Christiani Lehmanni inedita, publicanda, publicata
The nature of parts of speech
huius textus situs retis mundialis
dies manuscripti postremum modificati
occasio orationis habitae
Word classes. Nature, typology, computational
representations. Second Triple International Conference.
Università Roma Tre, 24.-26.02.2010
volumen publicationem continens
Sprachtypologie und Universalienforschung 66/2
The nature of parts of speech
Parts of speech have both semantic and structural aspects. The two sets of features are
essentially incommensurate, since the semantic features derive from the functions of language
in communication and cognition, while the structural features are essentially based in the
combinatorial potential of signs in a text. Consequently, the two sets of features are largely
independent of each other. Their combination in a language yields sets of parts of speech whose
systematicity is largely language-internal
To the extent that there is a functional motivation for parts of speech, three restrictions must
be made: 1) It is not, in the first place, a cognitive, but rather a communicative motivation. 2)
The functional motivation of word classes is not direct, but mediated by semantic and syntactic
categories of higher order. 3) Only the primary word classes (verb and noun) are motivated in
this way. The secondary classes (adjectives, adverbs etc.) and the minor word classes (pronouns,
subordinators etc.) increasingly have a system-internal structural rather than a universal
functional motivation. Given these heterogeneous functions and constraints, there is no uniform
nature to all parts of speech.
The problem of the nature of parts of speech may be articulated as the question for the forces which
are responsible for
• the existence of parts of speech in general
• particular parts of speech in different languages
• the assignment of a particular part of speech to a lexeme coding a given meaning.
As we shall see, different factors and motivations are behind these three aspects of the nature of
parts of speech.
On the one hand, there is a common basis to the part-of-speech systems of the languages of the
world; and on the other hand, there is no universal part-of-speech system that was represented in
every language. In this, parts of speech behave just like any other linguistic property of a semiotic
nature, i.e. one that concerns signs or categories of signs: their conformation is an affair of the
particular language as a historical and cultural activity. Such properties are therefore not
preassembled at the universal level. They do, however, obey universal principles since every
language is a system for the solution of a set of cognitive and communicative problems which, at an
appropriate level of abstraction, is the same for all languages and human beings.
1This paper was first presented as a keynote lecture at the Second TRIPLE International Conference on
Word Classes at Università di Roma III, March 24-26, 2010. I thank Raffaele Simone, the participants of the
conference, the Pavia PhD students colloquium, the Erfurt EPPP Sprachbeherrschung, the members of the La
Trobe University Research Centre for Linguistic Typology and two anonymous reviewers for helpful
Christian Lehmann, The nature of parts of speech 2
1.1 Formal constraints vs. cognitive and communicative functions
The language system is a semiotic system. As such, it is the result of the interplay of two essentially
independent forces (cf. Prandi 2004, IX-XVIII):
1. Formal constraints on structure: The constraints on a semiotic system and on the messages
constructed with it are of a heterogenous nature. Laws of logic and information theory
determine how signs may be selected and combined. Laws of physics determine the
composition and transmission of signs. These are complemented by other laws of nature in the
case of semiotic systems used by a particular species, e.g. homo sapiens.
2. Functions of communication and cognition: The world surrounding us which we
conceptualize is in many respects the same for every speech community; and the same goes for
the tasks of communication in such a community. These two domains provide the total of
content and its conveyance in the widest sense.
Thus, entities of grammar, including parts of speech, have a purely formal side determined by the
constraints imposed on any semiotic system. At the same time, this formal side is not empty, but is
laden with cognitive and communicative content. In more concrete terms: Grammatical categories,
relations, constructions and operations are necessary for a semiotic system to operate, and they do
have some purely formal properties. At the same time, those are categories like tense, relations like
the indirect object relation, constructions like the causative construction and operations like
nominalization; and none of these is purely formal, all of them have their semantic side. Putting it
yet another way: in a semiotic system, everything concerning the sign as a whole is meaningful.
The association of form and function in language is not biunique. A classification of semiotic
entities, including grammatical ones, by semantic criteria yields results different from a
classification based on formal criteria. This is true for word classes2 just as for any other
grammatical category. For instance, there is, in English, a distribution class that includes noun
phrases (like a bright girl), proper nouns (like Linda) and certain pronouns, among them personal
pronouns (like she), while it excludes nominals (like bright girl), common nouns (like girl) and
other pronouns (like one; cf. a bright one with *a bright she). The members of that distribution
class have no common semantic basis that would not also be shared by other kinds of nominal
elements. And on the other hand, a semantic criterion such as denoting an act would subsume
members of different word classes such as ask and question.
The double-sidedness of word classes has many methodological consequences. Two are of
immediate relevance here: First, definitions of word classes – just as of any other grammatical
category – are mixed definitions, combining semantic and structural criteria. Second, any analysis
of word classes aiming at understanding their nature has to take a double approach to them, a formal
and a functional approach. In §3, we will take the formal approach, and in §4, the functional
1.2 Interlingual word-class concepts
Grammatical concepts, including parts of speech, may be defined at different levels of generality.
The two levels that are of interest here are the language-specific and the interlingual (alias cross-
linguistic alias typological) level. These are levels of abstraction. Thus, the English perfect has
certain particular properties that it may not share with the perfect of any other language. It
2See §2 for the conceptual relation between ‘part of speech’ and ‘word class’.
Christian Lehmann, The nature of parts of speech 3
nevertheless instantiates an interlingual category of perfect, a concept which must be sufficiently
abstract and prototypical in nature in order to fulfill its methodological function of serving in the
description and comparison of more than one language.3
Now there is a difference between a single linguistic sign such as a lexeme or a particular tense
or case formative, on the one hand, and a category of signs such as the word class ‘adjective’ or the
paradigm of tense or case, on the other. The single language sign has a particular significatum
which, though general it may be, has its own specificity. The meaning of the category, however, is
what all of its members have in common semantically. The larger and more heterogeneous the
category, the more elusive becomes the attempt to identify a set of semantic features they all have in
Parts of speech of different languages are different; however, the extreme structuralist position
according to which they have no common interlingual basis4 is untenable. The English and the
Yucatec adjective are not just categories that happen to be homonymous in consequence of
terminological laziness or European bias; they do instantiate the same interlingual category (as
characterized in §184.108.40.206). If so, then a recognition of the parts of speech existing in a particular
language presupposes their definition at an interlingual level. That is the position taken here: parts
of speech will be conceived as interlingual categories, i.e. categories that may show up in individual
2 Parts of speech and the levels of grammatical structure
In modern linguistics, the traditional concept of ‘part of speech’ has mostly been equated with the
word class; and often the latter term is used instead of the former. Now the term part of speech is a
calque on the Latin pars orationis, which is a calque on the Greek méros lógou, all of which mean
literally ‘part of speech’ or ‘part of sentence’. The word classes of structural linguistics, instead, are
defined as lexeme classes. This notion is more abstract because a lexeme is an abstraction
corresponding to a class of word-forms and, therefore, a component of the system rather than of the
text. Consequently, lexeme classes, too, are essentially components of the language system. Thus, a
word class in the sense of ‘lexeme class’ is not actually a 'part of speech' (or of the sentence).
One must, however, bring to account that the ancient authors of the concept ‘méros lógou’ alias
‘pars orationis’ lacked a concept of the syntactic category in the sense of ‘category of syntagma’5
(“phrasal category”) (s. Himmelmann 2007:261), so that their concept comprised not only the word
3This is the distinction that Comrie (1976:3) and others mark by initial upper case and lower case, for the
names of language-specific and interlingual categories, resp. Haspelmath, in several recent publications, e.g.
2012, emphatically rejects the application of concepts like 'noun' and 'verb' at the interlingual level. Now it is
true that such concepts cannot serve as tertia comparationis in language comparison. However, from the fact
that such categories are not universal, it does not follow that they cannot be present in more than one
language. Haspelmath himself (o.c. p. 118) speaks of a nominative marker in Tagalog, certainly not implying
that Tagalog uses a Latin grammatical formative.
4Sapir 1921:125: “no logical scheme of the parts of speech – their number, nature, and necessary confines –
is of the slightest interest to the linguist. Each language has its own scheme. Everything depends on the
formal demarcations which it recognizes.”
5The term ‘syntagma’ is the immediate hyperonym for ‘phrase’, which is a continuous syntagma. In the
following, whenever ‘syntagma’ is meant, the word phrase will be used, as a concession to anglophone
Christian Lehmann, The nature of parts of speech 4
category, but also the syntactic category (to the extent the latter concept applies in Greek and Latin).
Phrases with their syntactic categories are indeed components of the sentence. We will therefore use
the term part of speech not as synonymous with word class, but as the hyperonym of word class
and syntactic category (similarly as in Vogel 2000, §2). Furthermore, the theoretical complication
involved in the concept of lexeme class just mentioned will be avoided, and instead we will
consider word classes as stem classes. Unlike lexemes, stems do occur in texts.
In languages with well-demarcated word classes, there is a systematic correspondence between
some major word classes and certain syntactic categories. This is well-established in structural
linguistics and need here only be recalled by way of the examples shown in T1.
T1 Syntactic categories and word classes in English
nominal verbal adverbial
syntactic category noun phrase verb phrase adverbial phrase
word class noun verb adverb
subclass proper noun intransitive verb adverbal adverb
The simplest possible relationship between a word class and a syntactic category is identity of
distribution. If and where it obtains, an adverb, for instance, can be defined as a word that has the
same distribution as an adverbial phrase.6 Alternatively, if the theory is based on word classes, an
adverbial phrase can be defined as a complex construction that has the same distribution as an
adverb. Identity of distribution between a word class and a syntactic category is guaranteed by
definition if the construction of that syntactic category is endocentric, with the word class in
question as its head (s. §3.2). However, for each of the syntactic categories in T1, there are subtypes
that do not fulfill this condition; for instance, a transitive verb phrase is not endocentric. And on the
other hand, most of the word classes in use are not so conceived. Actually, every word class splits
into a number of subclasses which differ in their distribution. Only one of them has the same
distribution as the corresponding syntactic category. In the case of the nominal category, that is – in
English and some other languages – the proper noun (see examples in §1.1), which is not even
considered a typical representative of the word class ‘noun’. The distinguished subclass is then
joined with other distribution classes under a common word class on the basis of semantic criteria
and membership of some words in more than one of these classes. For instance, English ad-
adjectival (e.g. very) and ad-verbal (e.g. hard) adverbs are subsumed under one class of adverbs
because they appear to be semantically similar and because a couple of adverbs such as partly are
members of both subclasses.
There are various ways how a biunique correspondence between word class and syntactic
category may fail to hold. First of all, there are languages which do not apply syntactic categories at
the root or even stem level. In Latin, roots are acategorial (Lehmann 2008). In Late Archaic Chinese
6The idea of conceiving a major class as a class of words that may substitute for one of the major
constituents in a clause is first expounded in Lyons 1968, ch. 7.6.2. There is, however, silence on the
problem that only a subclass of each major class actually has that potential. In its theoretically strictest form,
the idea amounts to the proposition that there is only one set of syntactic categories which apply both to
complex syntactic constructions and to words. This has received the name of ‘categorial uniformity
hypothesis’ (cf. Himmelmann 2007:249). It underlies X-bar syntax (Jackendoff 1977).
Christian Lehmann, The nature of parts of speech 5
(Bisang 2011, §5.3), Kharia (see below), Tagalog (Himmelmann 2007) and in Polynesian languages
like Samoan (Mosel & Hovdhaugen 1992) or Tongan (Broschart 1997), stems are largely
uncategorized in terms of syntactically relevant word classes. For a subset of these languages
(Chinese, Tagalog, Tongan), the authors claim that lexemes do fall into grammatical classes, but
these are not syntactic categories. In all of these cases, it is only the combination with categorized
expressions, esp. certain grammatical formatives (such as the tense-voice clitic to be seen in E2 –
E4.b below), that categorizes a root or stem in terms of a syntactic category. Such syntactic
categories, then, do not lexicalize into root or stem classes, resp. The same is true for particular
syntactic constructions in many other languages. For instance, Yucatec Maya has the word classes
of numeral (Num) and of numeral classifier (NumCl) and the syntactic category of numeral phrase
(NumP), as illustrated in E1. There is, however, no word class of the same distribution as the
E1 ka'-p'éel abal
YM [ [two-]Num [CL.INAN]NumCl ]NumP [plum]N
The correspondence between word class and syntactic category may also fail for the opposite
reason: certain word classes do not expand into phrases (do not “project”, as some would have it).
This is true for the Yucatec numeral and numeral classifier just illustrated. It is typically the case of
small closed classes, like the adjective or verb in languages which only have a small closed set of
these, and of classes of grammatical formatives like the articles and auxiliaries, in general.
Where categorial uniformity between syntactic categories and word classes does obtain, the
relationship between an endocentric construction and the stem that forms its head is reciprocal in a
1. On the one hand, the construction is an expansion (a “projection”) of its head. Since the head is
an item of the inventory, its category is given, and an endocentric expansion aims at a
construction that preserves the head’s combinatory potential.
2. On the other hand, the head is a lexical condensation of the construction. The category of the
construction is determined by syntactic principles. If the construction reduces to a stem, the
latter inherits the syntactic category, so that it becomes a stem category (a word class).
Note that these are not just a scientist’s alternative perspectives on his object, but there are actually
linguistic processes running in these converse directions:
1. The syntactic operation of modification affords the endocentric expansion of a stem.
2. Grammaticalization and lexicalization afford the condensation of a phrase into a stem.8
That means, in effect, that syntactic category and word class stabilize each other. One may
hypothesize that the part-of-speech system of languages such as most SAE languages, and in
particular their categorization at the stem level, is diachronically stable because it obeys categorial
The relationship, however, is not symmetric. Word classes exist and are such as they are
because they come about through grammaticalization of syntactic constructions and word
7Astonishingly, it is the Spanish loan numerals that have approximately the same distribution as a Yucatec
8More precisely: the transformation of a syntactic category into a word class is a grammaticalization
process; the transformation of a particular grammatical construction into a lexical item is a lexicalization
process; s. Lehmann 2004.
Christian Lehmann, The nature of parts of speech 6
formations. That is, they are the product of a reductive process. Syntactic categories, instead, have a
functional motivation in terms of the propositional operations of reference and predication, as we
shall see in more detail in §4.3. It is at the level of the sentence that these operations are situated and
marked as such by the speaker. The speaker using an expression clarifies whether he is using it as a
referring expression or as a predicate. Markers giving this kind of information essentially specify its
category in terms of parts of speech, roughly speaking, as a nominal or verbal category.
This may be seen clearly in languages which do not classify stems in terms of syntactic
categories, like the ones mentioned on p. 5. Here are a couple of illustrative examples from Kharia,
a strict predicate-final Munda language (Peterson 2005:394f). Clause-final position immediately
preceding the tense-voice clitic categorizes the stem as a verb stem. Position preceding the verb,
with no markers added, categorizes the stem as a nominal stem.
E2 a. lebu ɖel=ki
KHARIA man come=MED.PST
‘the/a man came’
b. bhagwan lebu=ki ro ɖel=ki
god man=MED.PST and come=MED.PST
‘god became man and came [to earth]’
E3 a. aʔghrom
KHARIA ‘Aghrom’ [a town]
‘became / came to be called “Aghrom”’
‘he/she made/named it “Aghrom”’
E4 a. am i karay=oʔb ?
KHARIA 2SG what do=ACT.PST.2SG
‘what did you do?’
b. am i=yoʔb ?
‘what did you do?’
Lebu in E2a is a referring expression, while in #b it is the core of the predicate. Aʔghrom in E3a is a
toponym; but in #b, it is the core of an intransitive predicate, and in #c, the core of a transitive
predicate. Finally, even an interrogative pronoun (i) may not only take the position of a nominal
dependent of the predicate, as in E4a, but also function as the core of a predicate, as in #b. Thus,
subject to semantic compatibility, most Kharia roots can be inserted either in the slot of a verb
complement, in which case they are heads of referential expressions, or in the slot of the predicate,
in which case they may combine with the middle or active voice clitics, with compositional changes
Such data show that the categories of nominal and verbal expression may, in some languages,
not be needed at the word level and only be formed at the sentence level.9 They may, in fact, even
be formed at the discourse level. This may perhaps be best grasped in languages with clearly
9Sapir (1921:133f) makes a similar point about Nootka.
Christian Lehmann, The nature of parts of speech 7
demarcated word classes. In such a language, the category of a stem can be used for the formation
of larger constructions, and this is economic in a certain sense. Often, however, the lexically given
word class is not taken advantage of, as shown by the following two sets of Spanish examples.
E5 a. Así formamos lo que es el barro.
SPAN ‘Thus we form what is the clay.’ (recorded in Guaitil, Costa Rica, 24/02/2010)
b. Así formamos el barro.
‘Thus we form the clay.’
The speaker who said E5a could have said E5b instead;10 the referential meaning would have been
the same. The direct object of the main verb is, of needs, a noun phrase. Its lexical head is the stem
barro ‘clay’, which is a noun. It only needs to be equipped with an article to form a noun phrase,
which is done in E5b. However, the speaker is talking about something that the predicate ‘clay’
applies to. He therefore first converts the noun barro into a predicate by making it depend on the
copula; and then he converts this predicating expression back into a referring expression by
nominalizing it by means of a free relative clause. He thus assigns the syntactic categories needed to
form a referring expression at the sentence level a lt ho u g h the item to be used already has the
category in question. The functions of this strategy are to be sought in information structure and
discourse planning. Ultimately, it is at this level that the speaker decides which components of his
utterance he needs in the nominal and which in the verbal category.11
E6 a. Lo que pasa es que la otra habitación está ocupada.
SPAN ‘What is happening is that the other room is occupied.’ (recorded in Heredia, Costa
b. La otra habitación está ocupada.
‘The other room is occupied.’
The speaker who said E6a could have said E6b instead, with no change in referential meaning.
Instead he nominalizes his proposition so that he can ascribe it the predicate of being the case
(pasa). However, this predicate is nominalized, too, by a free relative clause. So the speaker is left
with two nominal expressions which he now puts into a predicative relationship by a copula (es).12
In this copula clause, the idea of being the case forms the syntactic predicate. This, however, is
topicalized, so that the core predication, i.e. the one represented by E6b, becomes the comment. We
are faced, again, with a strategy of information structure which involves assigning the definitive
syntactic categorization of the linguistic units used only at the highest level of structure, the
discourse level. At the same time, the examples show what the ultimate function of categorization
in terms of syntactic categories is: it is the formation of referential expressions that one wants to
talk about, and of predicates that one wants to ascribe these referents.13
10 The speaker produced more tokens of the former construction during the conversation.
11 Cf. Simone 2006:387f, where nominalization is categorized not as a syntactic operation, but as a discourse
12 What in Spanish is an optional strategy freely employable by combining regular syntactic operations would
be completely grammaticalized into the basic principle of clause formation in Tagalog according to
13 Hopper & Thompson (1984:710) put it like this: “Categoriality ... is thus imposed on linguistic forms by
discourse.” The conclusion, however, that Hopper and Thompson draw from this, viz. that lexemes are
Christian Lehmann, The nature of parts of speech 8
The conclusion from such examples is that the speaker categorizes his expressions at the
highest grammatical level regardless of their categorization at lower levels. In a bottom-up
perspective, expressions are categorized as referring or predicating at the latest at the sentence level.
The principal difference among languages, in this perspective, resides in the possibility to anticipate
categorization at some lower level (Lehmann 2008). Thus, there are languages like German that
categorize already their roots in terms of word classes. Other languages like Latin leave roots
uncategorized and instead categorize stems. Yet other languages like Kharia and Tagalog leave
even most stems uncategorized and defer syntactic categorization to the level of the phrase. Low-
level categorization has the advantage of unburdening the syntax and freeing it for other kinds of
operations. However, if categorization is enforced already at the root or the stem level, it has the
disadvantage that much of that lower level categorization may not be what is wanted at higher
levels and therefore has to be undone by recategorization operations. For such a strategy to work, it
is therefore essential that the low-level categorization be “sensible”, a problem that we will come
back to in §220.127.116.11.
For typology, the issue is, thus, not whether the noun-verb distinction is universal. The
questions are, rather:
1. Which distinctions are required by the constraints introduced in §1.1?
2. Which of these are universally made at the level of grammar, i.e. in linguistic structure?
3. Which of these are universally made at the level of word classes?
4. In particular, given the task of marking the distinction between reference and predication: what
are the possibilities and limits of variation for solutions of this task?
5. Yet more in particular: Assuming that distinction #4 is one of the distinctions meant by
question #2: what are the conditions and consequences of marking it at different levels of
The present treatment is meant as a contribution to answering such questions.
3 Formal analysis: paradigmatic and syntagmatic relations among word classes
Like any other linguistic unit, a word bears paradigmatic relations to other words of its class and
syntagmatic relations to other words in the construction. A subset of these relations are proper to the
word class it belongs to. Therefore one should be able to speak, at a more abstract level, of
paradigmatic and syntagmatic relations among word classes. However, things are more complicated
3.1 Paradigmatic relations
The question of a paradigmatic relation between two entities arises only if they have the same
distribution. (This entails that they either occupy the same syntagmatic positions, or else they are in
complementary distribution, so they may be said to share their distribution in more abstract terms.)
This is a condition not generally met by entire cl as s e s of units if these are distribution classes. A
distribution class includes all items that fulfill the condition mentioned. There is, therefore, nothing
left outside the distribution class that this class could contract a paradigmatic relation with. Thus, if
two word classes were found to be in opposition or complementary distribution (discarding the
possibility of free variation), that would be an argument for subsuming them under a more general
precategorial (p. 747), does not follow; s. Lehmann 2008.
Christian Lehmann, The nature of parts of speech 9
common denominator; in other words, they would not be seen as distinct word classes in the first
place. The (putative) Kharia noun and verb reviewed in §2 provide an example of this.
A general principle of communication says that meaning presupposes choice: by using a certain
expression, a speaker can convey something only if he has a choice and might instead use a
different expression. On this is based a principle of method in structural linguistics which allows the
linguist to pin down a semantic or functional difference between two elements if they contrast in a
given context. This principle applies to individual signs. Applying it to categories of signs yields the
result that these do not meet the condition of substitutability. The principal raison d’être of parts of
speech is to c o mb in e with each other in the formation of sentences. Thus, the question of a
semantic contrast among them does not even arise in any straightforward way.14 The consequence
for the linguist who wants to find out about categorial meanings of word classes by applying the
methods of structural semantics is a methodological apory not easily overcome.
In some loose sense, the speaker does have a choice among word classes in certain contexts. In
the position of the predicate of a sentence, he may use a verb, or he may verbalize a noun or an
adjective by means of the copula. We will come back in §4.4.2 to such a substitution test, as it has
to do with the semantic side of word classes. There we will see that perfect minimal pairs of word
classes are impossible.
3.2 Syntagmatic relations
Viewed in terms of a formal constraint on a semiotic system, compositionality requires that
messages be composed of units that belong to categories that complement each other on the
syntagmatic axis. That is, the string must be segmentable into units that instantiate categories which
allow them to be grouped into larger units (constructions) by syntagmatic relations based on these
Syntagmatic relations between parts of speech may be conceived in terms of dependency
grammar (or its equivalent in other models of syntax, e.g. Gil’s (2000) categorial syntax). At the
highest taxonomic level, they subdivide into relations of sociation and dependency. The former may
serve to assess the role of certain minor parts of speech like the sociative particles, which we will
return to below (class 2a)). Dependency relations are recognized on the basis of the distribution of
the components contracting them. More specifically, each of these components belongs to some
category defined as its distribution class; and the resulting complex construction again belongs to
some such category. In a dependency relation, one of the members of the relation determines the
category of the resulting construction. That member is X in T2. Two cases may be distinguished:
either the complex category is simply the category of one of the members of the relation; or it is
determined by one of them without being identical to the latter’s category. T2 systematizes these
two dependency relations for the major parts of speech presently at stake. X’ means a category
differing from X in its distribution only by not combining with Y. The instantiations listed on the
right-hand side of T2 are interlingual categories in the sense of §1.2. The slash separates the phrasal
category from the word class as introduced in §2.15
14 The case of vowels and consonants in phonology is largely analogous.
15 A noun phrase is functionally caseless; a cased noun phrase is like an adpositional phrase, falling into the
distribution class of the adverbial phrase. Given this distinction, the conception may extend to multivalent
verbs beyond bivalent verbs: their first object may be a noun phrase, their second object, a cased noun
Christian Lehmann, The nature of parts of speech 10
T2 Dependency relations and parts of speech
X Y complex category
[ X Y ]XY modifies X nominal / common noun
verb phrase / verb
adjectival phrase / adjective
adverbial phrase / adverb
[ X Y ]X’ X governs Y bivalent verb
noun phrase / proper noun monovalent verb phrase
Each in the following set of examples illustrates one of the lines of T2:
E7 [ [ old ]Adjective [ house ]CommonNoun ]Nominal
E8 [ [ lives ]Verb [ in the house ]AdverbialPhrase ]VerbPhrase
E9 [ [ bought ]BivalentVerb [ the house ]NP ]VerbPhrase
E10 [ [ top ]RelationalNoun of [ the house ]NP ]Nominal
E11 [ [ devoid ]RelationalAdjective of [ meaning ]NP ]AdjectivalPhrase
E12 [ [ in ]Adposition [ the house ]NP ]AdverbialPhrase
Given the two configurations of the first column of T2, the two principal categories may be
characterized in purely structural terms, like this:
•There is a part of speech whose members can take the position of X, but not of Y in dependency
constructions. In other words, they may be modified, but not governed; and they govern, but do
not modify other elements. That is the verb phrase / verb.
•There is a part of speech whose members can only function as Y (the dependent) in
government. That is the noun phrase / proper noun.
The differential combinatorial potential of parts of speech is reified as their grammatical
relationality (which, in predicate logic, takes the form of argument places): a governor and a
modifier extend a relation to what they govern or modify, whereas a governed or modified element
contributes nothing to the relation and, instead, just occupies the argument place opened for it.
Grammatical relationality, in turn, is the structural correlate of conceptual relationality: the notion
designated by a non-relational noun (a “punctual concept” in the terms of Prandi 2004:122-124) is
autonomous, whereas the notions designated by verbs, relational nouns, adjectives, adverbs and
adpositions are dependent and refer to an autonomous notion that fills their argument place. This
cognitive aspect of grammatical relationality will be developed in §4.4.3.
Dependency relations define ranks for their members (cf. Jespersen 1924, ch. 7): the member
that determines the category of the construction is at a higher rank than the dependent. These ranks
translate directly into importance of these categories for sentence construction and, thus, for the
language system: The category that depends on nothing, viz. the verb, occupies the highest rank.
The category that directly depends on the former, but is autonomous in terms of relationality, viz.
the noun phrase, occupies the second rank. The category that always depends on something else and
is also not autonomous in terms of relationality, viz. the modifier, occupies the third rank.
The concept of modification can, thus, be defined on a purely structural basis, viz. on the basis
of an endocentric construction as represented in the first line of T2. As may be seen, for a semiotic
system to have categorial uniformity for some construction presupposes that this be endocentric,
Christian Lehmann, The nature of parts of speech 11
and this entails that there must be modification and, thus, modifiers. It may be anticipated here that
modification differs in this respect from the two propositional operations, reference and predication
(cf. §§4.3 and 18.104.22.168), whose basis is not in T2 and instead in functions of communication.
Government, i.e. governing relationality of the elements to be classified, is a subordinate
criterion in the structural classification. Suppose that, instead, the potential to take a complement
was a primary criterion in classification. Then transitive verbs, relational nouns and prepositions in
an ergative language might form a major distribution class. The class would exclude intransitive
verbs, non-relational nouns and adverbs. Such a class is not necessarily useless. There may be
grammatical rules that refer to it, and there may be stems that shift from one of the classes to
another just on the basis of acquiring or forfeiting the governing potential that is the basis for their
distinction. That is actually the case in Yucatec Maya.16 The #a examples of E13 – E15 illustrate the
three subclasses of that major distribution class.
E13 a. t-in ch'ul-ech b. h ch'úul-ech
YM PRFV-SBJ.1.SG wet(CMPL)-ABS.2.SG PRFVwet\DEAG(CMPL)-ABS.2.SG
'I wetted you' 'you got wet'
E14 a. in watan b. hun-túul atan-tsil
YM POSS.1.SG wife one-CL.AN wife-DEREL
'my wife' 'a wife'
E15 a. t-in paach b. paach-il
YM LOC-POSS.1.SG back back-ADVR
'behind me' 'behind'
As may be seen, there is an operation of derelationalization which blocks the governing slot present
and occupied in the #a examples to yield the non-relational stems appearing in the #b examples: an
intransitive verb, a non-relational noun and an adverb, resp.17 Although the derelationalizer displays
allomorphy, it applies to all the subcategories of that class in like fashion. However, the
dependencies filled by the categories in this distribution class (the ways in which they depend on
other items) are essentially different, and so are the ways that they themselves can be modified.
Therefore, in a hierarchy of parts of speech, the categories of verb, noun and adverb, regardless of
their valency, will be introduced at a higher level. Then the criterion of governing relationality will
apply to each of them to generate the subcategories of plurivalent verb, relational noun and
adposition. This will be taken up in §22.214.171.124.
The traditional class of particles s.l. (words that do not inflect) is not covered as such by the
foregoing description. That is a heterogeneous class not susceptible of a unified account. It may be
subdivided as follows:
1. A subset of the particles are modifiers. That concerns adverbs and their derivatives, adpositions
and subordinative conjunctions. They also differ from the particles of the second subset by
forming open, productive classes (s. §6). They are treated in §4.4.3.
2. The remaining subset, the particles s.s., comprises those particles which do not enter a
dependency relation. In consonance with this, there are also no productive ways of enriching
this class. This is, again, a negative definition which leaves two possibilities:
16 Yucatec Maya is not a (syntactically) ergative language, but its remnants of ergative morphology may
serve for the illustration presently required.
17 There are also, in Yucatec Maya, inverse operations to form transitive stems, relational nouns and
prepositions; however, they are structurally less uniform.
Christian Lehmann, The nature of parts of speech 12
a) A subclass of particles s.s. contracts various relations of sociation instead of dependency
relations. It comprises coordinative conjunctions like or, focus particles like not, yet,
discourse markers like however and maybe others. In the system to be presented in §7, they
would be introduced as particular subtypes of minor parts of speech, to be called sociative
b) The remaining subclass of particles do not integrate themselves into a sentence at all. These
are the interjections and ideophones. An interjection constitutes a sentence by itself; an
ideophone may appear in a sentence as a parenthesis or quoted speech. These holophrastic
particles are treated in §4.2.
All of the above are gross characterizations that pass over a lot of cross-linguistic and internal
variation. Their point is to show how a word-class system may fulfill the formal requirements
imposed on grammatical structure by a semiotic system.18 It is true that the syntagmatic properties
of parts of speech examined above also have to be the basis for their language-specific
distributional definition. This, however, is no straightforward matter:
1. A distributional definition defines X with reference to its context Y. Y, however, is of the same
nature as X: it is itself a distribution class. Thus, Y must have been set up in the same way. In
order for the definition system not to be circular, one needs to choose fixed points from which
to start. Such a fixed point may be established by non-distributional criteria. This means in
essence functional criteria of the kind introduced in §1.1. To the extent that such criteria cannot
be operationalized, starting points in the definition hierarchy just have to be stipulated.19
2. Such a fixed point may be a part-of-speech category. In an inflecting language, however, the
only way for stems as members of a word-class to occur in texts is provided with inflectional
morphemes. In such cases, there is no uniform syntactic context to base a distributional
definition of that word class on. Instead, it is the morphological paradigm appearing on the stem
class X that provides the immediate context for a distributional definition of X. Morphological
paradigms, however, are not part-of-speech categories. If such a paradigm is to fulfill this
function, it must, again, be either identified by other criteria or simply be taken for granted.20
We will return to hierarchical relations between parts of speech in §7.
18 Time and again (e.g. Beck 2002:18, Smith 2010, §2.1), criticisms are leveled against this kind of account
by examples of English nouns serving as modifiers, adjectives serving as verb complements, and suchlike.
Such examples contribute or detract nothing with respect to the theory at stake as long as the question has not
been asked what it is supposed to account for. The present theory is not meant to account for conversions
possible in English.
19 For instance, in more than one grammar, the noun is defined as the part of speech that combines with a
determiner to form an expression that may refer.
20 For instance, a Latin noun cannot be defined as a sign occurring in certain syntactic contexts, since it
would change its morphological form depending on the syntactic context. Again, a Latin noun stem may be
defined as a sign occurring in certain morphological contexts (essentially, declension endings). Then,
however, those morphological contexts would either have to be enumerated or to be replaced by an
abstraction like 'the grammatical categories of case and number'.
Christian Lehmann, The nature of parts of speech 13
4 Functional analysis: cognitive and communicative categories
4.1 Theoretical preliminaries
The general question of this section is what purpose is served by the categorization of meaningful
units in parts of speech. Assuming that communication and cognition are the two topmost functions
of language, it will be argued that this purpose lies more in the domain of communication than in
the domain of cognition. As a background to this claim, a minimum characterization of the two
domains is needed.
The communicative dimension of language is its social dimension, i.e. the dimension
connecting the speaker with the addressee. Functions subsumed under this concept concern contact
and social relations between the interlocutors in the speech situation, including speech acts, and
conveyance of content to the hearer (while excluding the content itself), more specifically,
manipulation of the universe of discourse, sequential management of the message, its coherence
including reference tracking, and its packaging in terms of information structure.
The cognitive dimension of language is the dimension connecting the interlocutors with the
(physical or imagined) world and concerning the content transmitted between them. Functions
subsumed under this concept concern apperception, thinking and orientation. It is structured in
terms of cognitive domains such as possession, spatial orientation, participation etc.
4.2 Holophrastic words
Both of the dimensions of cognition and communication concur when the speaker, on the basis of
some concern of his, forms a minimum message that he wants to convey to the hearer. At the initial
stage, the minimum message may remain grammatically inarticulate. We then get a holophrastic
utterance, as in E16.
E16 a. Gosh!
E16a features an interjection, E16b an ideophone. Both convey a minimum message which may be
explicated, but which is left inarticulate. The interjection conveys a proposition about the speaker as
he is in the current speech situation, while the ideophone conveys a proposition on any other
referent, including somebody else in the speech situation, or on any referent, including the speaker,
in a different situation. Interjections and ideophones are thus in complementary distribution.
Together they form the category of holophrastic words. These involve no articulation of the
message in the terms relevant in the next section. They are the primordial parts of speech or rather,
wholes of speech.21
From among the heterogeneous class of particles s.l., the functional approach thus singles out
interjections and ideophones. As was seen in §3.2, they may be characterized as words that contract
no syntagmatic relations at all.22
21 Cf. Gil 2000, §3, where the sentence is taken as basic for a theory of syntactic categories; and these, in
turn, are taken as more basic than word classes. Most other accounts of word classes do like Bisang 2011 in
glossing over ideophones and interjections. Heine & Kuteva (2007, ch. 2), in their theory of the evolution of
grammar, just forget them.
22 While this seems clear for interjections, ideophones may be used parenthetically. The present treatment
does not account for their use as predicates.
Christian Lehmann, The nature of parts of speech 14
4.3 Communicative functions
The minimum message is composed of a proposition and an illocutionary force. The latter is
generally coded at non-segmental levels or at most by grammatical formatives and therefore has
little to do with parts of speech. If it is coded syntactically, it takes the form of a proposition, and
hence the same considerations as for propositions apply. Thus, at the next stage of development of a
part-of-speech system, the proposition is decomposed.
At this point, the two propositional operations, reference and predication, come into play:23
the speaker distinguishes whether he uses a certain expression in order to refer to something or in
order to predicate something. This distinction is very general and manifests itself in linguistic
structure at different levels (cf. Meier 1979). One of these levels, already illustrated in §2, is
information structure, where it takes the form of topic vs. comment. Another level is syntax, where
it takes the form of the two basic syntactic functions of subject and predicate. These are
instantiated by two syntactic categories, the noun phrase and the verb phrase.24 And finally, at the
lexical level, they take the form of noun vs. verb. As is to be seen from T3, the functions fulfilled
at the semiotic levels of semantics and information structure translate into syntactic functions once
the level of the meaning-bearing systems of the language (grammar and lexicon) is reached. And
only there are they paired with parts-of-speech categories destined to fulfill them par excellence.25
Finally, these two basic syntactic categories are optionally mirrored in the inventory.
T3 Communicative functions and word classes
level functions categories
semantics reference vs. predication
information structure topic vs. comment
syntax subject vs. predicate noun phrase vs. verb phrase
lexicon noun vs. verb
It is at the end of this chain of relations that the two principal word classes may be characterized in
functional terms: a noun is a word of a category whose primary function it is to refer; a verb is a
word of a category whose primary function it is to predicate.26 Needless to repeat, these are
23 The functions of these operations have been termed ‘pragmatic’ functions in other accounts, e.g. Croft 1991, Anward
et al. 1997:172, Smith 2010, Bisang 2011, §1. They have nothing to do with pragmatics; propositional acts are part of
linguistic meaning and, thus, semantic in nature. Even calling them ‘discourse’ functions may be misleading. They do
concern the discourse in the sense that that term appears in the French parties du discours ‘parts of speech’; but they are
not located at a level above the sentence.
24 At this level, nominalization comes in as an operation converting a predicative expression into a referring
expression. This operation presupposes the loss of illocutionary force accompanying the subordination of a
sentence and then detracts further from its sententiality by suppressing the propositional act of predication.
25 Hengeveld (1992[P]) speaks of referential phrases and predicate phrases, assuming thus a categorial
instantiation of the discourse functions already at the semantic level. That, however, presupposes a semantic
representation in a particular formalism on which such concepts may be based.
26 The primary function of a stem is that function which it may fulfill without any structural apparatus. Any
function which requires additional structural means is then a secondary function (cf. Kuryƚowicz 1936:80,
Christian Lehmann, The nature of parts of speech 15
functional characterizations of interlingual concepts. Neither is it necessary that every language
implement the contrast between reference and predication at the lexical level (cf. Lyons 1977:429f),
nor does this functional basis provide much methodological help in identifying the classes of noun
and verb in a language.
These two categories thus find their ultimate motivation in the communicative functions of
language. All other categories are functionally subordinate to these two and therefore have an even
more indirect functional motivation or, rather, a predominantly structural motivation.
4.4 Cognitive functions
4.4.1 Notional theories
Parts of speech have a basis in cognition to the extent that the following presuppositions are
fulfilled: cognition has a categorial structure independently of linguistic structure, and both the
sheer existence of parts of speech and the specific parts of speech employed in the languages of the
world are motivated as representing this categorial structure. A theory based on these
presuppositions is a notional theory of parts of speech.
The major problem with such a theory has been observed repeatedly: a notion alone does not
determine the word class in which it is coded. That is true both at the level of the individual notion
and at the level of the notional category. At the former level, the argument that words of different
classes may represent the same notion was first made by the modistae. They used the example of
the notion of ‘whiteness’, which in Latin may take the forms of albus ‘white’, albedo ‘whiteness’
and dealbo ‘be white’ (Thomas of Erfurt 1972, §46). The modist doctrine holds that the meaning of
a part of speech is not among the semantic features of the lexeme in question and, instead, a modus
significandi, a ‘mode of signifying’. In other words, the part-of-speech category is no t given with
a notion, but something chosen for its linguistic representation.27 Jespersen (1924:91) makes the
same argument with exemplary incisiveness, illustrating with a whole sentence:
E17 a. He moved astonishingly fast.
b. He astonished us by the rapidity of his movements.
Jespersen offers 10 near-synonymous transformations of E17a (of which E17b is just one) by
converting each of the notions ‘move’, ‘astonish’ and ‘fast’ through the word classes of noun, verb,
adjective and adverb. At the level of the conceptual category, the analogous argument has often
been made with the concept type ‘property’. While notional theories of parts of speech would have
it that the part of speech ‘adjective’ is the structural reflex of the conceptual category ‘property’, in
actual fact, properties are coded both by adjectives like beautiful and by (abstract) nouns like
beauty. While relations of markedness may help in identifying one of alternate codings as more
basic (cf. Croft 1991:53-55), this does not yield uniform results either within a language or across
Dik 1989:162, Croft 1991, ch. 2, Hengeveld 1992[N]).
27 The approaches reviewed in Bisang 2011, §2.1, which identify a part of speech by concepts which are its
prototypical members, fail by disregarding this. Putting English words such as SEE, BIG in capitals in order to
designate concepts is of no avail here: While it may be a useful methodological approach to identify, in the
target language, the translation equivalents of such English words as see, big, concepts such as SEE, BIG are
insensitive to word classes, i.e. they cover equally see and sight, big and size.
Christian Lehmann, The nature of parts of speech 16
languages and would, in the example at hand, identify the noun as the part of speech that basically
codes the property of beauty.
4.4.2 Categorial meaning
What has been said so far does not encourage the search for categorial meaning, i.e. the intension of
a word class. However, it still befits us to briefly review some evidence for categorial meaning of
parts of speech that has been brought forward in the literature. It is confined to three major parts of
speech, noun, verb and adjective; and it comes from languages that possess a productive class of
adjectives and in which linguists can control finest shades of meaning. Methodologically, this
section takes up where the discussion of paradigmatic relations among parts of speech in §3.1 left
126.96.36.199 Noun, adjective and verb
Nouns and adjectives may alternate in a couple of contexts. One of these is in the predicate of a
copula clause, as in E18 (from Jespersen 1924:75-77):
E18 a. c’est rose
FRENCH ‘it’s pink’
b. c’est une rose
‘it’s a rose’
E18a only entails ‘it is colored’, thus, the hyperonym of ‘pink’, while E18b not only entails ‘it is a
flower’, but also ‘it has thorns’, ‘it has pinnate foliage’ etc. More in general: Given a proposition of
the form ‘X is P’; then if P is an adjective, the proposition entails only hyperonyms of P; if it is a
noun, then it entails a sometimes heterogeneous set of more or less specific predicates.28 It is the
combination of these that constitutes the higher ontological autonomy of what is signified by a noun
as against an adjective.
Another difference between the two word classes becomes clearer in E19 (example from Bally
1921:305 taken up in Jespersen 1924:77):
E19 a. vous êtes impertinent
FRENCH ‘you are impertinent’
b. vous êtes un impertinent
‘you are an impertinent guy’
The substantivization of E19b has the effect of subsuming the subject under an established class,
thereby characterizing it in an essential way, i.e. forestalling the interpretation of a contingent state.
This is also illustrated by E20.
E20 Having been a Conservative Liberal in politics till well past sixty, it was not until
Disraeli’s time that he became a Liberal Conservative. (Jespersen 1924:78)
The wording obviously presupposes that liberals and conservatives are classes and that the subject
is subsequently subsumed as an element under either of these classes. In E20, these classes are
28 Cf. Jespersen 1924:75: “the adjective indicates and singles out one quality, one distinguishing mark, but
each substantive suggests … many distinctive features”.
Christian Lehmann, The nature of parts of speech 17
presumably stabilized by party membership. Being essentially a member of either of these classes,
the subject is secondarily characterized by a property. The underlying principle is that a substantive
says what an entity primarily and essentially is, whereas an adjective only attributes a property (or
just a state) to it which may be compatible with many other properties of the same significance.
These examples are apt to show that predicating a noun on a referent implies its inclusion in a
conceptually stable class, its subsumption under a kind, with the associated stereotyping effect.
Predicating an adjective on a referent implies ascribing it a property or state as a more or less stable
characteristic without, however, categorizing it in any essential way (s. Wierzbicka 1986).
Now as for adjectives and verbs, a direct opposition between them in predicative position may
be obtained in a language that possesses a set of roots from which either an adjective or a verb stem
may be formed. This is the case in Latin, as illustrated by the examples in T4 (cf. Lehmann 1995,
T4 Verb and adjective in Latin
form meaning form meaning
ūmēre be wet ūmidus wet
valēre be strong validus strong
līvēre be blue līvidus blue
frīgēre be cold frīgidus cold
In general, given a root X, then the verb stem X-ē- designates the state X, whereas the adjective
stem X-ido- designates the property X. E21 provides a minimal pair:
E21 a. bracchia līvent
LATIN ‘the arms are blue’
b. bracchia līvida sunt
‘the arms are blue’
In E21a, the arms are temporarily blue, perhaps having been tossed. The arms of E21b instead are
permanently blue, being perhaps the arms of a painted statue. Thus, the difference between the
categorial meaning of adjectives and verbs in Latin is, again, one of time-stability: the proper locus
for the adjective is a property; for notions with lesser time-stability, a verb is employed.
These semantic differences between nouns, verbs and adjectives can be related to their primary
function: A noun subsumes its referent under a class. This operation presupposes that the class has
members which have essential traits in common. An adjective does not do that; it just attributes a
predicate to its referent. This predicate is a property or a state, thus, less time-stable than the
predicates conveyed by nouns. An adjectival predicate involves no class-formation and therefore
does not imply that what it predicates characterizes the referent in an essential way. This is nicely
shown by examples such as E19f. Finally, a verb says that its argument is temporarily in some
situation (situation is a hyperonym of event) which may change.
Christian Lehmann, The nature of parts of speech 18
188.8.131.52 Grammatical meaning and types of concepts
In examples such as the above, most clearly perhaps in E20, there is no structural motivation for the
choice between one word class or another. The speaker is entirely free here in his choice among
categories. Accounting for the choice implies either finding a semantic motivation or pleading for
free variation or extralinguistic factors. The latter, however, is excluded by the nature of examples
such as E20. Such examples only allow the conclusion that word classes may have a semantic side,
even if this be only a contextually conditioned effect.
Observe, however, that by the strictest structuralist standards, we have not been able to come up
with a minimal pair contrasting two word classes. In E18f, the noun is in a different context than the
adjective, viz. following an article. The latter is an overt recategorization operator, which
contributes something to the meaning difference between the #a and #b versions. In E20, the
adjective is in prenominal position; the substantive is not. Finally in E21, the root in question
precedes an -ē- formative in the #a version, but an -ido- suffix in the #b version. Each of these
makes some contribution to the meaning difference between the two forms. An absolute minimal
pair featuring a given stem in two different categories in the very same context is logically
impossible: there would by definition be nothing whereby one could recognize the categorial
In assessing the semantic phenomena demonstrated in the preceding two sections, we have to
bear in mind that if a linguistic unit has some semantic potential, this does not entail that it conveys
that meaning on every occasion of its use. A categorial meaning is a grammatical meaning which
has no expression of its own. This kind of meaning is extremely fragile and easily overridden by
other factors. To render this clearer, we will briefly compare two related areas, markedness and the
contrast between lexical features and features coded separately.
First consider the case of markedness oppositions: In certain contexts, the present tense means
‘at the time of this speech act’. It has this sense primarily when it contrasts syntagmatically with a
more marked tense whose meaning is incompatible with it, as in E22a.
E22 a. Just war, as it was and is. (Johnson, James T., First things, January 2005.)
b. from time to time the information involved is very sensitive (www.lingue.de)
Whenever there is no such contrast, the semantic feature may remain inactive, as in the timeless
(“gnostic”) present. And it may even be overridden by some contradictory feature coded more
explicitly in the context. Thus, reference to the moment of the speech act is excluded if a present
tense verb is accompanied by a temporal adverb like the one of E22b.
Second, consider such features as constitute verbal characters and aktionsarten. They may be
coded at different grammatical levels in different degrees of explicitness. The verbal character may
be determined already at the root level, so that root verbs behave differently depending on it. Or
else the verbal character may be fixed by an aktionsart derivation. Again, that kind of meaning may
be conveyed by inflectional aspect. And finally, there is the possibility of determining such things
as telicity by syntactic operations, e.g. by combining a verb with a definite direct object. The verbal
character of a root may become effective in contexts that allow it to develop. And it may be
overridden by overt higher level operators. This is illustrated by E23. The verbal character of the
German root verb schlafen ‘sleep’ is atelic (durative), as shown in the diagnostic context of #a. In
the compound verb einschlafen ‘fall asleep’ shown in #b, the aktionsart is fixed as telic (ingressive),
as again proved by the diagnostic framing adverbial. This categorization is, again, undone in #c,
where the periphrastic progressive aspect forces atelicity on the verbal complex.
Christian Lehmann, The nature of parts of speech 19
E23 a. Linda schlief sieben Stunden.
GERM ‘Linda slept for seven hours.’
b. Linda schlief innerhalb von Sekunden ein.
‘Linda fell asleep in a few seconds.’
c. Linda war gerade am einschlafen, als das Telefon klingelte.
‘Linda was just falling asleep when the phone rang.’
What such examples show is that a grammatical or semantic feature that is not coded separately is
fragile. The same is true of the semantic features associated with part-of-speech categories. These
are totally implicit and therefore subliminal. They come out in such contrasts as examined in
§184.108.40.206; but otherwise they remain dormant. They may easily be overridden by operations of
recategorization such as those illustrated in §2. And wherever the speaker does not have a choice, a
potential contrast is neutralized. There is, thus, no contradiction between examples such as E17 and
examples like E19 and E21.
The conclusion from this is that parts of speech are primarily not semantic, but s y n ta ct i c
categories. Only secondarily, namely if they are lexicalized in the form of stems, does the question
arise which kinds of notions it would make sense to have available in the inventory in which word
class. In other words, the essence and raison d’être of a part of speech is not some kind of highest
hyperonym for all of its members. The role of notions in the formation of a part-of-speech system is
that notions of a certain kind are typically needed in one of the communicative functions so that it
makes sense to store the respective categorization with their lexemes, i.e. to assign them “already”
in the lexicon the word class corresponding to that function. The communicative functions reviewed
in §4.3 have, thus, priority in a functional account of word classes, while cognitive kinds play an
Cognitive kinds may be distinguished by the parameter observed to be operative in §220.127.116.11,
viz. time-stability (see Givón 1979, ch. 8 and Lehmann 1991, §3.4). It constitutes a scale on which
concepts may be arranged. Time-stability of a concept correlates in an essential way with its
conceptual relationality, as follows:
•The most time-stable concepts are those of the lowest degree of relationality, thus
representations of objects (in the widest sense of the word). As these objects are time-stable,
concepts of them characterize them in an essential way.
•The least time-stable concepts are those of the highest degree of relationality, thus
representations of events. Since events are volatile, such concepts do not characterize or
classify objects involved in them in any essential way and instead provide information on
•Concepts of an intermediate degree of relationality also display an intermediate degree of time-
stability; they represent properties and states which characterize objects more essentially or
Besides the relational functions, to be reviewed in the next section, time-stability constitutes the
most important cognitive parameter that is relevant for parts of speech.
29 At this point, the present account follows Hopper & Thompson 1984:708 in “that the lexical semantic facts
about N's and V's are secondary to their discourse roles” and derivative of the latter. By the same token, it
diverges from the accounts presented, among others, in Croft 1991, ch. 2 and Gil 2000:197, where cognitive
categories are directly associated with syntactic categories (including parts of speech).
Christian Lehmann, The nature of parts of speech 20
The general statement that word classes only have a derivative, if any, cognitive basis and
therefore only a weak, if any, common semantic denominator is subject to one exception, which
concerns the numerals. These are the only word class definable on a purely semantic basis, viz. as
words designating numbers. Thus, they do form a lexical field which, although lacking an
archilexeme, is based on the common semantic denominator of designating the cardinality of a set.
As is to be expected, a word class constituted in such a way is in an orthogonal relation to the other
word classes, which are not constituted by notional criteria, but by their function in structuring a
message. And true enough, numerals may behave as nouns, adjectives or verbs in different
languages. Even inside a given language, the set normally falls apart into subsets that share
properties with different word classes: the lowest numerals tend to lack syntactic autonomy, while
the higher numerals are more noun-like (Lehmann 2010). A consistent theory would therefore not
posit the numeral as a separate part of speech. For English, they may be subsumed under the
nominal category and then subdivided into more substantival and more adjectival numerals.
4.4.3 Conceptual relationality
In the course of the syntagmatic structural analysis performed in §3.2, it was seen that some parts of
speech can be conceived as categories fulfilling specific functions in dependency relations. We are
here particularly concerned with the modifiers and governors of T2. As was said there, these are
equipped with grammatical relationality. The latter has a cognitive basis, to which we now turn.
18.104.22.168 Modifying relationality
Concepts may be modified in order to be used for reference and predication. Modification, thus,
produces operands of these two operations. As we saw in §3.2, the formal basis of modification is
modifying grammatical relationality, defined as the potential to function as Y in [X Y]X. Again, the
question arises what kinds of concepts are predestined for such a syntagmatic function. The answer
lies in the kind of conceptual relationality that enables a concept to contribute to the function of
Consider first modification of predicative concepts. Situation concepts are primarily coded in
the verbal sphere and, to that extent, lexicalized as verbs. There may be languages with an all-
embracing class of verbs which leave little room for anything else (Hengeveld’s (1992[P]:69) type
7). Examples include Hengeveld’s (l.c.) Tuscarora and Sasse’s (1993) Cayuga.30 However, specific
situation concepts are composed of certain basic features which are modified by more specific
features. For instance, sneak is move stealthily. Such specific semantic features may be coded
syntactically as modifiers, that is, as some kind of adverbial, as in the paraphrase given. Often, there
is the alternative of coding the specifying feature by a higher verb. For instance, where English says
appeared again, coding the repetition by an adverb, Spanish says volvió a aparecer (“returned to
appear”), coding it by a higher verb. Languages make use of these possibilities to different extents.
Some languages like Spanish and Yucatec Maya rely predominantly on verbs. German, although
certainly not poor in verbs, prefers adverbial modification over higher-level predicates in certain
functional domains (Lehmann 1990). Other languages abide by a small set of verbs and code all
more specific situation types in some kind of verbal dependent. There are several subtypes of this
latter strategy, having to do with the particular word class assigned to the specific situation
30 although these are probably virtual rather than actual examples; s. Mithun 2000
Christian Lehmann, The nature of parts of speech 21
concepts. They may be adverbs or “preverbs” or converbs, as in Jaminjung.31 They then act
syntactically as modifiers of the main verb. Or else they may be treated like abstract nouns. In that
case they form some kind of inner dependent of the main verb (a light verb), as they do in Persian
and Korean. While such a pattern may remain stable over a long time, combinations of a verb with a
dependent that represent a specific kind of situation tend to lexicalize as verbs. This leads to an
enrichment of the verbal lexicon. One may therefore hypothesize a long-term cycle of enrichment
and depletion of the verbal lexicon.
The same goes for the modification of referential concepts. To the extent that the inventory
does not provide a particular referential concept needed in the discourse, one may form one by
combining a hyperonym with a modifier. For instance, German Schimmel is English white horse.
The part of speech functioning in this operation is the adjective. Adjectives are often similar to one
of the primary parts of speech, either nouns or verbs, and may even be a subcategory of one of these
(cf. Bhat 1994, Wetzer 1995). In Latin and English, the adjective is a nominal category, in Thai, it is
a verbal category. Some languages have a very small class of adjectives (cf. Dixon 1976); Yukaghir
only has ‘big’, ‘small’, ‘old’ and ‘young/new’. And some languages, including Goemai (Hellwig
2007), Korean (Evans 2000:714) and Lao (Enfield 2004), lack adjectives altogether.32
The adjective and the adverb are alike in their primary function of modifying another concept.
Consequently some languages abide by a generic category of modifier, which may be combined
indiscriminately with nouns and verbs. Hixkaryana is an example (Derbyshire 1979, ch. 2.1.4).
Furthermore, conversion between adjective and adverb is often conditioned by rules of grammar.
For instance, nominalization of a verbal clause like E24a entails the conversion of the modifier of
the verb into an adjective, as it appears in E24b.
E24 a. Linda works heavily
b. Linda’s heavy work
While nominalization may have certain semantic effects like suppressing predication (cf. E5f), the
accompanying conversion of the adverb into an adjective is an automatic and obligatory
consequence of this syntactic operation. This is further evidence that, in such languages, adjective
and adverb do not differ in their categorial meaning33 but, instead, exclusively in their syntactic
It follows from the above discussion that the concepts of modification and modifier are
paradigm examples of mixed concepts in the sense of §1.1. A purely semantic definition of
modification has proved difficult (Smith 2010) because it is hard to capture the difference between
modification and predication without reference to formal structure. The intuition is, anyway, the
following: Given concepts X and Y such that X either refers or predicates. Then Y modifies X iff it
predicates on X while, at the same time, subordinating itself to the function of X. Modification,
thus, implies a distinction of levels of force in semantic structure: A modifying expression may, in
itself, have a referring or predicating potential. This is, however, subordinated to the referring or
predicating function of the modified. This kind of self-subordination is the nature of modifying
31 The terminological problem is telling here. The words in question code the bulk of what in SAE languages
are verbal meanings. Structurally, however, they are not verbs but, quite to the contrary, they presuppose a
verb that they combine with.
32 In these three languages, properties and states are primarily lexicalized as stative verbs.
33 Things may be different in languages like Latin and Italian, where there are minimal pairs like Ital.
cammina veloce ‘walks quick’ and cammina velocemente ‘walks quickly’.
Christian Lehmann, The nature of parts of speech 22
relationality. At the same time, it provides the reason for us not to accord modification the same
status as the propositional operations of reference and predication.34 It is here treated as a syntactic
operation, thus, as an operation with a semantic and a structural side.
As is usual with such language-independent definitions, it does not suffice for the purpose of
singling out and delimiting particular constructions. As seen in §3.2, the concepts of modification
and modifier have a formal correlate in the concept of the endocentric construction. However, that
concept offers no clue for delimiting attributive and adverbial modification against other
endocentric constructions like apposition or even disjunctive coordination. The syntactic operation
of modification constitutes the set of constructions that emerges in the intersection of the functional
and the formal perspectives. In particular, the concept of self-subordination needed in the
characterization of modification is precisely the role of Y in [ X Y ]X (cf. T2).
The conceptual basis of a modifier, then, does not lie in its intermediate degree of time-
stability; that is only the conceptual category that most easily lends itself to that function. Instead,
its essence is the kind of relationality that it is equipped with: the conceptual relationality of a
modifier is such that it attaches to the concept of the modified. This function of the modifier in
messages translates into a stabilizing function in the part-of-speech system: As we saw in §2,
endocentric modification affords categorial uniformity. This occasions the hypothesis, already
mentioned in §3.2, of a typological correlation between presence of modifiers and categorial
uniformity in a language.
22.214.171.124 Governing relationality
We now come to the functional notions that correspond to the lower half of T2 and, thus, to the
conceptual correlate of government. Verb, adjective and adverb are semantically relational in that
they refer to some entity constituted independently of themselves, on which they provide more
specific information and which they are, thus, ascribed to as predicates in the logical sense. That
entity is what is designated by the subject of the verb, by the head nominal of the adjective and by
the verb phrase modified by the adverb. However, these parts of speech may also be relational in
another sense: There may be yet another referent, which serves as a reference point for the situation
and whose relation to the latter is mediated by the verb, adverb or adjective. The same holds of a
nominal concept which may be individuated by reference to such an external fixed point.
Conceptual relationality comprising such an argument position spells out grammatically as
governing relationality. Governed referential expressions do refer independently, but they serve as
a reference point for the governing concept, which they thus help delimit and individuate.
Verbs having governing relationality in addition to their subject place are plurivalent (and
possibly transitive), adjectives possessing governing relationality are relational (like reminiscent [of
that achievement]), adverbs become adpositions (like behind [the door]), and likewise there are
relational common nouns (like [Sue’s] aunt); the brackets enclose the complement. In terms of
categorial grammar, the construction consisting of a word of one of these latter categories and its
nominal complement is of the same category as an intransitive verb, a non-relational adjective, an
adverb or a non-relational noun, respectively (cf. the operation of derelationalization illustrated on
p. 11). This is shown in S1, an alternative representation of the lower half of T2. The curved arrow
34 Croft 1991, ch. 3.2.2 hesitates in positing modification as a propositional operation on a par with reference
Christian Lehmann, The nature of parts of speech 23
reads ‘governing’. The constructions on the right-hand side of the equal sign provide a source for
the lexicalization of new members of the categories on the left-hand side.
S1 Subclassification according to governing relationality
verb intransitive verb = transitive verb
adverb adverb = adposition
adjective absolute adjective = relational adjective
noun absolute noun = relational noun
Further differentiation may refer to the type of complement governed by these words. Just as a verb
may govern either a noun phrase or a subordinate clause, the same is true for an adposition. An
adposition governing a clause is a subordinative conjunction.35
126.96.36.199 Conceptual relationality as a cognitive basis of parts of speech
Modifying conceptual relationality is the most important cognitive basis of adjectival and adverbial
modifiers, thus, indirectly, of adjectives and adverbs. Governing conceptual relationality is the most
important cognitive basis of plurivalent verbs, adpositions (including subordinating conjunctions),
relational adjectives and relational nouns. As is evident, this comprises most of the major parts of
speech except the primary ones, noun phrase / (absolute) noun and verb phrase / (intransitive) verb.
As was shown in §4.3, these have their functional basis elsewhere, viz. in the propositional
operations of reference and predication. Reference is an exophoric relation, in other words, it
involves no conceptual relationality of referential expressions and, consequently, no syntagmatic
structural relations. The relational function of the noun (phrase) is therefore a purely negative one:
that is the part of speech that lacks any such function.
The case of predication is less straightforward. Apart from avalent predications of the type ‘it is
raining’, a predicate is attributed to a referent. However, there is no particular dependency (or other)
relation destined to be the structural reflex of the predicative relation. Instead, there are, even within
one language, more than one structural manifestation of this relation, depending on the categorial
nature of the predicate. For nominal predicates, some kind of equative construction may be used,
establishing just a sociative relation between the subject and the predicate. With adjectival and
adverbial predicates, their modifying potential may be used, and such predicates may then differ
from modifiers only by word order or prosody. For verbal predicates, the case is most complicated
because mirroring the bipartite semantic structure of a referent and a predication in a verbal clause
requires introducing a binary subdivision among the verbal dependents, with one of them being the
subject and the others being oblique. If the verb has valency, then that subject is governed. And
again depending on the language, one of these subject-predicate constructions may be used as a
model for any or all of the others.36 In other words, while the propositional act of predication
indirectly provides the communicative function for the part of speech ‘verb’, there is no conceptual
relationality corresponding to this. Predication differs in this conceptually from the relational
functions of modification and government.
35 One of the first to postulate this is Jespersen (1924:88f).
36 The subject relation in SAE languages is a peculiar combination of modification and government; see
Lehmann 1983, §3.2.
Christian Lehmann, The nature of parts of speech 24
This result is a facet of a theoretical framework for the parts of speech which provides a set of
different – partly mutually independent, partly interconnected – motivations for them, such that the
motivation of one part of speech may be composed of a subset of these factors differing from the
motivation of the next part of speech.
4.5 Combining communicative and cognitive criteria
As communicative functions of parts of speech, we have identified the propositional functions of
reference and predication. As their cognitive functions, to the extent they have any, we have the
relational functions of modification and government and the degrees of time-stability. Now these
parameters are logically independent of each other. They might be conceived as orthogonal,
creating a cross-classification of expressions that designate some kind of concept and fulfill some
propositional function (Croft 1991:53). However, communication and cognition go hand in hand in
language, and thus there is one kind of concept particularly apt for functioning in either of the two
propositional operations. For each of these two associations of a type of concept with a
propositional operation, there is a syntactic category. Finally, where categorial uniformity obtains,
there is a word class instantiating, at the lexical level, each of these two syntactic categories. This
yields the two cross-level associations whose communicative aspect already appears in T3:
•Functional bases of the noun:
•Referring expressions are categorized as noun phrases. The noun is the lexical representative
of the noun phrase. Its primary function therefore is to form the basis of a referring
•Concepts of the highest time-stability (objects) lend themselves most easily to reference.
Therefore, a noun phrase and, derivatively, a noun typically designate an object.
•Functional bases of the verb:
•Predicating expressions are categorized as verb phrases. The verb is the lexical
representative of the verb phrase. Its primary function therefore is to form the basis of a
•Concepts of the lowest time-stability (events) lend themselves most easily to predication.
Therefore, a verb phrase and, derivatively, a verb typically designate an event.
In other words, the communicative functions of reference vs. predication map, on the one hand, on
the two cognitive functions of high vs. low time-stability and, on the other hand, on the nominal vs.
verbal syntactic categories. The association of high time-stability with the nominal category, and of
low time-stability with the verbal category, is therefore not direct, but mediated by the
Most if not all languages have the noun and the verb at the poles of the scale of time-stability.
The majority, however, does not exhaustively divide the continuum up between these two word
classes, but leaves room in the middle for one or two additional, adjective-like categories. Some
published accounts of these cross-level associations of parts of speech (e.g. Croft 1991, Lehmann
1995) therefore include the adjective at an intermediate position on the time-stability scale. This
move is in consonance with a theory that treats modification as the third propositional operation
beside reference and predication. It is not taken up here, for the following reasons: First, as argued
in §4.4.3, modification is on the same level as and contrasts minimally with government. If
government is a syntactic relation and not a propositional operation, then so is modification.
Second, modification subdivides into adverbal and adnominal modification, yielding adverbials as
Christian Lehmann, The nature of parts of speech 25
adverbal modifiers and adjectivals as adnominal modifiers. These differ only by the criterion of the
category of the modified. There is, on this basis, no sufficient reason to accord adjectives a
privileged status in the theory over adverbs. Third, one might draw the conclusion from this that
propositional operations as a basis for major parts of speech should be complemented by conceptual
relationality as their secondary basis, and that therefore the triple ‘noun – adjective – verb’ should
be extended by including the adverb as the fourth and equal partner. While this will be done in the
hierarchy of parts of speech to be proposed in §7, adverbs have nothing to do with time-stability.
That is, prototypical adverbs (notions primarily categorized as adverbs in many languages), such as
fast, hard, can simply not be assigned a position on that scale. The reason for this appears to be the
following: The concepts on the time-stability scale can be predicated on first-order objects and then
characterize such an object in a more or less time-stable way.37 Adverbials, however, make no
predication on first-order entities and instead on second-order entities. The essential parameter for
the concepts providing such predicates is yet to be found; it is not time-stability.
Moreover, the class of adverbs is utterly heterogeneous: an adverb may modify a verb, an
adjective, a sentence, another adverb and (in German at least) even a noun. A distributional
approach will come up with different classes of adverbs which have little in common (s. Pecoraro &
Pisacane 1984 for Italian and Cinque 1999 for some more languages). Consequently, there is no
conceptual core to this traditional word class. And if one limits the analysis to modal adverbs, as is
sometimes done, one has made an antecedent semantic delimitation, so that the question of a
common conceptual core of the class is then no longer an empirical one. The approach to be taken
is a semasiological analysis of each distribution class of adverbs, esp. the adverbal adverbs.
The discussion has made it clear that the semantic force of a part-of-speech category is
derivative by a couple of intervening steps. The propositional functions are fulfilled primarily not
by words, but by components of information structure and of syntactic structure. These are typically
represented by nominal and verbal phrases, and these finally may shrink down to nouns and verbs.
These are entities belonging to levels that differ in nature. Cognitive structures exert even less
determining force on part-of-speech systems: time-stability is only a factor that favors primary
categorization of certain concepts in parts of speech motivated by other forces, and conceptual
relationality comes into play only at lower levels of the part-of-speech system. Consequently word
classes only conserve traces of the semantic force associated with cognitive and communicative
motivations. The labile character of the categorial meaning of word classes observed empirically in
§4.4.2 follows from the indirect character of the cognitive and communicative motivation for word
classes established here.
5 Combining formal and functional analysis
At several points (especially in §§1.1 and 4.4.1), it was argued that since there is no biunique
mapping of meaning onto expressions in language, the semasiological and onomasiological
approaches taken in §§3 and 4, resp., are mutually independent; both have to be taken wherever
meaningful linguistic phenomena are at stake and instantiate hybrid (form-function) concepts. On
the other hand, the mapping is not entirely arbitrary. At least two correspondences between the
formal analysis and the functional analysis should be noted.
In §3.2, dependency relations were used to establish ranks of major parts of speech. By these
formal criteria, verbs, nouns and modifiers are ranked in this order. This is not exactly mirrored, but
37 See Lyons 1977, ch. 11.3 for the ontology of “naïve realism”.
Christian Lehmann, The nature of parts of speech 26
easily compatible with the result of the functional approach presented in §4: the primary parts of
speech according to functional criteria are noun and verb, while modifiers are secondary. In both
approaches, holophrastic words (§4.2) remain outside the ranking.
Secondly, in §3, it was seen that a structural analysis of parts of speech has to concentrate on
their syntagmatic relations, since they do not bear paradigmatic relations to each other. In §4 it was
seen that a functional analysis has to concentrate on the communicative functions of parts of speech,
since their cognitive correlates are derivative. These two findings hang together at an abstract level.
As explained in §4.1, cognition means grasping the world by systematizing it in terms of concepts.
This involves arranging them on the paradigmatic axis of the system. Communication, on the other
hand, means creating community among the interlocutors by orienting their awareness to the same
ideas over a stretch of time. This involves arranging these ideas on the syntagmatic axis of the
message. This makes us understand that communicative operations and the categories involved in
them have their primary reflex on the syntagmatic axis. Thus, given the primary motivation of parts
of speech by their communicative functions, the concentration of their formal analysis on the
syntagmatic axis follows. The character of the theory brought to bear on our subject is a
consequence of the fact, already underlined in §2, that we are accounting for parts of speech, not for
parts of the system.
The kinds of motivation relevant for the formation of part-of-speech systems may now be
summarized in S2:
S2 Functional and formal factors conditioning parts of speech
S2 is just meant to graphically summarize the functional and formal bases of parts of speech
discussed so far. §6 will add nothing that could be integrated into S2. The different role of the
factors mentioned is not shown, nor are particular form-function associations such as the mapping
of conceptual relationality on grammatical relationality.
6 Major and minor parts of speech
The terms ‘major vs. minor word class / part of speech’ have been around at least since they were
introduced in Lyons 1968, ch. 7.1.3. There the noun, verb, adjective and adverb are called major
parts of speech, whereas preposition and conjunction are mentioned as minor parts of speech. Ch.
Christian Lehmann, The nature of parts of speech 27
9.5.2 then says that these notions may be explicated as open vs. closed classes of elements, and this
is offered, at the same time, as an operationalization of the distinction between lexical items and
grammatical items. We thus get the correspondences shown in T5:
T5 Major and minor parts of speech (Lyons 1968)
major class = lexical class = open class
minor class = grammatical class = closed class
Finally, a closed set is defined (p. 436) “as one of fixed, and usually small, membership”, while “an
open set is one of unrestricted, indeterminately large, membership”. These definitions are
sufficiently precise to overthrow at once the assignments made to the supercategory ‘minor class’.
Adpositions and conjunctions are open classes in all modern European languages and certainly in
many other languages. For instance, Lehmann & Stolz 1991:14f enumerates 140 German
adpositions, with no claim to completeness.38 This is much more than many languages can summon
for adjectives or verbs. The number of members is, however, just a consequence of the productivity
of the class: there are regular operations of syntax and word formation to generate new adpositions.
Productivity is the decisive criterion for the distinction between major and minor class. This
criterion is, in turn, operationalized as requiring that there be, at the synchronic stage in question, at
least one word formation process that generates members of the class in question. In short, a major
class is one that may be enriched by word formation, and a minor class is one that cannot.
By this criterion, it turns out that the association of major classes with lexical items (content
words) and of minor classes with grammatical items (formatives, function words) can be upheld in
principle (i.e. barring very small and unproductive lexical classes like Yukaghir adjectives or
Jaminjung verbs39): there are no operations of word formation to generate grammatical formatives.
If a certain grammatical class receives new members, this may happen by processes of grammati-
calization and other kinds of grammatical change. These are not rules that would be part of the
language system, and instead they change the language system and are therefore usually accounted
for in a diachronic perspective. However, since these processes are universal, the distinction
between major and minor classes is universal (cf. Bisang 2011, §4). In other words, all languages
have content words and function words, though the latter may differ in their degree of
grammaticalization and, thus, constitute typological differences among languages.40
38 The majority of these are, to be sure, secondary adpositions such as angesichts ‘in the face of’ and even
phrasal adpositions such as in bezug auf ‘with respect to’. However, there is no way of keeping these out. For
one thing, one would then, by analogy, have to exclude phrasal verbs from the class of verbs and phrasal
compounds from the class of nouns. For another, there is no other category available in which such
expressions could fall.
39 The latter may not even be an exception, as they approach grammatical formatives in their status, like the
light verbs of other languages.
40 A language all of whose grammatical formative candidates are lexical items in an incipient phase of
grammaticalization (the purely isolating language), and a language all of whose grammatical formatives are
bound morphemes (the purely synthetic language), are not meant to be excluded on theoretical grounds.
Some existent languages come close to these extremes. They just testify to the dynamic character of the
distinction relevant here.
Christian Lehmann, The nature of parts of speech 28
The dynamic relationship between lexical and grammatical classes of words has the following
1. A closed class is fed by an open class by grammaticalizing the latter’s members. Now the
distribution of an item does not change categorically by its grammaticalization (Lehmann 2005,
§4); it just gets less sensitive to semantic properties of its context. This means essentially that a
certain closed class is the most grammaticalized subclass of a certain open class; it is that
subclass of the latter with the most general distribution.
2. Given the gradual nature of grammaticalization, major and minor word classes are not
categorically different. The familiar word classes like nouns, verbs, adjectives, adverbs,
adpositions, conjunctions do not divide up into a major and a minor subset (cf. Lehmann 2002).
Instead, there is a set of word classes such as the ones named; and ea c h of these contains, as a
proper subset, a minor word class. The remainder may then be called a major class. This is
illustrated by T6.
T6 Lexical and grammatical subclasses of English word classes
category example category example
noun person pronoun one
adjective red pro-adjective such
verb exist pro-verb be
adverb behind pro-adverb there
preposition notwithstanding grammatical preposition of
conjunction supposing (that) grammatical conjunction that
interjection gosh! grammatical interjection yes
It should be clear that T6 contains only a subset of the lexical and grammatical categories even of
that one language. Here a few more grammatical categories will be mentioned which, in different
languages, have the status of minor classes grammaticalized from some major class.41 The class of
pro-verbs at least includes verbs of existence, positionals, copulas, auxiliaries, modals, light verbs
and coverbs. Numeral classifiers and possessive classifiers are minor classes corresponding to
absolute and to relational nouns, respectively, as their feeding major class. Quantifiers may be
treated as the grammatical counterpart to numerals. They share with the latter their indeterminacy in
terms of distribution class. Where they are of nominal character, at least one relevant grammatical-
ization relation is amply documented, viz. the grammaticalization of the numeral ‘one’ to the
For each class of grammatical formatives, there is at least one major class which feeds it
through grammaticalization. Grammaticalization is not among the forces creating and delimiting
parts of speech adduced in the preceding sections and assembled in S2. It differs categorically from
these, just as some of the factors joined in that diagram differ categorically among each other. This
is just one more occasion to recall that part-of-speech systems owe their existence and shape to a set
of incommensurable factors and are therefore internally heterogeneous. That is, however, not to say
that the factors assembled in S2 are irrelevant for the minor parts of speech. Instead, with increasing
41 A rather comprehensive overview of the ways in which minor classes evolve from major classes by
grammaticalization is found in Heine & Kuteva 2007, ch. 2.
Christian Lehmann, The nature of parts of speech 29
grammaticalization, their motivation in terms of cognitive and communicative functions fades
away, while their motivation by purely formal factors remains and increasingly becomes their only
7 A hierarchy of parts of speech
We have now assembled the theoretical basis for a dynamic model of the development of a part-of-
speech system. S3 is the first step in such a model. It starts from the holophrastic word and
comprises the primary parts of speech – noun and verb – and those secondary parts of speech
directly dependent on these by modifying them – adjective and adverb, resp.
S3 Hierarchy of parts of speech I: major parts of speech
The idea of S3 and the following diagrams is the gradual building of a part-of-speech system from
top to bottom (similar models are proposed in Hengeveld 1992[N]:47-72, Anward 2000 and Gil
2000, §3). The nature of this model is systematic-genetic. The principle leading from top to bottom
is a dynamism of increasing system complexity. The model accounts for the stepwise extension of a
part-of-speech system in the sense of Jakobson’s (1968) unilateral foundation. A given part of
speech presupposes the existence in the system of the parts of speech higher up in the tree.
However, apart from S5, the model does not determine the specific way in which any of the
secondary elements in the tree come into existence. That is, it does not say that each such element
evolves out of its respective superordinate element. Although such a model has repercussions in
typology, diachrony and language acquisition, these are not direct, since there are many intervening
Given that, for syntactic aspects, the present approach essentially relies on concepts of
dependency grammar, it is insufficient to account for relations whose head is a complex phrase. The
sociative particles mentioned in §3.2 – no matter whether they are lexical or grammatical
42 For example, Lehmann 2010 argues that the primary motivation of Yucatec Maya numeral classifiers is not
a functional one (“individuation of the concept designated by the counted noun”), but a structural one (to
serve as a prop for affixal numerals).
Christian Lehmann, The nature of parts of speech 30
formatives – would have to be provided for at the clause or even sentence level. Here, it must
suffice to mention them beside the category of adverb.
Expanding a part-of-speech system by secondary parts of speech like modifiers, adpositions,
conjunctions and different kinds of particles, as will be done in the remaining two schemata, means
moving down the hierarchy from the universal to the language-specific. While the primary parts of
speech find an extra-grammatical motivation in terms of propositional acts, those secondary parts of
speech are motivated with respect to the primary ones. This kind of motivation refers not so much
to cognitive or communicative functions of language and more to formal constraints on a semiotic
system and to system-internal functions.
In a second step, the major classes developed on the basis of the holophrastic word get an
additional governing slot motivated by conceptual relationality. This leads to the subclasses
proceeding from the main classes in S4.
S4 Hierarchy of parts of speech II: relational parts of speech
The third logical step is the evolution of minor parts of speech out of these major ones by
grammaticalization. In this case, a given minor part of speech does develop out of its corresponding
major part of speech, although other sources are not excluded. S5 gives an overview of the system
proposed so far, leaving a few items out for want of space.
Christian Lehmann, The nature of parts of speech 31
S5 Hierarchy of parts of speech III: minor parts of speech
The diagram symbolizes, in the vertical dimension, the hierarchical relations between the major
classes, and in the horizontal orientation, the specific dependence of a certain category on another.
In a dynamic perspective, word classes may now be seen as the product, to a large extent, of the
omnipresent processes of lexicalization and grammaticalization:
•Apart from other operations of enrichment of the lexicon, the major classes are fed by
lexicalization of word-formations and syntactic constructions,
•the minor classes are fed by grammaticalization of members of the major classes.
This dynamic model may generate quite diverse sets of synchronic systems. Among the perhaps
less expected outcomes is a language that possesses some minor class without having the
corresponding major class. This may happen if the minor class first comes into existence by the
grammaticalization of elements of some major class, but the major class gets lost afterward. Thus, it
is possible for a language to have exclusively simple adpositions, but no complex adpositions and
no productive process for their formation. Classical Latin is a case in point. An admittedly extreme
case would be a language that acquires pronouns – by grammaticalization of nouns or noun phrases
– and then gives up its category of nouns. At the ensuing synchronic stage, it would have no lexical
nouns, but only pronouns. That is the situation claimed to obtain in Hengeveld’s (1992[P]:69)
Tuscarora and Sasse’s (1993) Cayuga.
Finally, as we saw in §2, words are at an intermediate level of the complexity hierarchy of
meaningful units. The transition from major to minor class words points towards the next lower
level, which is the level of morphemes, including affixes. A word class system not only has to be
Christian Lehmann, The nature of parts of speech 32
related, as we have done, to the next higher level system, which is the system of syntactic
categories, but also to the next lower level system, the system of inflectional categories. Auxiliaries
grammaticalize to conjugation affixes, postpositions grammaticalize to case suffixes, determiners
grammaticalize to definiteness markers, and so forth. A complete account of the word-class system
of a language would include the systems of both of these adjacent levels.
The raison d’être of parts of speech lies in the semiotic necessity of structuring the message in
terms of categories and relations in order to assure compositionality. The categories and relations
are therefore related to the syntagmatic axis and devised in such a way that the categories of
syntagmatically related elements complement each other to form a higher whole.
Given these premises, the question arises which these categories are. There is no universal set
of them; instead only the principles underlying their development are universal. Compositionality
itself is not an absolute goal, but subservient to mutual understanding. There are situations where
compositionality is unnecessary even at the highest syntactic level. For such situations, all
languages have holophrastic utterances, which involve no categorization, made up of interjections
Compositionality is necessary to the extent that inferencing is insufficient to create the intended
sense. Linguistic structure guarantees compositionality and thus guides inferencing; but the extent
to which it does so and the functional domains in which it does so are largely language-specific
(LaPolla 2003). Consequently, there are differences among languages in the extent to which they at
all categorize expressions in structural terms, in the structural level – between sentence and root – at
which they do so, and in the communicative and cognitive categories which they use to functionally
motivate the structural categorization. However, this variation is guided by a couple of universal
First, understanding is essentially holistic. In other words, if I understand your utterance, then
neither of us will care whether I understood its components. Therefore linguistic structure is most
concerned about securing understanding at the highest level and is most compositional at that level.
Therefore languages have means to mark off categories and relations at the sentence level. They
may or may not do so at lower levels, including, in particular, the stem level. In that sense, fixing
parts of speech at the stem level (in the form of word classes) amounts to downscaling the solution
of a task. That is a strategy available at the typological level which may suit the type of the
particular language (cf. §2). This projection of syntactic categories into the lexicon happens by the
joint action of grammaticalization and lexicalization.
Second, since all of this concerns the structure of the message (as opposed to the system), the
functional principles filling up structural categories with content are more principles of
communication than of cognition. The highest-level communicative operations are the propositional
operations of reference and predication. Therefore much of linguistic structure is oriented towards
these; and that is true for parts of speech, too. Therefore all languages distinguish the categories of
referring and of predicating expressions. If these are marked off as structural units, they yield the
syntactic categories of the noun phrase and the verb phrase and, in a derivative way, their lexical
manifestations, the noun and the verb. These two word classes are populated with members
essentially on the basis of the cognitive category of time-stability.
From there on, extension of the part-of-speech system is guided by universal and then
increasingly language-specific structural constraints. The next step in the extension of the system is
Christian Lehmann, The nature of parts of speech 33
concerned with expanding the range of concepts used in reference and predication. All languages
can do this, some languages, however, only at the level of modifying syntactic operations of
attribution and adjunction. Now if a language opts for categorial uniformity, it needs modifiers.
Here is another field where it can be economic to store prefabricated modifiers as a lexical class.
This yields adjectives and adverbs, which make use of the structural device of modifying
relationality. Similarly, the structural device of governing relationality is put to use in order to
create subclasses of the classes generated so far which differ in their valency and thus afford more
flexibility in syntagmatic combination. This then opens a rich field of further subdivision according
to grammatical selection restrictions and, thus, to the subcategory of the complement.
Finally, the overall burden of categorization and relationalization cannot be born by the lexicon
alone. There must be flexibility in recategorizing items and putting them into new relations. Apart
from the purely isolating language, all languages derive minor classes from the lexical classes by
grammaticalization. Their members help in pinning down the category that an expression belongs
to, thus introducing redundancy into the message. Some of these minor classes, like demonstrative
and interrogative pronouns, are again motivated by universal principles of communication. In
principle, however, their organization is a matter of language-internal structure.
The notion constituting the title of the present article – the nature of parts of speech – is not a
unified notion. They are of very different nature.
Anward, Jan 2000, “A dynamic model of part-of-speech differentiation.” Vogel & Comrie (eds.) 2000:3-45.
Anward, Jan & Moravcsik, Edith & Stassen, Leon 1997, “Parts of speech. A challenge for typology.”
Linguistic Typology 1:167-183.
Bally, Charles 1921, Traité de stylistique française. Heidelberg: C. Winter (Indogermanische Bibliothek, 3).
Beck, David 2002, The typology of parts of speech systems. The markedness of adjectives. New York &
London: Routledge (Outstanding Dissertations in Linguistics).
Bhat, D.N.S. 1994, The adjectival category. Criteria for differentiation and identification. Amsterdam &
Philadelphia: J. Benjamins (Studies in Language Companion Series, 24).
Bisang, Walter 2011, “Word classes.” Song, Jae Jung (ed.), The Oxford handbook of linguistic typology.
New York & Oxford: Oxford University Press; 280-302.
Broschart, Jürgen 1997, “Why Tongan does it differently: categorial distinctions in a language without nouns
and verbs.“ Linguistic Typology 1: 123-165.
Cinque, Guglielmo 1999, Adverbs and functional heads. A cross-linguistic perspective. New York etc.:
Oxford University Press (Oxford Studies in Comparative Syntax).
Comrie, Bernard 1976, Aspect. An introduction to the study of verbal aspect and related problems.
Cambridge etc.: Cambridge University Press.
Croft, William 1991, Syntactic categories and grammatical relations. The cognitive organization of
information. Chicago: Chicago University Press.
Derbyshire, Desmond C. 1979, Hixkaryana. Amsterdam: North-Holland (Lingua Descriptive Studies, 1).
Dik, Simon C. 1989, The theory of functional grammar. Dordrecht: Foris (Functional Grammar Series, 9).
Dixon, Robert M.W. 1976, “Where have all the adjectives gone?“ Studies in Language 1: 19-80.
Enfield, Nick 2004, “Adjectives in Lao”. Dixon, Robert M.W. & Aikhenvald, Alexandra Y. (eds.), Adjective
classes. A cross-linguistic typology. Oxford: Oxford University Press; 323-347.
Evans, Nicholas D. 2000, “Word classes in the world's languages.” Booij, Geert & Lehmann, Christian &
Mugdan, Joachim (eds.), Morphologie. Ein internationales Handbuch zur Flexion und Wortbildung. 1.
Christian Lehmann, The nature of parts of speech 34
Halbband. Berlin & New York: W. de Gruyter (Handbücher zur Sprach- und Kommunikations-
wissenschaft, 17.1); 708-732.
Gil, David 2000, “Syntactic categories, cross-linguistic variation and universal grammar.” Vogel & Comrie
Givón, Talmy 1979, On understanding grammar. New York etc.: Academic Press (Perspectives in
Neurolinguistics and Psycholinguistics).
Haspelmath, Martin 2012, "How to compare major word-classes across the world's languages." Graf,
Thomas & Paperno, Denis & Szabolcsi, Anna & Tellings, Jos (eds.), Theories of everything. In honor of
Ed Keenan. Los Angeles: University of California at Los Angeles (UCLA Working Papers in
Linguistics, 17); 109-130.
Heine, Bernd & Kuteva, Tania 2007, The genesis of grammar. A reconstruction. Oxford etc.: Oxford
Hellwig, Birgit 2007, "Fieldwork among the Goemai in Nigeria: discovering the grammar of property
expressions." Aikhenvald, Alexandra Y. (ed.), Focus on: linguistic fieldwork. Berlin: Akademie-Verlag
(Sprachtypologie und Universalienforschung, 60/1); 67-80.
Hengeveld, Kees 1992, Non-verbal predication. Theory, typology, diachrony. Berlin & New York: Mouton
Hengeveld, Kees 1992, “Parts of speech.” Fortescue, Michael & Harder, Peter & Kristoffersen, Lars (eds.),
Layered structure and reference in a functional perspective. Papers from the functional grammar
conference, Copenhagen, 1990. Amsterdam & Philadelphia: J. Benjamins (Pragmatics & Beyond, New
Series, 23); 29-55.
Himmelmann, Nikolaus P. 2007, “Lexical categories and voice in Tagalog.” Austin, Peter K. & Musgrave,
Simon (eds.), Voice and grammatical relations in Austronesian languages. Stanford, CA: Center for the
Study of Language and Information (Studies in Constraint-Based Lexicalism); 247-293.
Hopper, Paul J. & Thompson, Sandra A. 1984, “The discourse basis for lexical categories in universal
grammar.” Language 60:703-752.
Jackendoff, Ray S. 1977, X-Syntax. A study of phrase structure. Cambridge, MA & London: MIT Press
(Linguistic Inquiry Monographs, 2).
Jakobson, Roman 1968, Child language, aphasia and phonological universals. The Hague & Paris: Mouton
(Janua Linguarum Series Minor, 72).
Jespersen, Otto 1924, The philosophy of grammar. London: G. Allen & Unwin.
Kuryłowicz, Jerzy 1936, “Dérivation lexicale et dérivation syntaxique (contribution à la théorie des parties
du discours).” Bulletin de la Société de Linguistique de Paris 37:79-92.
LaPolla, Randy J. 2003, “Why languages differ: variation in the conventionalisation of constraints on
inference.” Bradley, David & LaPolla, Randy & Michailovksy, Boyd & Thurgood, Graham (eds.),
Language variation. Papers on variation and change in the Sinosphere and in the Indosphere in honour
of James A. Matisoff. Canberra: Pacific Linguistics; 113-144.
Lehmann, Christian 1983, "Rektion und syntaktische Relationen." Folia Linguistica 17:339-378.
Lehmann, Christian 1990, “Strategien der Situationsperspektion.” Sprachwissenschaft 16:1-26.
Lehmann, Christian 1991, “Predicate classes and PARTICIPATION.” Seiler, Hansjakob & Premper, Waldfried
(eds.), Partizipation. Das sprachliche Erfassen von Sachverhalten. Tübingen: G. Narr (Language
Universals Series, 6); 183-239.
Lehmann, Christian 1995, “Latin predicate classes from an onomasiological point of view". Longrée,
Dominique (ed.), De usu. Études de syntaxe latine offertes en hommage à Marius Lavency. Louvain:
Peeters (Bibliothèque des Cahiers de l'Institut de Linguistique de Louvain, 70); 163-173.
Lehmann, Christian 2002, “New reflections on grammaticalization and lexicalization.” Wischer, Ilse &
Diewald, Gabriele (eds.), New reflections on grammaticalization. Amsterdam & Philadelphia: J.
Benjamins (Typological Studies in Language, 49); 1-18.
Christian Lehmann, The nature of parts of speech 35
Lehmann, Christian 2005, “Wortarten und Grammatikalisierung”. Knobloch, Clemens & Schaeder, Burkhard
(eds.), Wortarten und Grammatikalisierung. Perspektiven in System und Erwerb. Berlin & New York:
W. de Gruyter (Linguistik - Impulse & Tendenzen, 12); 1-20.
Lehmann, Christian 2008, “Roots, stems and word classes.” Studies in Language 32:546-567.
Lehmann, Christian 2010, “On the function of numeral classifiers.” Floricic, Franck (ed.), Essais de
typologie et de linguistique générale, offerts au Professeur Denis Creissels. Lyon: Presses de l'École
Normale Supérieure; 435-445.
Lyons, John 1968, Introduction to theoretical linguistics. Cambridge: Cambridge University Press.
Lyons, John 1977, Semantics. 2 vols. Cambridge: Cambridge University Press.
Meier, Georg F. 1979, “Wortklassen als Basis jeder Grammatiktheorie. Zur Methodologie der Wort -
klassifikation in asiatischen und europäischen Sprachen.” Zeitschrift für Phonetik, Sprachwissenschaft
und Kommunikationsforschung 32: 24-35.
Mithun, Marianne. 2000. "Noun and verb in Iroquoian languages: Multicategorisation from multiple
criteria." Vogel & Comrie (eds.) 2000: 397-420.
Mosel, Ulrike & Hovdhaugen, Even 1992, Samoan reference grammar. Oslo: Norwegian University Press.
Peterson, John 2005, “There’s a grain of truth in every “myth”, or, Why the discussion of lexical classes in
Mundari isn’t quite over yet.” Linguistic Typology 9(3):391-441.
Pecoraro, Walter & Pisacane, Chiara 1984, L'avverbio. Bologna: Zanichelli.
Prandi, Michele 2004, The building blocks of meaning. Amsterdam & Philadelphia: J. Benjamins (Human
Cognitive Processing, 13).
Sapir, Edward 1921, Language. An introduction to the study of speech. New York: Harcourt, Brace &
Sasse, Hans-Jürgen 1993, „Das Nomen - eine universale Kategorie?“ Sprachtypologie und Universalien-
forschung 46: 187-221.
Simone, Raffaele 2006, “Classi di costruzioni.” Grandi, Nicola & Iannàccaro, Gabriele (eds.), Zhì. Scritti in
onore di Emanuele Banfi in occasione del suo 60° compleanno. Cesena: Caissa Italia (Serie linguistica);
Smith, Mark 2010, “Pragmatic functions and lexical categories.” Linguistics 48:717-777.
Thomas von Erfurt 1972, Grammatica speculativa. Edited with an introduction, a translation, and a
commentary by G.L. Bursill-Hall. London: Longman.
Vogel, Petra M. 2000, “Grammaticalisation and part-of-speech systems.” Vogel & Comrie (eds.) 2000:259-
Vogel, Petra M. & Comrie, Bernard (eds.) 2000, Approaches to the typology of word classes. Berlin & New
York: Mouton de Gruyter (Empirical Approaches to Language Typology, 23).
Wetzer, Harrie 1995, Nouniness and verbiness. A typological study on adjectival predication. Nijmegen:
Universiteitsdrukkerij Nijmegen (PhD thesis).
Wierzbicka, Anna 1986, “What’s in a noun? (or: how do nouns differ in meaning from adjectives?)”. Studies
in Language 10:353-389.