on the categorial analysis. This is obviously an impossible situation. Finally, I would like to enumerate the topics which were mentioned in passing but which would deserve a more detailed discussion. One is the hypothesis that a clause is actually a series of slots; it would be interesting to know whether it is possible to enumerate and name these slots, and see what determines their number and their syntactic nature. Second, I suggested that it might be worth taking a quick look at what I called broad categories — construcThe Even Yearbook 8 (2008),Department of English Linguistics, Eötvös Loránd University, Budapest ISSN 1218-8808, http://seas3.elte.hu/delg/publications/even, © 2008, Lajos Marosán

Categorial ambiguity 20

tions that are in paradigmatic relation with respect to a syntactic position. This second issue corresponds to the previous topic in as much as it concerns syntactic positions. Third, the relation between class and syntactic function was also mentioned — a question that is related to how fine-grained we want our lexical categorisation to be. Also, statements on how the parts of speech of a language relate to syntactic functions can be controversial since the establishment of the classes themselves already contain syntactic information. Further, the problem of how polysemy relates to syntax was raised. In semantic literature polysemy is discussed separately from syntax but it might be interesting to find out whether polysemy is not a type of word-level implicature, which implies that what is identified as polysemy is, in fact, very similar to conversational implicature, where various meanings arise dependent on the context of utterance. A final issue which I find an interesting research topic is the question of what a syntax looks like which is based on mixed principles.

References Bursill-Hall, G. L. 1971. Speculative grammars of the Middle Ages. The Doctrine of ‘partes orationis’ of the Modistae. The Hague: Mouton. Emonds, Joseph. 1987. Parts of speech in generative grammar. Linguistic Analysis 17: 3-42. Fries, Charles Carpenter. 1952. The structure of English. London: Longman Hadrovics, László. 1969. A funkcionális magyar mondattan alapjai. Budapest: Akadémiai Kiadó. Kratochvíl, Peter. 1967. Modern Standard Chinese. Lingua 17: 129-152. Kury owicz, Jerzy. 1935. Dérivation lexicale et dérivation syntaxique. Bulletin de la Société de Linguistique de Paris 37: 79-92. Lengyel, Klára. 2000. Az igenevek helye a szófaji rendszerben. Nyelvtudományi Értekezések 146. sz. Budapest: Akadémiai Kiadó. Martinet, André. 1960. Elements of a functional syntax. Word 16: 1-10. Schachter, Paul. 1985. Parts-of-speech systems. In: Timothy Shopen (ed.) Language typology and syntactic description. Clause structure. Cambridge: Cambridge University Press. Vol. I: 3-61. Telegdi, Zsigmond. 1961. Elméletek a szófajok természetér l. In: Sulán Béla (ed.) Nyelvtani tanulmányok. Budapest: Tankönyvkiadó. pp 25-48. Lajos Marosán Department of English Linguistics ELTE marosan.lajos@gmail.com