Three key notions of linguistics: Lexemes, inflection, and derivation

My paper on inflection and derivation as traditional comparative concepts has just been published (in Linguisticsin open access, 2024). What made me revisit this old question? (which had occupied me back in 1996, and of course in my 2002 morphology textbook, Chapter 4)

Linguists have many technical terms that they rely on in their everyday work, such as lexeme and morpheme, but when pressed about their meanings, they are often at a loss to give precise answers, or even vague answers that sound satisfying. For example, the definitions of lexeme and word given in Peter Matthews’s dictionary of linguistics (2007) are hardly satisfying because they do not say how one identifies “words in the lexicon” as opposed to “word forms”:

Lexeme: A distinction is often drawn between lexemes as words distinguished in the lexicon (e.g. the verb ‘to sing’) and the individual word forms that they subsume (past tense sang, present participle singing, etc.).”

Word form: The form of a specific word, either phonetic or orthographic. Distinguished as such from the word as a lexical unit or lexeme: e.g. ran is one of a set of word forms (run, runs, ran, running) each of which is a ‘form of’ the lexeme ‘(to) run’.”

In a naïve conception of language systems, there are (i) grammatical rules and (ii) a lexicon as a simple list (without regularities), and on such a view, one would be forced to decide whether a “word” is part of the lexicon (= a lexeme) or is a “form” that is “subsumed” by a lexeme. But linguists have long passed such a stage, and there are at least three “official” views which are incompatible with this naïve conception: the Lexicalist view on which there are many “lexical rules” in addition to ordinary grammatical rules; the Distributed-Morphology view on which there is not one single lexicon (but “lexical information”is “distributed” over multiple parts of the architecture); and the Parallel-Architecture view, on which there is a continuum from fully regular to fully specific stored elements (e.g. Jackendoff 2023).

So once we leave the naïve view, what do we do with the distinction between lexemes and word-forms? For a long time, I thought that it didn’t really matter that we had no good way of distinguishing between them, because there is so much uncertainty anyway – after all, we do not really know how to distinguish generally between words and phrases, and between clitics (= bound words) and affixes (Haspelmath 2011). One may want to say that the foundation of the lexeme/word-form contrast is the distinction between derivation and inflection, but how exactly do we distinguish between these two subtypes of morphology? This has often been discussed (Scalise 1988; Plank 1994; Spencer 2016), but the distinction has appeared to be just as elusive as the distinctions between affixes, words, and phrases.

So we have no clear definitions of some of our key terms – but is this a situation that we must (and can) live with?

I no longer think so: It became clear to me at some point that the distinctions between grammar and lexicon, between word-forms and lexemes, and between inflection and derivation are powerful stereotypes. Regardless of the technical details that particular approaches may offer, each generation of linguists acquires these stereotypes, and it takes a lot of effort to unlearn them – and of course, not everyone has the time for such a process of unlearning. (Psycholinguists may want to focus on experimental designs, corpus linguists on statistical techniques, and language documenters on primary data collection and primary description.)

But science means that we leave our everyday stereotypes behind, and it can hardly be a good idea to introduce the younger generation to stereotypical notions which we know are not well-founded. So I think that we need non-stereotypical (precise) definitions of the terms, which roughly correspond to the general idea that is shared among linguists, also because the earlier literature is full of references to “lexemes”, “inflection”, and “derivation” – we cannot do without these terms for the time being (and we also need Wikipedia articles for them, and so on). Moreover, it might be that these notions are not completely unnatural after all, maybe as “prototypes” if not as sharply delimited concepts. Or maybe as “focal zones” on a continuum. (See Dressler 1989 for a clear articulation of inflection and derivation as “prototypes”.)

In the case of “grammar vs. lexicon”, it is easy to identify the historical source of this stereotype: It comes from our habit of describing rules and words in different kinds of books (grammar books vs. dictionaries, as I discuss in §3 of the new paper). And presumably the distinction between inflection and derivation comes from these books: Inflectional paradigms are the patterns that are described in grammar books, and derivational rules are regularities that one can formulate over entries in a dictionary. Thus, while inflectional paradigms have been a mainstay of grammar for many centuries, derivational rules have come into full focus only in the 20th century.

Such historical considerations are interesting, but how do they help us with the stereotypes? I think they make us realize that the definitions need not be “natural”, in the sense of corresponding to deep theoretical notions. Thus, in my new paper I define inflection and derivation as follows, in an admittedly somewhat unnatural way (pp. 23-24):

  • inflection:
    An inflectional construction is a construction in which an inflectional meaning (role, person, number, gender, tense, mood, evidentiality, polarity) is expressed by an affix or nonconcatenatively.
  • derivation:
    A derivational construction is a construction in which a meaning other than an inflectional meaning is expressed by an affix or nonconcatenatively.

So what I propose is to define derivation as non-inflectional affixation, and inflection as affixation expressing a meaning from one of eight specified semantic domains. This is similar to the proposal according to which a root is defined as a morph denoting an action, an object or a property, as it also makes reference to three specified meanings. Clearly, these eight meanings are somewhat arbitrary, but as I note in the paper: Attempts at providing natural (non-arbitrary) definitions for inflection-derivation have failed (just as they have failed for the notion ‘root’), and I now think that it’s better to provide n arbitrary definition than none at all.

Once we have a definition of inflection (and derivation), we can provide a definition of lexeme (p. 27):

  • lexeme:
    A lexeme is the set of forms that minimally contain the same lexeme-stem, or one of its suppletive counterparts, and that may only contain inflectional affixes in addition.

A lexeme is thus a set of forms, not a kind of form, but the definition is based on the definition of lexeme-stem, which is a kind of form (p. 26):

  • lexeme-stem:
    A lexeme-stem is a root, or a compound, plus possibly derivational affixes, that can combine with inflectional affixes if the language has any but does not contain any inflectional affixes.

To understand the aspects of this definition that are not immediately apparent, it will be helpful to refer to my 2023 paper “Defining the word” (where I also give a definition of compound). A larger point to notice is that these definitions (unlike those given by Matthews) do not make reference to “lexicon” (or “lexical”), a term that has no clear meaning.

This blogpost is not the place for more detailed technical discussion, but the basic message that I want to get across is: The key terms inflectionderivation and lexeme do not seem amenable to simple intuitive definitions that are at the same time precise. They can be defined, but in an unnatural way that many experienced linguists will find unintuitive or downright strange. This seems to mean that they have no deeper significance for general linguistic theorizing, and that theories should not rely on them. The terms will be used in the future just as they have been used in the past, in language teaching contexts, in descriptive grammars, in psycholinguistics, in computational linguistics, and so on. But it seems best not to try to build deeper explanatory stories on them, as they seem to be based primarily on our traditional ways of describing languages in two kinds of books (dictionaries and grammar books).


Dressler, Wolfgang U. 1989. Prototypical differences between inflection and derivation. Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung 42(1). 3–10.

Haspelmath, Martin. 1996. Word-class-changing inflection and morphological theory. In Geert Booij & Jaap van Marle (eds.), Yearbook of morphology 1995, 43–66. Dordrecht: Kluwer.

Haspelmath, Martin. 2002. Understanding morphology. London: Arnold.

Haspelmath, Martin. 2011. The indeterminacy of word segmentation and the nature of morphology and syntax. Folia Linguistica 45(1). 31–80.

Haspelmath, Martin. 2023. Defining the word. WORD 69(3). 283–297.

Haspelmath, Martin. 2024. Inflection and derivation as traditional comparative concepts. Linguistics

Jackendoff, Ray. 2023. The Parallel Architecture in language and elsewhere. Topics in Cognitive Science n/a(n/a).

Matthews, P.H. 2007. The concise Oxford dictionary of linguistics (Oxford Paperback Reference). Oxford, United Kingdom: Oxford University Press.

Plank, Frans. 1994. Inflection and derivation. In R.E. Asher (ed.), Encyclopedia of Language and Linguistics, 1671–1678. Oxford: Pergamon.

Scalise, Sergio. 1988. Inflection and derivation. Linguistics 26(4). 561–582.

Spencer, Andrew. 2016. Two morphologies or one? Inflection versus word-formation. In Andrew Hippisley & Gregory Stump (eds.), The Cambridge handbook of morphology, 27–49. Cambridge: Cambridge University Press.

Cite this blog post
Martin Haspelmath (2024, January 8). Three key notions of linguistics: Lexemes, inflection, and derivation. Diversity Linguistics Comment. Retrieved June 13, 2024, from

12 thoughts on “Three key notions of linguistics: Lexemes, inflection, and derivation

  1. ”as they seem to be based primarily on our traditional ways of describing languages in two kinds of books (dictionaries and grammar books)”.
    So you are assuming that this tradition is arbitrary and does not reflect any properties of the languages speakers of which are responsible for its emergence? In other words, for instance Hesych – who in the antiquity collected rare words of Ancient Greek – just arbitrarily decided to collect strings of morphemes of various length (while it would have been better to rather collect rare roots or rare noun phrases)?

    • Yes, I think that this tradition is somewhat arbitrary. Hesychius might as well have collected fixed expressions in general (and indeed these are popular items of collection, e.g. Of course, shorter fixed expressions may well have somewhat greater salience, but this would be a matter of degree.

      • Did Hesych also collected (strings) of morphemes shorter than what we usually call a word-form? I.e. something like ‘dzel = bal in Arcadian’ for the verb meaning ‘throw’ or perhaps ‘ei = i in Arcadian’ for the dative singular ending of athematic noun stems?

        • I don’t know, but probably he should have 🙂 It may well be an accident that the Western tradition worked with words and larger units until 1880 (when Baudouin came up with the “morpheme” idea). Didn’t Panini identify morphemes even in antiquity?

          • But Panini also identified sounds of speech and word-forms. It’s certainly possible to identify units smaller than word by means of linguistic analysis (recurrent parts of many words), as it is possible to identify units smaller than morph. I don’t see why this would imply that words don’t exist in the relevant language. To be sure, word boundaries are not always obvious in a language but so are the morph boundaries.

    • That Panini identified morphs doesn’t imply that words don’t exist, and yes, there is some uncertainty about morph boundaries. However, a morph is a logically necessary concept if one works with the idea of a linguistic expression or form (namely, a morph is a minimal form). There is nothing necessary about the “word” concept, and if we define it as in my 2023 paper, it is a truly strange kind of thing. I suspect that it’s a spelling-based artifact.

      • You are suggesting to work only with concepts which are logically necessary (for a decsription to be possible), starting with the minimal form? I.e. morphs + assembling rules but no words? Then, you are probably also against such concepts as ‘syllable’, ‘molecule’, etc.?

        • I think that traditional concepts such as “letter” or “syllable” or “word” should be treated with skepticism – they have a long history and are entrenched in our practice, but we may need to update it. We successfully replaced “letter” by “sound segment” (or phoneme), and we partly succeeded in replacing “word” by “morph(eme)”. Maybe “syllable” also needs to be replaced (see – I don’t know. It’s very likely that higher-level concepts will also turn out to be important, but each concept that is not logically necessary needs to be justified.

  2. Interesting post as always!
    Assuming that the distinction between inflection and inflection indeed has no deeper significance for general linguistic theorizing, do you think it makes sense for introductory Linguistics classes to continue introducing this distinction at all (with the caveat that this is merely a historical distinction with no real justification?)

    Or would it be better to just avoid this confusing distinction altogether, and reserve it for a “history of linguistics” class or possibly some other more advanced classes?

    • Ideally, the latter. But I’m not optimistic that this will happen soon, so for the time being, I think it’s better to make people aware that the distinction is not well-founded. When I got the reviews for my paper, one of the reviewers actually commented that this was not news, and I don’t claim that it’s news – Plank (1994) made the same point fairly clearly, and morphologists tend to know about this. But it hasn’t made its way to general awareness yet, so I was happy that the paper was accepted despite this valid point.

  3. The distinction between closed and open word classes often comes up in trying to distinguish between grammar and lexicon. That distinction is not well-defined. Have you written about that particular issue?

    • Yes, that’s a distinction that comes up in many textbooks and elsewhere, but how do you identify a “class”? The major classes have many different types of subclasses, and many of these are closed (e.g. in many languages: ditransitive verbs, inalienable nouns, pluralizable nouns; in English: strong verbs, nouns with umlaut plurals, etc.). So I don’t think it’s particularly useful.

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.