Against (lexical-)categorial typology: Why school grammars are basically right

Syntax textbooks often emphasize that word classes must be identified by grammatical tests, rather than by their meanings – in contrast to traditional school grammars and basic pedagogical texts, which often still say, for example, that “Nouns are a part of speech that indicate people, places, things, or ideas“.

Here I point out that school grammars are not as wrong as linguists often say, and that word categories identified on the basis of grammatical diagnostics are not as general as they are often thought to be. General cross-linguistically oriented syntax can hardly rely on such diagnostic tests, so for general-theoretical purposes, semantic definitions are still needed.

In my 2023 paper on word class universals (in van Lier’s handbook of word classes) I state this insight rather bluntly (§(2.)5):

(1) A noun is an object-denoting root, a verb is an action-denoting root, and an adjective is a property-denoting root.

This may be rather puzzling at first sight, so I explain it again in this blogpost, from a perspective that is a bit different from the perspective adopted in the 2023 paper. The perspective here is the much more general rejection of categorial typology, as stated in (2).

(2) Cross-linguistic grammatical research cannot be based on categories of grammar.

Thus, it is not just diagnostic-based word classes (or “lexical categories”, as they are also called; cf. Baker & Croft 2017) that are not a good basis for general grammar, but more generally, the structural grammatical classes (or categories) that linguists identify in languages are not a good basis for typological comparison. Instead, comparison relies on substantive comparative concepts: semantic concepts, purely formal concepts (i.e. shape-based concepts), and complex concepts built upon these (Haspelmath 2010). These substantive concepts allows us to define kinds of forms in such a way that they can be identified across languages (e.g. clitics, affixes, person indexes, inflectional affixes, which I have written about in earlier work). But these substantively defined kinds of forms do not (necessarily) correspond to distributionally identified classes/categories in particular languages.

Thus, what I’m saying is that school grammars are right about identifying nouns, verbs and adjective as general categories, but of course, within each particular language, we need diagnostics in order to establish language-particular categories/classes. Now since grammar teaching in schools has traditionally been concerned with multiple languages (e.g. French and Latin, German and French, or English and Spanish), it makes good sense for teachers to emphasize those properties that are shared among several languages. For more rigorous descriptive approaches (e.g. American descriptivism, e.g. Gleason 1961), a different set of criteria will of course be relevant.

For over two decades, typologists have emphasized the language-particular nature of grammatical categorization (e.g. Dryer 1997; Croft 2001; Haspelmath 2007), and this insight has been gaining more and more ground. But one consequence of it has not been fully realized yet: It does not make much sense to typologize languages according to their lexical categorizations. Many linguists still think, e.g., that there are two types of languages: those that make a verb-adjective distinction, and those that lack a verb-adjective distinction. But there is no general basis for such a typology, because “verb” and “adjective” cannot be defined formally (via morphosyntactic diagnostics) as general categories. Thus, a question such as “Do all languages have a distinction between nouns and verbs?” (e.g. Davis et al. 2014) cannot be answered (as I emphasize in §(2.)4 of my 2023 paper). “Distinction” is simply too vague, so the only way in which we can understand the terms noun, verb and adjective independently from particular systems is through their meanings. (And if understood as aprioristic categories of universal grammar, they cannot be tested rigorously.)

The usual objection to a semantic definition is that not all nouns in English denote “people, places or things”, and that not all adjectives in English denote “properties”. For example, replacement denotes an action (not a thing), and parental denotes a kind or a relation (not a property). This means that these forms do not fall under the comparative-concept definition in (1), and this may seem unsatisfactory. But it must be recalled that comparative research is always partial – one can compare languages (and most other things) only with respect to certain properties, not with respect to everything. Event nominalizations such as replace-ment show a wide range of different properties across languages, and they are often treated as inflected forms, not as distinct noun lexemes. They can probably be left aside when formulating generalizations about nouns across languages. And relational adjectives such as parent-al seem to be uncommon in the world’s languages, and they are thought to be a peculiarity of Romance and Slavic languages. So for good reasons, they have not played a major role in typological work on adjectives. The definitions in (1) thus concern the shared core of nouns, verbs and adjectives as we can compare them across languages (on shared-core definitions, see §3.2 of my paper on definining the word).

That different languages have different criteria that might be relevant for categorization is brought home by an interesting paper on adpositions for the same handbook (van Lier (ed.) 2023), by Borja Herce. He examines adpositions (in the broad sense) in Hungarian, Basque, German and Spanish, and finds that in all these languages, about 20 different subclasses of adpositions can be identified if 8-11 different criteria are taken into account (shown in the columns in the figures below).

Herce uses these results to argue for a multivariate approach to typology (Bickel 2010), which is another term for the comparative concepts argued for by Dryer, Croft and Haspelmath (Bickel’s 2007 “fine-grained variables”; see also his “typological metalanguages” in Bickel 2015).

Herce’s paper refers to DeLancey’s (2005) interesting work on the lack of adpositions in Klamath, where the general problem is stated very clearly at the outset (emphasis added):

“In traditional grammar syntactic categories like noun and preposition are given partially or completely notional definitions. Throughout most of the 20th century orthodox linguists shunned such definitions, insisting that syntactically relevant categories could only be defined on purely distributional grounds. This approach has the advantage of preventing the careless analyst from introducing into the analysis of one language categories which are assumed to exist on the basis of their occurrence in some other. On the other hand, it notoriously makes it difficult to compare the categories of different languages, and thus to come to any objective conclusions about the universality of any particular category.“ (DeLancey 2005: 185)

For comparative purposes, and for general-theoretical claims (which must be based on universals; Haspelmath 2021b), we thus need function-based definitions. For example, we find that nouns (as defined above) have a tendency to require extra coding (a “copula”) when used predicatively, and verbs (as defined above) have a tendency to require extra coding (a “nominalizer”) when used referentially. Such extra coding elements are called function indicators in my 2023 paper, using a term that is inspired by Croft (1991). Croft also identified the reasons why the function indicators are distributed the particular ways in which we observe them in languages: They occur when the association between pragmatic function (predication, reference, modification) and semantic class (action, object, property) is unexpected, for reasons of efficient coding (Haspelmath 2021a).

DeLancey (2005) notes that

“the problem with distributional definitions is that they have no explanatory value – the [distributional] facts about English which let us identify an unknown form as a verb or noun [regardless of its meaning] do nothing to explain why there should be such things as verbs or nouns.” (DeLancey 2005: 186)

For functional explanations, we need to identify grammatical properties that are universal across languages (because the functions of languages are themselves universal), so it is only universal functional properties such as semantic class or pragmatic function that are truly important.

Now what does this mean for adpositions, the topic of Herce’s paper? Actually, Herce is not concerned with the difference between adpositions and case affixes, and in fact he includes case markers that are written as affixes in Hungarian and Basque. So all the points he makes apply to the more comprehensive concept of of flags. It does seem that basically all languages have flags, i.e. either case suffixes or adpositions (or both). If we look at the combination of the two WALS maps by Dryer (2005) (on adpositions) and Iggesen (2005) (on case markers), we see that almost all languages have either both (the blue dots), or only adpositions (the red dots), or only case (the yellow dots).

Thus, we can say that flags are (almost) universal. Now one might ask: Isn’t this a categorial statement, similar to the statement that all languages have nouns and verbs (which I criticized above)? After all, I just said that typology shouldn’t be category-based.

The answer is no: The notion of ‘flag’ is not identified by distributional criteria, which are by definition language-particular. Like the definitions of noun, verb and adjective in (1) above, the definition of flag involves a notional component (indicating a semantic role or syntactic function) and a formal component (being a bound morph):

(3) A flag is a grammatical marker (= a bound morph that is not a root) that occurs on a nominal and that indicates its semantic role or syntactic function.

While not all the elements that Herce (2023) discusses for his four languages (Hungarian, Basque, German, Spanish) are flags by this definition, most of them are, but they fall into a range of language-particular categories, as Herce shows. Like nouns, verbs and adjectives, flags are defined in substantive terms, and comparative studies of flags do not compare categories.

Thus, I think that quite generally, we can say that morphosyntactic typology compares kinds of forms across languages, defined by their substantive properties (relating to shape, position, and meaning/function), not distributional categories (or classes).

Addendum (May 25th)

This is the corresponding map based on Grambank (






Baker, Mark C. & Croft, William. 2017. Lexical categories: Legacy, lacuna, and opportunity for functionalists and formalists. Annual Review of Linguistics 3. 179–197.

Bickel, Balthasar. 2007. Typology in the 21st century: Major current developments. Linguistic Typology 11(1). 239–251.

Bickel, Balthasar. 2010. Capturing particulars and universals in clause linkage: A multivariate analysis. In Bril, Isabelle (ed.), Clause-hierarchy and clause-linking: the syntax and pragmatics interface, 51–102. Amsterdam: Benjamins.

Bickel, Balthasar. 2015. Distributional Typology. In Narrog, Heiko & Heine, Bernd (eds.), The Oxford handbook of linguistic analysis. Oxford: Oxford University Press. (10.1093/oxfordhb/9780199677078.013.0046)

Croft, William. 1991. Syntactic categories and grammatical relations: The cognitive organization of information. Chicago: University of Chicago Press.

Croft, William. 2001. Radical Construction Grammar: Syntactic theory in typological perspective. Oxford: Oxford University Press.

Davis, Henry & Gillon, Carrie & Matthewson, Lisa. 2014. How to investigate linguistic diversity: Lessons from the Pacific Northwest. Language 90(4). e180–e226.

DeLancey, Scott. 2005. Adpositions as a non-universal category. In Frajzyngier, Zygmunt & Hodges, Adam & Rood, David S. (eds.), Linguistic diversity and language theories, 185–202. Amsterdam: Benjamins.

Dryer, Matthew S. 1997. Are grammatical relations universal? In Bybee, Joan L. & Haiman, John & Thompson, Sandra A. (eds.), Essays on language function and language type: Dedicated to T. Givón, 115–143. Amsterdam: Benjamins.

Dryer, Matthew S. 2005. Order of adposition and noun phrase. In Haspelmath, Martin & Dryer, Matthew S. & Gil, David & Comrie, Bernard (eds.), The world atlas of language structures, 346–349. Oxford University Press. (

Gleason, Henry Allan. 1961. An introduction to descriptive linguistics. New York: Holt, Rinehart and Winston.

Haspelmath, Martin. 2010. Comparative concepts and descriptive categories in crosslinguistic studies. Language 86(3). 663–687. (doi:10.1353/lan.2010.0021)

Haspelmath, Martin. 2021a. Explaining grammatical coding asymmetries: Form-frequency correspondences and predictability. Journal of Linguistics 57(3). 605–633. (doi:10.1017/S0022226720000535)

Haspelmath, Martin. 2021b. General linguistics must be based on universals (or nonconventional aspects of language). Theoretical Linguistics 47(1–2). 1–31. (doi:10.1515/tl-2021-2002)

Haspelmath, Martin. 2023. Word class universals and language-particular analysis. In van Lier, Eva (ed.), The Oxford handbook of word classes, 15–40. Oxford: Oxford University Press. (

Herce, Borja. 2023. Adpositions. In van Lier, Eva (ed.), The Oxford handbook of word classes, 420–442. Oxford: Oxford University Press.

Iggesen, Oliver A. 2005. Number of cases. In Haspelmath, Martin & Dryer, Matthew S. & Gil, David & Comrie, Bernard (eds.), The world atlas of language structures, 202–205. Oxford: Oxford University Press. (

van Lier, Eva (ed.). 2023. The Oxford handbook of word classes. Oxford: Oxford University Press.

Cite this blog post
Martin Haspelmath (2024, May 18). Against (lexical-)categorial typology: Why school grammars are basically right. Diversity Linguistics Comment. Retrieved June 13, 2024, from

2 thoughts on “Against (lexical-)categorial typology: Why school grammars are basically right

  1. I wonder what your definitions of word classes can tell us about word pairs such as “stupid” and “idiot”, which at least at the language-specific level belong to different word classes (adjective and noun, respectively) but which seem to have similar meanings.

    • Nothing, really – at the language-particular level, we all agree that English “stupid” is an English Adjective, and English “idiot” is an English Noun. Note that my definition is a shared-core definition, so it does not apply to the fringes. Wierzbicka and Langacker (and many others, no doubt) have said insightful things about English minimal pairs like “stupid/idiot”, but I don’t think these help us understand the general picture. In general, I think that languages have a very large amount of parochial features, and that linguists tend to greatly overestimate the significance of general linguistics to linguistic analysis.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.