Scott DeLancey

LSA Summer Institute, UC Santa Barbara, 2001

Lecture 2





   Lexical Categories


All grammars leak.  Functionalist grammars leak like sieves.


A basic empirical fact about language is that morphemes can be sorted into categories according to their syntactic behavior:


It is taken to be a truism, an "absolute universal" in Greenberg's sense of a "design feature of language" in Hockett's sense, that all natural language utterances are made up of distinct units that are "meaningful" and that all natural language systems divide those units into a series of two or more classes or SYNTACTIC CATEGORIES.  In fact, it would be safe to say that the nature of syntactic categories is at the very heart of grammar.  (Croft 1991:36)


The words child and write can each occur in a range of positions in an English sentence.  There are many thousands of words with essentially the same privileges of occurrence as child, and many thousands with essentially the same potential distribution as write.  But there is very little overlap between the range of child and that of write.

This suggests a neat and simple model of syntactic structure consisting of a defined set of lexical categories and a set of rules which define the range of occurrence of each of them, that is, a phrase-structure grammar.  The construction of such a grammar would, in principle, be a simple matter of identifying the lexical categories by their syntactic behavior, and writing a set of formulas which generate these combinatorial patterns.  But what seems so simple in principle turns out to be impossible for any actual language--and the reasons for this impossibility are of fundamental importance to our understanding of language.



Defining categories


There are three kinds of definition which are given for lexical categories like noun and verb (cf. Croft 1991, Payne 1999:142-3).  American (and European) structuralists relied entirely on structural definitions, i.e. the definition of a category is the set of behaviors shared by the members of the category (Payne's Type 1):

32.Def.  The positions in which a form occurs are its functions.[1]

Thus the word John and the phrase the man have the functions of 'actor', 'goal', 'predicate noun', 'goal of preposition', and so on.

33. Def.  All forms having the same functions constitute a form-class.


37. Def.  A form-class of words is a word-class.

(Bloomfield 1926/1957:29)


Generative theory offers another category of "explanation", the a priori explanation (Payne's Type 2):


The question of substantive representation in the case of grammatical formatives and the category symbols is, in effect, the traditional question of universal grammar.  I shall assume that these elements too are selected from a fixed universal vocabulary, although this assumption will actually have no significant effect on any of the descriptive material to be presented.  (Chomsky 1965:65-6, emphasis added)


That is, the categories are simply stipulated by the theory; the linguist's task includes identifying them, but there is no need to define them.  In current Generative theory categories are defined in terms of syntactic "distinctive features", e.g. "N, "V; while in theory these may be taken as simply stipulated by Universal Grammar, in practice they are identified by syntactic behaviors, even if these behaviors may be regarded as "tests" for the presence of an a priori category rather than defining qualities of an inductive one.

Radically different in form and spirit are definitions in terms of the function of a category (Payne's Type 3), like the traditional "person, place or thing," or the definition of noun and verb in terms of "time stability" (Givón 1984), actual (Croft 1991) or potential referentiality (Hopper and Thompson 1984), or different types of hypothesized conceptual representations (Langacker 1987).  For generations breath and ink have been expended arguing about which of these two is the "right" kind of definition (which particular definition is the best is, of course, a separate question):

Some grammarians, feeling the failure of such [functional] definitions as those just given have been led to despair of solving the difficulty by the method of examining the meaning of words belonging to the various classes:  and therefore maintain that the only criterion should be the form of words. (Jespersen 1924:60)


In fact, most of us regularly spend time trying to convince our beginning linguistics students of the superiority of structural definitions over the traditional functional one.  But there is no logically necessary conflict between the two types of definition, which do very different kinds of work.

Structural definitions are diagnostic--they allow us to identify a noun, verb, etc., when we see one.  And the most persistent and important argument raised against the legitimacy of functional definitions is that, without exception, they are spectacularly unable to do this in any non-circular way--the only evidence for a claim that 'fire' is a thing, and 'burn' an event, is that fire is a noun, and burn a verb.  What functional definitions are is explanatory--once we discover, through structural analysis, that a language has--or that many or even all languages have--a particular category, a functional definition of the category is an attempt to provide an account of why languages might have it.  And, just as the fatal weakness of functional definitions is that they are not operationalizable, so the traditional and inescapable criticism of purely structural definitions is precisely that they are not capable of providing such an explanation.

But surely we need to be able both to identify categories and to explain their existence.  If all nouns have a certain set of behaviors in common, we can hardly claim to have an explanatory linguistic theory without an account of why those particular behaviors cluster together.  Of course structural analysis must come first--there's no point in trying to explain the facts before we know what they are--but just as obviously it is only the first step in constructing an explanatory theory of lexical categories.  This is not an issue for those theoreticians who explicitly eschew explanation of the sort that we are interested in.



Structural categories


If linguistic categories are anything, they are at least categories.  That is, they are characterizable in terms of linguistic properties common to their members.  Any linguistic form, from morpheme up, has an internal structure, and a set of possible higher-order structures in which it can occur.  A linguistic category is defined by those structural and combinatorial properties which its members share.  Thus any morphological or syntactic construction or process constitutes a feature which is part of the definition of each of the categories to which it refers.  Let us refer to shared characteristics as syntactic properties of the category which shares them.  Thus grammatical number, possessive inflection, and eligibility for subject, object, or prepositional argument status are among the syntactic properties of nouns in English.  A major methodological innovation of Generative Grammar has been the development of more sophisticated methods of syntactic analysis which permit more, and more subtle, generalizations to be discovered.

If we approach the problem inductively, any generalization about a language which refers to some subset of the morphemes or words of the language thereby defines a class of morphemes or words.  Just as in phonology, our expectation will be that each such class is a natural class, i.e. can be characterized by some motivated syntactic property independently of the particular generalization which defines it (cf. Jackendoff 1977:31).  If in some language we can define a particular category by the fact that its members, and no other words, inflect for tense, we assume that there is something about the members of that category, which distinguishes them from all other words, which makes them an appropriate locus for tense marking.  And just as in phonology, where we find rules which define to non-natural classes, it is a result of diachronic processes which have obscured the motivation for what was once a natural generalization.  Just as in phonology, the most natural classes are those defined by the largest number of and/or the most basic (however that may be determined) generalizations.

But it is quickly evident that syntactic properties must be hierarchicized somehow--that some characterize more basic categories than others.  So, we speak of categories and subcategories.  For example, all true nouns in English share certain fundamental behaviors, in particular, the ability to head a NP.  But within that category, mass and count nouns are distinguished as subcategories by the set of possible determiners occurring with the noun when not inflected for plural:  the/a/some child, the/some/*a mud.  Relator nouns like top, back, front, place, behalf, are similarly distinguished by lacking morphological noun properties--in particular, they do not inflect:


1)     on Suzie and Fred's behalf/behalves


2)     on behalf/*behalves of Suzie and Fred


But more significantly, they are distinguished by the fact that they can head only a very specific NP structure, with an obligatory modifying PP and no other dependents.  Thus they do not occur with other NP components, except for their characteristic dependent PP: 

3)     I will be there in her rather difficult place.


4)     *I will be there in rather difficult place of her.



Among the external combinatorial properties which characterize a category, we can distinguish between mention in "basic" and derived constructions, essentially equivalent to old-fashioned kernel and transformed sentences.  For example, Preposition in English has two syntactic properties common to all its members:  occurrence directly before a NP, and occurrence in sentence-final position--but the latter is possible only in the derived preposition-stranding topicalization construction.   English Auxiliary, on the other hand, is defined solely by the latter kind of property:  Auxiliaries are those words which participate in a specified way in negative and question constructions.  Other than this they have nothing whatever in common.  Have and be conjugate irregularly, the modals not at all.  The modals take bare-infinitive complements, like make, let, come and go.  Progressive be, like like, etc.  takes an -ing-complement, while have and passive be take a past participle.



Nouns and Verbs


Nouns and verbs are regularly cited as the universal word classes by authors of every era and most persuasions, even the most resolutely empiricist:


A major form-class distinction reminiscent of "noun" versus "verb" is universal, though not always at the same size-level.  (Hockett 1963:23)


(Cf. Sapir 1921:126, Vendryes 1925:117, and many of the authors discussed below, inter alia; for some of the older and more recent history of these concepts see Robins 1952, Hopper and Thompson 1984, Croft 1991).  The only significant doubts about the universality of these categories has arisen in connection with the analysis of certain languages from the Northwest of North America.  As will turn out to be the case many times in our this course, this controversy turns out to be an issue of theoretical preconceptions rather than of substantive fact.



Nouns and Verbs as Universal Categories


The question of the universality of Verb and Noun, like similar issues which will come up later, is in part a matter of how we define the categories.  In a very basic sense, the universality of Noun and Verb follows directly from the universality of predicate-argument structure.  In every language there are constructions consisting of, at least, a predicate and one or more arguments.  Predicates and arguments have different morphosyntactic behaviors.  These behaviors are, then, diagnostics for Verb- and Noun-hood.  Thus if predicate and argument are universal functions, then Verb and Noun are universal categories.

This line of argument is an old one, though earlier generations made more of the difference between predicate nominals and other predicates as the fundamental and universal diagnostic:


The distinction between verb and noun, which is not always apparent in an English or Chinese word standing alone, is revealed as soon as the word is placed in a sentence; it is not a question of form but of use.  In other words, we must go back to the formation of the verbal image, where the elements of the parts of speech are combined, in order to justify the distinction between verb and noun.  Although there are languages where the noun and verb have no distinct forms, all languages are at one in distinguishing the substantive from the verbal sentence.  (Vendryes 1925:120)


So there is no serious issue of the universality of noun and verb functions.  But in most languages of the world, there are certain stems that can only be nouns, and others that can only be verbs.  In some languages, like English, there are many stems which can serve both functions, while in others there may be few or none.  But even in English, which in cross-linguistic context is quite promiscuous in this respect, there are limitless numbers of purely nominal (child, realty, lizard, prairie, measles) and purely verbal (write, ask, engage, agree, pray) stems.  So the universals question, properly asked, is whether a grammatical distinction between noun and verb words is universal.

In purely structural terms, the only meaningful interpretation of this question is, do all languages have separate noun and verb lexicons, or are there languages which have only an undifferentiated lexicon of lexical stems which can serve either function at need?  We can avoid the question of whether, in a language which has no grammatical distinction, there will not still be stems which, because of their meaning, are more likely to be used as arguments, and others which are more likely to occur as predicates.  And we can thus defer the issue of the relevance of such statistical facts to syntactic theory--but it will be back soon.

The universality of a lexical distinction between noun and verb has been challenged, on the basis of data from languages of the Northwest Coast of North America--most famously Nootkan--where stems do not appear to be intrinsically specified as Verb or Noun (see Jacobsen 1979 for a history of the issue).  Any stem can be inflected as, and have the syntactic function of, either category (exx. from Sapir and Swadesh 1939, cited from Jacobsen 1979:87):


5)     wa»a:k-ma qu:?as     'a man goes'

go-INDIC  man


6)     qu:?as-ma           'he is a man'



7)     ?i:-ma:            'he is large'



8)     wa»a:k-ma ?i:      'a large one goes'

go-INDIC  large


Here we see the stem qu:?as 'man', which we would expect to be a noun stem, occurring as such in ex. (5), but inflected with the verbal indicative suffix -ma and used as a predicate in (6).  And, conversely, ?i: 'large', which we would expect to be a predicate, occurs as an inflected verb in (7), but as an uninflected noun stem in (8).[2]

As has often been pointed out (at least since Robins 1952; see Jacobsen 1979 for further citations and discussion), it remains the case in Nootkan and other Wakashan languages, as well as in other Northwest languages with similar grammars, that any word in use can be easily identified as to its category (cf. Hockett 1963:4).  A Wakashan stem either is inflected as a verb, or it isn't.  Jacobsen (1979) shows, quite unsurprisingly, that nouns and verbs--i.e. actual words--in Nootkan are equally distinguishable by their syntactic behaviors.  In this respect, Nootkan is no more a counterexample to the universality of Verb and Noun than is the productive process of zero derivation in English:


It is, however, very important to remark that even if round and love and a great many other English words belong to more than one word-class, this is true of the isolated form only:  in each separate case in which the word is used in actual speech it belongs definitely to one class and to no other.  (Jespersen 1924:62)


But this is no more than where we began--Verb and Noun functions, i.e. predicate and argument, are universal.

But Jacobsen also elegantly demonstrates that a more sophisticated syntactic analysis does show actual morphosyntactic differences between a set of stems which function as arguments with no morphological marking, and stems which require nominalizing or other derivational morphology in order to be arguments.  In other words, there is a set of lexical stems in Nootkan which naturally occur as arguments of predicates, and another set which are formally marked in that function.  That is to say, a set of nouns and a set of verbs.

Thus, the widespread belief in the universality of noun and verb as lexical categories holds up empirically--as far as we know, there actually are syntactically distinguishable lexical categories of noun and verb in every human language.  If we are not willing to accept this fact as some how simply "stipulated" by a "theory" (which actually says nothing more than the original proposition, i.e. that nouns and verbs are universal), then we need to consider the question of why particular categories are such a fundamental part of language.



What are Nouns and Verbs?


The most obvious, and popular, explanation for the universality of a noun/verb distinction is that the two lexical categories label cognitively distinct types of concept.  For example, Givón (1979, 1984) motivates the existence of Nouns, Verbs, and, where they occur, Adjectives, in terms of a scale of "time-stability":


Experiences--or phenomenological clusters--which stay relatively stable over time, i.e. those which over repeated scans appear to be roughly "the same", tend to be lexicalized in human language as nouns.  The most prototypical nouns are those denoting concrete, physical, compact entities made out of durable, solid matter, such as 'rock', 'tree', 'dog', 'person' etc. ...

At the other extreme of the lexical-phenomenological scale, one finds experiential clusters denoting rapid changes in the state of the universe.  These are prototypically events or actions, and languages tend to lexicalize them as verbs.  (Givón 1984:51-2, emphasis original)


Some syntactic properties of the word classes fall out from Givón's account--why verbs but not nouns can have tense, for example.  And it allows an intuitively satisfying interpretation of many examples of category shift.  Consider, for example, the nominal and verbal uses of the English stem mother.  In our construal of the world, once a mother, always a mother, so that while the state of motherhood has an onset, it has no end.[3]  So, by Givón's account the word which refers to such an individual should be a noun, and it is.  Mothering, on the other hand, is an activity which some individuals (not all or only actual mothers) engage in from time to time, but, by the very nature of human existence, it cannot be continuous or "time-stable"--you simply cannot be mothering someone while you are sleeping, for example.

But the time-stability analysis does not seem to be up to the task of explaining everything which there is to explain about nounhood and verbhood.  Most conspicuously, it does not offer a ready explanation for the most fundamental, defining behavior of verbs and nouns--their function as predicates and arguments.  Moreover, if this is the only motivation for the existence of nouns and verbs, it would seem to predict considerably more gradience between categories than we actually see.  In fact the division is usually quite sharp:  nouns with ephemeral referents, like spark or fit, are not in any respect more verb-like in their behavioral properties than other, more "time-stable" nouns, nor are verbs like endure (or adjectives like eternal) characterized by any noun-like properties (cf. Newmeyer 2000).  When, as is quite common, we find syntactic gradience in the membership of either category, it is between one or the other and some derivative category such as adjective or adposition; I will discuss examples of this sort in the next lecture.  And, finally, as easy as it is to find examples like mother where the difference between the noun and verb uses nicely exemplifies the time-stability concept, it is not in the least difficult to find examples which don't--the verb and noun love, for example, don't seem to me to show any difference in time stability.

Langacker (1987a, b) grounds the noun/verb distinction in a conceptual distinction between THINGS and RELATIONS.  This formulation is instantly compatible with the argument/predicate distinction, and indeed sounds as though it could be directly based on it.  But Langacker intends these categories to have direct conceptual content.  THINGhood is easily identified in "the conceptualization of a physical object involv[ing] some reference to the continuous spatial extension of its material substance" (1987a:63).  The cognitive scanning process which identifies such continuous spatial extension can also be applied to more abstract domains, so that the concept of nounhood is fundamentally based on the structure of actual perception.

There is ample psychological evidence that human cognition distinguishes between object representations and event representations at every level, from perception to memory.  One of the major contributions of Gestalt psychology has been the understanding that recognizing objects is a fundamental characteristic of perception, rather than something derived through experience (Köhler 1929).  As Miller and Johnson-Laird put it, "The most compelling fact of perception is that people see objects" (1976:39), that is, bounded regions of the visual perceptual field which are interpreted as coherent objects.  This is precisely the characterization of noun concepts presented by Langacker (cf. Jackendoff).

People also perceive events:


... sensory systems demonstrate an acute sensitivity to change, as if change carried information of great biological significance.  Sensitivity to change, and a conservative tendency to attribute changes to intelligible sources, is characteristic of the perceptual system at every level of its functioning. (Miller and Johnson-Laird 1976:79)


This sounds very like Givón's "time-stability" dimension, located where it belongs, in perception and cognition.

Langacker is very explicit that THINGhood and RELATIONhood, like other conceptual categories, are matters of construal, not of intrinsic qualities:


If nominal predications crucially involve interconnections, what distinguished them from relational predications?  The essential difference, I maintain, is that a relation predication puts the interconnections in profile (rather than simply presupposing them as part of the base).  The distinction between a nominal and a relational predication does not necessarily imply any difference in the inventory or the organization of constituent events, but only in their relative prominence.  (Langacker 1987b:215)


Once again, modern Functionalism consists in large part of rediscovering and refining the wisdom of our elders:


The so-called parts of speech are distinctions among words based not upon the nature of the objects to which they refer, but upon the mode of their presentation.  Thus the name of anything presented as a thing is a 'noun', and the name of anything presented as an action or ... as a process, is a 'verb'.  In the verb to cage, reference is made to the thing called a cage, but it is not presented as a thing but as an action.  In the noun assassination reference is made to an action, but it is not presented as an action but as a thing.

(Alan H. Gardiner, The Theory of Speech and Language (1932), cited in Jespersen 1933:11)


Hopper and Thompson (1984) carry this question considerably further, discussing in detail the fact that a form--a clause, for example--may be treated as a noun to varying degrees, depending, in effect, on how much THINGiness the speaker needs to imbue it with in order to organize her utterances within a discourse (cp. Givón 1980).  Our present concern, however, is only with a priori nounhood; nominalization we will have to save for later.

The main point of this section is that intelligent interpretations of the notional basis of noun and verb, induced from the analysis of linguistic structure and behavior, lead us toward a conception which closely matches standard psychological models of perception and memory.  This is reassuring; and the most obvious explanation for why it should be is that the psychological phenomena directly inform the linguistic structures.  But, by themselves, the psychological models do not directly motivate all the relevant linguistic facts.  Many indubitable nouns denote concepts which cannot appear in any physical perceptual field, and therefore must be nouns for some reason other than the perceptual structure of their referents--in English, think of anger, help, honor, music.  Here, for the first of a number of times, we see another recurrent issue in functional explanation.  In many domains of grammar we can show a clear motivation, independently establishable on psychological grounds, for that subset of "basic" uses of a construction which have concrete, physical reference.  But all constructions can be, and regularly are, extended to refer to abstract domains which are based on the physical.  There is ample evidence emerging from the study of semantics within Cognitive Grammar for the metaphorical structure of human thought, with abstract domains always grounded in concrete physical domains (Lakoff and Johnson 1980, Lakoff 1987, 1990).  The essential mystery of language is not where how grammar is motivated, which is often reasonably self-evident.  The mystery is the mystery of human thought--how the categories which we have for thinking about the physical world are extended into the abstract realm.





It has long been argued that Verb and Noun are the only universal categories:


No language wholly fails to distinguish noun and verb, though in particular cases the nature of the distinction may be an elusive one.  It is different with the other parts of speech.  Not one of them is imperatively required for the life of language.  (Sapir 1921:126)


Pursuing this process of elimination, we end by leaving intact only two "parts of speech", the noun and the verb.  The other parts of speech all fall within these two fundamental classes.  (Vendryes 1925:117)


The other major categories--Adjective, Adposition, Adverb--frequently originate from Verbs or Nouns, and/or acquire new members by recruitment Verbs or Nouns.[4]  Diachronic sources for other categories, e.g. complementizers or subordinators, include Verbs and Nouns, as well as Adpositions or Adverbs.  Since these latter categories ultimately trace to Verbs and Nouns, Verb and Noun, the only universal categories, are also the likely diachronic source for all other categories.  (I am here speaking about the diachronic renewal and replenishment of categories, which is a constant and, indeed, synchronic process, not about the ultimate origin of categories in glottogenesis--although the answer is no doubt the same, speculation about language origins can't bear directly on the subject matter of this course).  We will discuss adpositions at length in subsequent lectures, and touch on adjectives again later on in connection with questions of constituent order; my primary aim in this section is to document the claim that the adjective category is not universal, and introduce in a preliminary way some of the implications of this.  To discuss this issue, we need separate ways of referring to adjectival function and adjectival form.  I will adopt Thompson's (1988) useful phrase property concept word to refer to the functional category of concepts which in adjective-forming languages turn up as adjectives, and the adjective to refer to a syntactically distinct category of such words.



Verbal and Nominal Property Concept Words


It is well-known that there are languages with no distinct identifiable adjective category (Dixon 1977/1981).  Dixon, and many authors since, have documented the existence of a number of such languages.  To my knowledge, however, no one has documented a language with no discernable subcategory of property concept words.  In most languages without adjectives (or, what seems to be equally common, with a very small, closed class of distinctly adjectival forms), the property concept words occur as a subcategory of verbs and/or nouns.

To take an example from the recent literature, Prasithrathsint (2000) shows neatly that in Thai there is no distinct adjective category; there are no significant syntactic criteria which can be invoked to distinguish the semantically stative verb dii 'good' from the semantically active yím 'smile'.  Exx. (9-10) show dii 'good' functioning as a predicate and as a modifier within an NP; (11-12) show yím 'smile' in exactly the same constructions:


9)     khaw dii

s/he good

'S/he is good.'


10)     khon   dii  ch]]p phom

person good like  I(masc)

'Good people like me.'


11)     khaw yim

s/he smile

'S/he smiles/is smiling.'


12)     khon   yim   ch]]p phom    

person smile like  I(masc)

'Smiling people like me.'


Now, it is often the case that property concept words constitute an identifiable subcategory of verbs.  In Mandarin, as in Thai, there is no distinct adjective class; property concept words negate, inflect for aspect, and in other respects behave as verbs:


13)     ta  pao-le

3rd run-PERF

'S/he ran off.'


14)     ta  gao-le

3rd tall-PERF

'S/he got tall.'


15)     ta  bu-pao

3rd NEG-run

'S/he doesn't run, isn't running.'


16)     ta  bu-gao

3rd NEG-tall

'S/he's not tall.'


But there are still stigmata by which property concept verbs can be distinguished from the rest of the verb category.  For example, only they occur with certain intensifiers or degree adverbs such as hen 'very', zwei 'the most', and -ji-le 'surpassingly:


17)     ta  hen  gao

3rd very tall

'S/he's very tall.'


18)     *ta hen pao


And only the property concept verbs can occur in the comparative construction:

19)     ta  bi   wo gao

3rd than I  tall

'S/he's taller than me.'


20)     *ta bi wo pao


In fact, by a combination of these criteria we can distinguish a property concept subcategory of verbs in Thai as well.  We cannot replicate in Thai examples like (17-18), since Thai allows the same intensifiers with all verbs:


21)     khaw dii maak

3rd good much

'S/he's very good.'


22)     khaw yim   maak

3rd  smile much

'S/he smiles a lot.'


But ordinary verbs can occur in the comparative construction only with an intensifying or other qualifying adverbial, while property concept verbs occur in this construction only without:


23)     khaw dii  kua  phom

3rd  good than I(masc)

'S/he's better than me.'


24)     *khaw yim kua phom


25)     khaw yim   maak kua  phom

3rd  smile much than I(masc)

'S/he smiles more than me.'


And the intensifier l@@y has the same distribution, i.e. it can occur with ordinary verbs only with maak or some other qualifying adverbial:


26)     khaw dii  l@@y

3rd  good indeed

'S/he's really good!'


27)     *khaw yim l@@y


28)     khaw yim   maak l@@y

3rd  smile much indeed

'S/he smiles a whole lot!'


Clearly the basis for these differences is semantic.  Ordinary verbs do not name a quality, but an activity which can be thought of as exhibiting a number of difference qualities.  Thus sentences like (20, 24, or 27) are inherently vague as to the quality which is being compared or intensified, and in these languages they are ungrammatical unless the quality in question is specified.  In contrast, the property  concepts which are lexicalized as adjectives in a language like English are intrinsically gradable (Givón 1970)--they name a quality, and thus when they occur in a comparative or intensifying construction, there is no vagueness about what quality is being compared or intensified. 

The recognition of some degree of underlying commonality between adjectives and verbs has precedents in modern linguistics (Lakoff 1970, Chomsky 1970).  Recognition of parallelisms between the noun and adjective categories is considerably older and more deeply entrenched in Western linguistics, because of the strong morphological similarities and apparent common origin of the two categories in Indo-European:[5]


The adjective again, is often very poorly distinguished from the substantive.  In the Indo-European languages both appear to have sprung from a common origin, and, in many cases, to have preserved an identical form ... Substantives and adjectives are interchangeable in this way in all languages, and, from a grammatical point of view, there is no clear-cut boundary between them.  They may both be grouped together in a single category, that of the noun.  (Vendryes 1925:117)


And the similarities remain strong enough in English to inspire observations down to the present:


It will emerge that there are many rules which generalize across supercategories of N and A, and many to V and P ... Since the combination of N and P [i.e. behaviors which these categories have in common] is so rare, and the combination of V and A at least equally rare, we will feel justified in provisionally accepting (3.4a) as the major division of lexical categories.  (Jackendoff 1977:31, emphasis added)


This claim of a special relationship between nouns and adjectives is reasonable as long as we are analyzing languages like English and other European languages, where there is a strong historical connection between these categories.  But, as we have already seen, it cannot be universal, as there are well-known languages where property concept terms are verbs, not nouns.



Adjective as a Functional Sink


When we find evidence for a category regularly developing out of, or recruiting new members from, more than one source, we are looking at what I will call a functional sink--that is, a function which is important enough, cross-linguistically, that in language which does not formally express it with dedicated grammatical machinery, any construction or lexical means which expresses a related function is a likely candidate for grammaticalization.  In the case of adjectives, it seems that noun modification is such a functional sink.  Human beings describe things--and therefore there is a constant need in human discourse to find ways to describe nouns as possessing certain properties not inherent in their meaning.  It is not necessary that a language have a distinct syntactic category designed for this function; languages have ways of marking nouns and verbs as modifiers of nouns, i.e. genitive marking and relativization--which in some languages are the same thing (DeLancey 1999).  But the function is always there, and frequent and easily-accessible constructions which can be used to express it are automatic candidates for routinization.  Like most of what I will be presenting in this course, this is hardly a new idea:


This brief survey has shown us that though the formal distinction between substantive and adjective is not marked with equal clearness in all the languages considered, there is still a tendency to make such a distinction.  It is also easy to show that where the two classes are distinguished, the distribution of the words is always essentially the same: words denoting such ideas as stone, tree, knife, woman are everywhere substantives, and words for big, old, bright, grey are everywhere adjectives.  This agreement makes it highly probable that the distinction cannot be purely accidental:  it must have some intrinsic reason, some logical or psychological ("notional") foundation ... (Jespersen 1924:74).


This is exactly Dixon's conclusion:


We suggest that the lexical items of a language fall into a number of 'semantic types' ... the division into types can be justified in terms of the syntactic/morphological properties of the members of each type; in addition, a non-disjunctive definition can be given for the overall semantic content of each type.  These types are almost certainly linguistic universals.  By this we mean that each languages has the same array of types, with more-or-less the same overall semantic contexts; however, the morphological/syntactic properties associated with particular types will vary from language to language, and must be learnt for each individual language.  (Dixon 1977:25)


With Jespersen and Dixon (and Givón 1984, Thompson 1988, and others) I am arguing that, while adjectival structure, as Dixon has shown, is not in any useful sense universal, adjectival function is.  And further, that even in a language in which 'bright' and 'gray' are formally nouns, we may expect to see them functioning as noun modifiers more often than sister nouns such as 'tree' or 'knife'.[6]

While this is reminiscent of our conclusions concerning noun and verb, the result is not exactly the same.  Predicate and argument function, and to that extent verb and noun, are formally distinct in any language.  But, since both verbs and nouns can serve as modifiers within a NP, there is not an inescapable need for a distinct modifying construction for a distinct adjectival category.  This theme, of universal function with non-universal syntactic realization, is one to which we will return often.



Problems for a Theory of Minor Categories


One issue which is peculiar to Generative theory is the question of how many, and what, lexical categories there are (Jackendoff 1977:2).  This was not an issue for American Structuralists, who in their descriptive practice were happy to identify whatever and however many different form-classes might be required by the data of a given language, and who had no particular expectation that the inventory of categories in one language should be like that of another.  And it is not an issue for functional theory, which makes no claim that it is possible even in theory to exhaustively list all of the functions which could possibly be grammaticalized.  But Generative theory assumes that the categories of every language are drawn from a fixed set (or, put otherwise, are defined in terms of a fixed set of syntactic distinctive features) defined by Universal Grammar.

While it is hard to pick just one empirical inadequacy from a body of doctrine as reckless and empirically irresponsible as Generative theory, it could be well argued that right here is the most prominent and vulnerable empirical Achilles' heel of the formalist enterprise.  Surely anyone who has tried to develop an informal account of the actual syntactically distinguishable categories of some significant part of as many as two or three languages will quickly conclude that a "fixed" set of possible categories is an impossibility.



Categories of one


In the first lecture, we talked about the anomalous case of English better--a category of one, whose set of defining syntactic properties are shared with no other form in the language.  Let's now look at a less familiar example of the same sort, involving a fairly basic and universal functional category--the comitative marker in Klamath, a nearly extinct (as of this writing I know of one living fluent speaker) Plateau Penutian language of southern Oregon.

Klamath marks the comitative relation with a form dola:[7]


29)     doscambli          hoot sa  ?at,

dos    ‑cn'  ‑ebli hood sa  ?at‑along‑back that 3pl now


sqel   c'asgaayas   dola.

sqel   c'asgaay‑'as dola

Marten Weasel  ‑OBJ with


... now they ran back, Marten together with Weasel.  (Barker 1963b: 10:127)


Dola usually[8] takes object case in any nominal which can express it, such as the human c'asgaay 'Weasel' (a myth character).  It has no obvious categorial assignment in Klamath.[9]  According to X' theory, since it governs case it must be either a verb or an adposition.  Both of these are well-attested among comitative markers across languages.  But Klamath (as we will see later) has no adposition category, unless dola is it.  And its form makes it highly likely that it is etymologically a verb, so we might consider the verbal analysis first.  Dola has the form of a verb in the simple indicative tense, and its syntactic behavior is in many ways what one would expect of a Klamath verb.  The order of a Klamath verb and its arguments is quite free (Underriner 1996).  The same is true of dola and its object.  It usually follows its argument, as in ex. (29), but it can also precede it:


30)     q'ay honk s?aywakta      kakni           

q'ay honk s?aywg‑otn‑a   RE  ‑ka     ‑ni 

NEG  HONK know  ‑on ‑IND DIST‑someone‑ADJ


hoot sa  kat     dola honks. 

hood sa  ka ‑t   dola honk‑s 

that 3pl who‑REF with DEM‑OBJ


... those who were with him did not know that.  (Barker 1963b 10:108)


In this example the relative kat is a subject form, and the demonstrative honks is an object form, and is thus the argument of dola.

However, dola can at best be a highly defective verb, as it only ever occurs in that form--that is, out of the efflorescent inflectional and derivational possibilities of the Klamath verb (DeLancey 1991), the hypothetical *dol- stem uses only one.  And it is not even the most likely one--Klamath does not serialize finite verbs, and the synchronically expected form for a subordinated verb would be the non-occurring *dolank.  Moreover, its syntactic behaviors also include some which are not consistent with verbal status.  We do not ordinarily find sequences of finite verbs within a Klamath sentence, but dola frequently occurs following the verb gena 'go', with no overt nominal argument:


31)     coy  honk ?at hok sn'eweeck'a     c'osak

coy  honk ?at hok sn'eweec'‑'aak' c'osak

then HONK now HOK woman    ‑DIM   always


gena         dola,  gankankca

gV‑e_n  ‑a   dola   gan ‑okang ‑a 

go‑hence‑IND with   hunt‑around‑IND


'Now then that little girl always went with [him], went hunting.' (Barker 1963b, 4:69)


32)     coy  sa   naanok gen‑a   dola,

coy  sa   naanok gV‑en‑a dola

then they all    go      with


kat     ?aysis dola swecandam@n‑a    

ka-t    ?aysis dola swe-cn'-damn-a

REL‑REF Aisis  with gamble‑while.going‑HAB‑INDIC


'They all went with [him], gambling with Aisis on the way.'  (Gatschet XXX)


And ex. (33) casts further doubt on the synchronic identification of dola as a verb, as there is no productive construction in Klamath of a finite verb followed by the copula gi:[10]


33)     hoot hok dola gi, sqel'am'c'as. 

hood hok dola gi  sqel  ‑?m'c‑'as

that HOK with be  Marten‑AUG ‑OBJ


'He was together with, Old Marten'. (Barker 1963b 10:92)


So dola is not synchronically a verb, though it undoubtedly was one once.  And there is no reason to call it an adposition, given that it doesn't need to be adjacent to, or even to have, an argument.  Of course, there are no other adpositions to compare it to, so we don't really know how adpositions behave (or, would behave) in Klamath.  And it could well be that, had Klamath survived, dola was destined to be the entering wedge for the development of a new, innovative postposition category.  But we can hardly maintain that in its attested form it has already grammaticalized to that extent.

So what dola is is one more example, like better, of a categorially unique form--a working part of the language which does not fit into any larger category.  Now, if every language had just one of these, what does that imply about the "universal set" of categories from which languages get to draw their own?  And more fundamentally--why would a phrase-structure grammar have such things?  Why would it--how could it--allow such things?



Universal categories?  The adposition story


The secondary nature of the Adposition category has been long noted (e.g. Vendryes 1925:164-5), and its universality strongly called into doubt by the demonstration of its diachronic connection to relator noun and serial verb constructions (Givón 1979, Mallinson and Blake 1981:388-9, Heine and Reh 1984:241-4, Starosta 1985, Bybee 1988, Aristar 1991, DeLancey 1994, Harris 2000).  Still, even in languages like Chinese or Akan, where characteristic adpositional functions are carried out by a set of more-or-less grammaticalized verbs, there are typically a few members of the set that are so thoroughly grammaticalized that they can no longer be categorized as verbs, and might as well be considered to represent a distinct category of prepositions.  And the fact that a category may regularly draw new recruits from other categories is not by itself an argument that the category cannot be universal.

Just so as to lay this particular issue to rest, I want to describe a language--Klamath--which simply lacks the category altogether, and carries out the typical adpositional functions by quite different means.  The primary function typically associated with adpositions--specification of location or path--is expressed in Klamath by a set of "locative-directive stems", which occur in what have been called "bipartite" complex verb stems (DeLancey 1991, 1996, 1999, to appear).[11]  The most numerous type of bipartite stem, and the one relevant to our present concerns, consists of a lexical "prefix" and a locative-directive stem (LDS):[12]


34)            on top       in water          underneath


living object:     ksawal-        ksew-          ksodiil‑


round object:     lawal‑         lew‑           lodiil‑


long object:     ?awal‑         ?ew‑           ?odiil‑


In stems of this type the lexical prefix is a classify­ing element referring to a category of object; the final element, the LDS, describes a motion, location or path of that object.  These stems are indifferently stative, eventive intransitive, or transitive, accord­ing to context; thus ex. (35) could refer to a dog sitting in water, running into the water, or being given a bath:


35)     wac'aak ?a  ks-ew-a

dog     IND living.obj-in.water-INDIC

'dog is/goes/is put in(to) water'


When the clause has a distinct NP corresponding to the path or location indicated by the LDS, this is marked with the locative case suffix |dat|:[13]


36)     coy  honk    naanok Gees cewam'c‑     am 

then DEM.OBJ all    ipos Old Antelope‑GEN


?i-Gog          a     mna‑tant       y'agi‑ dat

pl.-in.container‑INDIC 3sPOSS‑OBL.LOC bas­ket‑LOC


'Then [she] put all Antelope's ipos into her basket.' (Gatschet XXX)

37)     s?as?abam'c qtan‑ a   ks-elwy‑ank               

Old.Grizzly sleep‑IND‑HAVING





'Old Grizzly slept, lying by the fire.' (Gatschet XXX)


In the English glosses for these examples, the prepositions into and by encode both the abstract relational concept LOCATION and more specific lexical information describing the precise spatial relation predicated between the THEME and the LOCATION.  In Klamath, LOCATION is expressed by the case suffix {dat}, and the lexical information (not, obviously, exactly the same as that expressed by any particular English word) in the LDS.

I have argued above against the identification of the comitative marker dola as an adposition.  That aside, unless one wants to start grabbing odd particles at random in order to find content for an a priori category, there are simply no plausible candidates in the language for adposition status.  Of the functional categories commonly expressed by adpositions, benefactive is indicated by an LDS.  In ex. (38), the benefactive suffix {oy}, here surfacing as -ii-, adds a benefactive argument to the verb (note the object marking on tobaks 'man's sister'):


38)     coy  mna    tobaksa          slambli:ya       

     coy  mna    tobaks      ‑a   sla_n‑ebli‑i: ‑a 

     then 3sPOSS man's.sister‑OBJ mat  ‑back‑BEN‑IND


      Naykst'ant                loloqs.

      Nay   ‑ksi‑t  ‑y'e:n'‑t   loloGs

      beside‑LOC‑LOC‑NOMZ  ‑LOC fire


'[He] laid down a bed for his sister on one side of the fire ...' (Barker 1963b 4:15)


Instrumental, like locative, is marked with a case suffix, and also in some cases by "instrumental prefixes" in the verb.  In (39), the instrumental prefix s- 'sharp instrument' provides some information about the nature of the instrument, while -tga marks the instrumental noun 'knife':


39)     hohasdapga                       deqiistga

RE  ‑s_e ‑s         ‑dV ‑obg‑a   deqiis-tga

DIST‑REFL‑sharp.inst‑hit‑DUR‑IND knife‑INST


'[They] stabbed one another with knives.'


At least in this area of the grammar the difference between Klamath and more familiar languages like English does not necessarily reflect any fundamental difference in the conceptualization of motion and location.  The difference seems to be essentially typological.  The Klamath LDS and the English preposition category are in many ways quite comparable, in terms of semantic function and range, numbers, and degree of openness of the class.  The essential difference between the languages is that in English--a "configurational" language if ever there was one--these forms form a constituent with the NP which is their semantic argument, while in the quasi-polysynthetic Klamath they incorporate in the verb.



Aristar, Anthony. 1991. On diachronic sources and synchronic pattern:  An investigation into the origin of linguistic universals.  Language. 67:1-33.

Bloomfield, Leonard. 1926.  A set of postulates for the science of language.  Language 2,153-64.  repr. in M. Joos, ed., Readings in Linguistics I.  Chicago: University of Chicago Press, 1957.

Bybee, Joan. 1988. The diachronic dimension in explanation.  in Hawkins, ed., pp. 351-79.

Chomsky, Noam. 1965. Aspects of the theory of syntax.  Cambridge, MA: MIT Press.

     . 1970. Remarks on nominalization.  in Jacobs and Rosenbaum, eds., pp. 184-221.

Croft, William. 1991.  Syntactic categories and grammatical relations.  Chicago: University of Chicago Press.

DeLancey, Scott. 1991. Chronological strata of suffix classes in the Klamath verb.  International Journal of American Linguistics 57:426-445.

     . 1994. Grammaticalization and linguistic theory.  Proceedings of the 1993 Mid-America Linguistics Conference and Conference on Siouan/Caddoan Languages, pp. 1-22.  Boulder:  Dept. of Linguistics, University of Colorado.

     . 1996.  Penutian in the bipartite stem belt:  Disentangling areal and genetic correspondences.  Proceedings of the Twenty-Second Annual Meeting of the Berkeley Linguistics Society:  Special Session on Historical Topics in Native American Languages, pp. 37-54.

     .  1997.  Bipartite verbs in western North America. presented at the Second Meeting of the Association for Linguistic Typology (ALT II), Eugene, September 11, 1997.

     .  1999.  Lexical prefixes and the bipartite stem construction in Klamath.  International Journal of American Linguistics 65.1.56-83.

     . to appear.  Argument structure of Klamath bipartite stems.  26th Annual Meeting of the Berkeley Linguistics Society:  Special Session:  Syntax and Semantics of the Indigenous Languages of the Americas.

Dixon, R.M.W. 1977. Where have all the adjectives gone?  Studies in Language 1:19-80. (repr. (1981) in Dixon, Where have all the adjectives gone? and other essays in semantics and syntax.  NY: Mou­ton).

Givón, T. 1979. On understanding grammar.  New York: Academic Press.

     . 1980. The Binding Hierarchy and the typology of complements.  Studies in Language 4:333-77.

     . 1984. Syntax:  A functional-typological introduction.  Amsterdam: Benjamins.

Hakulinen, Lauri. 1961. The structure and development of the Finnish language. (Indiana University Publications in Uralic and Altaic Series, vol. 3).  Bloomington: Indiana University Publications.

Harris, Alice. 2000. Word order harmonies and word order change in Georgian. in Rosanna Sornicola, Erich Poppe, and Ariel Shisha-Halevy, eds., Stability, variation and change of word-order patterns over time.  Amsterdam: Benjamins.

Hawkins, John, ed., Explaining language universals.  Oxford: Blackwell.

Heine, Bernd, and Mechtild Reh. 1984. Grammaticalization and reanalysis in African languages.  Hamburg: Helmut Buske.

Hockett, Charles. 1963.  The problem of universals in language.  pp. 1-29 in J. Greenberg, ed., Universals of language.  Cambridge: MIT Press.

Hopper, Paul, and S.A. Thompson.  1984.  The discourse basis for lexical categories in Universal Grammar.  Language 60:703-52.

     , and      . 1985.  The iconicity of the universal categories 'noun' and 'verb'.  pp. 151-86 in John Haiman, ed., Iconicity in syntax.  Amsterdam: John Benjamins.

Jackendoff, Ray. 1977.  X' Syntax:  A study of phrase structure.  Cambridge, MA: MIT Press.

Jacobs, Roderick, and Peter Rosenbaum, eds. 1970. Readings in English Transformational Grammar.  Waltham, MA: Ginn.

Jacobsen, William. 1979.  Noun and verb in Nootkan.  in B. Efrat, ed., The Victoria Conference on Northwestern Languages (British Columbia Provincial Museum Heritage Record No. 4), pp. 83-155.

     . 1980. Washo bipartite verb stems.  American Indian and Indoeuropean Studies:  Papers in honor of Madison S. Beeler, ed. by K. Klar et. al., 85-99.

Jespersen, Otto. 1924.  The philosopy of grammar.  London: Allen & Unwin; New York: Holt & Company.

     . 1933. The System of Grammar.  XXX

Köhler, Wolfgang. 1929. Gestalt Psychology. New York: H. Liveright.

Lakoff, George. 1970. Irregularity in syntax. XXX

     . 1987. Women, fire, and dangerous things.  Chicago: University of Chicago Press.

     . 1990.  The Invariance Hypothesis:  is abstract reason based on image-schemas?  Cognitive Linguistics 1:39-74.

Langdon, Margaret. 1990. Morphosyntax and problems of reconstruction in Yuman and Hokan.  Linguistic Change and Reconstruction Methodology, ed. by Philip Baldi, 57-72. (Trends in Linguistics, Studies and Monographs: 45).  New York: Mouton de Gruyter.

Langacker, Ronald. 1987a.  Nouns and verbs.  Language 63:53-94.

     . 1987b. Foundations of Cognitive Grammar.  Stanford: Stanford University Press.

Mallinson, Graham, and Barry Blake. 1981. Language typology: Cross-linguistic studies in syntax.  Amsterdam: North-Holland.

Miller, George, and Philip Johnson-Laird.  1976.  Language and perception.  Cambridge: Harvard University Press.

Noonan, Michael. 1999. Non-structuralist syntax.  in M. Darnell et. al., eds., 11-31.

Payne, Doris. 1999.  What counts as explanation?  A functionalist approach to word order.  in M. Darnell et. al., eds., pp. 137-65.

Prasithrathsint, Amara. 2000.  Adjectives as verbs in Thai.  Linguistic Typology 4.2:251-71.

Robins, R.H 1952. Noun and verb in universal grammar.  Language 28:289-98.

Sapir, Edward. 1921. Language: an introduction fo the study of speech.  New York: Harcourt, Brace.

Starosta, Stan. 1985. Relator nouns as a source of case inflection. in V. Acson and R. Leed, eds., For Gordon Fairbanks, pp. 111‑33. Honolulu: U. Hawaii Press.

Talmy, Leonard. 1972.  Semantic structures in English and Atsugewi.  Ph.D. dissertation, University of California, Berkeley.

Thompson, Sandra A. 1988.  A discourse approach to the cross-linguistic category 'Adjective'. in Hawkins, ed., pp. 167-85.

Underriner, Janne. 1996. Nonconfigurationality in Kla­math. MA Thesis, Department of Linguistics, University of Oregon.

Vendryes, J.  1925.  Language: A linguistic introduction to history.  New York: Alfred Knopf.

     [1]Bloomfield is using the word function in a different sense than ours.  In the older sense used by Bloomfield, function refers to syntactic function, e.g. as subject or object (Bloomfield's "actor" and "goal"), etc.  Thus his function is equivalent to our syntactic property (see below); I use function in a sense closer to Bloomfield's class meaning.

     [2]Note that these data also contradict Vendryes' claim that "all languages are at one in distinguishing the substantive from the verbal sentence".

     [3]Even with the end of the mother--in a very real sense Eleanor of Aquitaine will always be the mother of Henry, Geoffrey, Richard, and John Lackland (and a dozen or so more whose names I can't remember), even though none of them are any longer, or ever will be again, physical objects.

     [4]I am not speaking in glottogenetic terms here.  The point is not that somehow in the evolution of language other categories emerged from primordial nouns and verbs, but that all other categories originate and are regularly renewed by recruitment from the basic major categories.

     [5]And in Uralic:


It is known that in the Indo-European group the adjective was not differentiated from the noun until comparatively recently and the same appears to be true of Finno-Ugric.  (Hakulinen 1961:50)

     [6]The fact that in such a language, all nominal modifiers--both ordinary and property concept nouns--will be marked as genitive should not confuse the issue.

     [7]All Klamath forms are written in the practical orthography adopted by the Department of Culture and Heritage of the Klamath Tribes.  The orthography is essentially Barker's (1963a) phonemic orthography with a few self-evident typographical changes.  Examples taken from Barker's Klamath Texts (1963b) are cited with text and sentence number, i.e. 4:69 is sentence (69) in text #4.  Examples from other sources are cited with page numbers.

     [8]But not, for example, in the second clause of ex. (32)

     [9]Barker assigns it to his "residue" category, and labels it only as an "enclitic".

     [10]Barker's transcription of a comma in this sentence implies that sqel'am'c'as is an afterthought.  Note that it is still in object case.

     [11]There is a number of languages of this general type in western North America (DeLancey 1996, 1997, see also Talmy 1972, Jacobsen 1980, Langdon 1990); I don't know whether any of them show evidence of a distinct adpositional category.

     [12]In Talmy's (1985) analysis of the isomorphic structure in the nearby Hokan language Atsugewi, the "lexical prefix" is an initial verb stem which lexicalizes the shape of a THEME, and the LDS's are called satellites.  The differences between this and the bipartite stem analysis are irrelevant to the present argument.

     [13]The underlined |d| in Barker's morphophonemic representation indicates an underlying /d/ which assimilates to any preceding consonant.

As suggested by Talmy (1985), these morphemes--at least in Atsugewi and Klamath--are probably not shape classifiers of instrumental arguments, as sometimes assumed, but action classifiers reflecting a characteristic type of motion.  In a non-mechanical technology the use of particular types of implement will be characteristically associated with particular body movements.  Even so, this category of verbal element does provide information about the instrument, in the same way that LDS's do about the LOCATION.

North American languages show a strong tendency to combine a great deal of grammatical material with the verb in a single phonological word, which we may take as a (thoroughly informal) definition of a polysynthetic language.  Klamath is polysynthetic by this definition, though not by others.