Two senses of “lexicon”: The inventorium and the lexemicon

[Update Jan 2024: There is now a draft of a paper that makes the same proposal in a more formal way, available on Lingbuzz:]

This blogpost proposes two new terms for what Mark Aronoff  (1988) called “idiosyncratic-lexical” items and “categorial-lexical” items: the inventorium is the set of all morphs, constructions and phrasemes of a language (i.e. all idiosyncratic meaningful elements), and the lexemicon is the set of all lexemes of a language, i.e. the members of the major lexical categories noun, verb and adjective. I think that by using these two terms (and one further term, as discussed below), we can avoid confusions that have often been a problem.

In his little-known (1988) paper, but also in the first chapter of his well-known 1994 book, Aronoff points out that these two sets need not coincide, and that one of the reasons for the confusions around “lexicalism” is the non-distinction of the two concepts. His “word-based morphology” (Aronoff 1976) was really meant as “lexeme-based morphology”, covering word-formation of major lexical categories. The lexeme-formation rules of a language are open-ended, and most of the lexemes that they can create are not in most speakers’ mental lexicons (just as most of the sentences that they can create are not stored anywhere). In his 1994 book, Aronoff proposes that the terms “lexicon” and “lexical” should only be used for the idiosyncratic lexicon (Bloomfield’s (1933: 274) “list of basic irregularities”), i.e. for what I call inventorium here. For the lexemicon, he has no name, and he actually rejects earlier suggestions to distinguish the concepts by using the terms “permanent lexicon” (for the inventorium) and “conditional lexicon” (for the lexemicon):

“…we are not dealing with two types of lexicons, one permanent and one conditional, but with two largely unrelated theoretical entities: an arbitrary list of signs that a speaker happens to hold in memory [=inventorium, MH] as opposed to a well-defined infinite set [=lexemicon, MH]… The term lexicon should … be understood, in this work at least, as referring only to the permanent lexicon: the list of all idiosyncratic signs, regardless of their category or complexity [=inventorium]. The endless list of all lexemes [=lexemicon], by contrast, will remain nameless.” (Aronoff 1994: 22)

But I would think that, if one wants to contrast two different concepts that were confused because they have been known by the same name, it is not a successful strategy to propose a name only for one of the concepts. In fact, in his earlier (1988) paper, Aronoff did propose a term for lexical in the sense of lexemic, namely umlical. An umlic is an “uninflected member of a major lexical category” (UMMLC), so that a “lexical rule” that derives lexemes could be called an “umlical rule”. But as we saw, a few years later Aronoff gave up this term and said that the lexemicon should “remain nameless” – which is a pity, because the term lexemicon is readily available. (Maybe in 1994, the suffix -icon had not become productive yet, because neither were constructicons well-known at the time, nor did the Concepticon exist.)

For me, the immediate impetus for writing this blogpost did not actually come from Aronoff’s old paper or his book, but from a very nice paper on compounds and similar binominals (see Pepper 2020 for the latter term) that was published by Livio Gaeta and Davide Ricca in Rivista di Linguistica in 2009. Gaeta & Ricca (G&R) note that the discussion of forms such as those in (1)-(3) is often confused because compounds are expected to be both morphologically generated and part of the “lexicon”.

(1) English life vest, or German Schwimm-weste [float-vest]

(2) German Entscheidungs-ort [decision-place], or English apple-juice seat

(3) Italian luna di miele [moon of honey] ‘honeymoon’

According to G&R, only the forms in (1)-(2) are compounds in the sense of being the outputs of a morphological rule of compounding, while only (1) and (3) are lexicalized, and in this sense part of the “lexicon”. Compounds such as Entscheidungsort ‘decision place’ are created on the fly in a fully regular way, like syntactic phrases, but according to morphological (rather than syntactic) regularities. Linguists have often regarded Italian forms such as luna di miele ‘honeymoon’ as compounds as well, because they have a meaning that is not (fully) compositional, just like morphological compounds such as life vest. Quite rightly, Gaeta & Ricca say:

“We have to distinguish carefully between what we want to attribute to the lexicon (i.e. to consider lexicalized), and what we would like to see as the result of a morphological operation” (2009: 38)

So it seems clear that what G&R mean by “lexicon” is what I call inventorium, because this is the set of idiosyncratic meaningful elements of a language. By contrast, the lexemicon of a language is open-ended, just as the set of sentences, and contains ad hoc compounds such as German Entscheidungsort, or English China report. G&R note that Italian V+N compounds may also be of this type, and they give examples such as astensione salva-Prodi [abstention save-Prodi] ‘Prodi-saving abstention’ (Romano Prodi was an Italian prime minister).

One of the reasons I started to think seriously about the terminology was a recent invitation to a conference on “Grammaticalization and lexicalization in Greek and Latin” at UCL London. I was probably invited because of my earlier work on grammaticalization, but for this conference, I decided to focus on “lexicalization” (Haspelmath 2022). I realized that this term has multiple meanings (just like the terms “lexicon” and “lexical”), with one prominent meaning better described as lexification (I also gave a talk about this phenomenon earlier this year, in Uppsala). The type of change that leads to relatively opaque conventional compounds such as life vest (or even honeymoon) is best called inventorialization, because it involves a change from an expression that is created on the fly (such as China report) to an expression with idiosyncratic, not fully predictable properties. Inventorialization can affect lexemes (derived lexemes such as re-place, which no longer means ‘place again’ or ‘place back’; or compound lexemes such as life vest), or it can affect phrases (as in idiomatic phrasemes such as pay attention). Inventorialization is thus a change in which an element is added to the inventorium, and if this is a lexeme, it must have been part of the lexemicon before. (As an aside, it can be noted that there is also a change that we may call lexemization, where a syntactic phrase is turned into a non-inventorialized lexeme; this is a sub-case of univerbation, as recently discussed by Lehmann 2020.)

I hope that what I said is clear enough so far, because the plot thickens even more: The inventorium as I defined it is not exactly what Aronoff says in the quotation above (“an arbitrary list of signs that a speaker happens to hold in memory”). In the definition that I propose here, memory storage plays no role, but the defining feature of inventorial elements is that they are idiosyncratic. This is in line with Aronoff’s characterization of the first sense of “lexical” as “what is not predictable but must be memorized” (1988: 3), and with Bloomfield’s definition of the lexicon as “the total stock of morphemes of a language” (though Bloomfield did not include constructions and idiomatic expressions).

Now crucially, there is a well-known distinction between “what must be memorized” (= what is idiosyncratic) and “what a speaker happens to hold in memory”. The latter, often called the mental lexicon, contains not only all the idiosyncratic forms and constructions of a language (when constructions are included, the term extended lexicon has been used, as in Jackendoff & Audring 2020), but also many regular combinations that are used frequently and thus entrenched. Since Bybee (1985), linguists have generally been aware that the kinds of diachronic changes that we find (plus many results from psycholinguistics) can only be explained if (some) regular forms are stored (by some or many speakers). The reason why the (extended) mental lexicon (what I called the mentalicon in my London talk) cannot be equated with the inventorium is that the inventorium is a property of a language (i.e. what must be learned), while the mental lexicon is a feature of a particular speaker’s knowledge of language – and different speakers may of course differ in the degree to which they have entrenched/stored regular forms. These two aspects of a language are often conflated, but they must of course be distinguished sharply (Chomsky’s distinction of E-language vs. I-language roughly covers the same concepts). If lexicalization is a diachronic process, it must be a change to the inventorium of a language, not to any speaker’s mentalicon. So while I was generally impressed by Gaeta & Ricca (2009), I cannot agree with the following passage from this paper:

“Notice that we do not intend to consider as lexicalized only items which display any sort of (formal or semantic) idiosyncrasy. Rather, we are inclined to consider that being lexicalized or stored depends very much on behavioural properties like token frequency and/or naming force.” (2009: 38)

Here they equate “lexicalized” with “stored”, but the latter is a property of the cognitive states of particular speakers, not a property of a language. However, “lexicalization” is generally regarded as a type of diachronic change, so it must be equated with inventorialization. (The counterpart process at the level of an individual’s knowledge of language might be called “mentalicization”, and while the inventorialization of a composite expression surely depends on some speakers having mentalicized it, this is by no means a sufficient condition.)

While many authors conflate the notions of inventorium and mentalicon, Aronoff (2018) actually contains an interesting passage where he talks about both, in different terms [emphasis mine]:

“The study of word-formation is concerned at its core with how words are created, how they are formed, and how they are added to the language. Unlike sentences, words, once formed, accumulate, and this accumulated storehouse has an effect on new words. Words accumulate both in the mental lexicon of an individual speaker and in the collective lexicon of the larger linguistic community.” (Aronoff 2018: 12)

What Aronoff calls “the collective lexicon of the community” must be what I propose to call inventorium, even though Aronoff does not mention the key property of (Bloomfieldian) idiosyncrasy in this context. There cannot be any “collective mental lexicon”, and what is (in some sense) “stored by the community” can only be those forms and constructions that are not completely predictable from other forms and constructions: The inventory that is needed by the community is a set of social norms or conventions (as I discussed in this earlier blogpost), not a set of cognitive routines.

(Postscriptum: This blogpost got longer than was originally intended, and it now actually talks about three senses of “lexicon”: inventorium, lexemicon, and mentalicon. But the distinction between inventorium and lexemicon should be seen as the current focus; more will have to be said about how they relate to the mentalicon.)


Aronoff, Mark. 1976. Word formation in generative grammar. Cambridge, MA: MIT Press.

Aronoff, Mark. 1988. Two senses of lexical. 1–11. (

Aronoff, Mark. 1994. Morphology by itself: Stems and inflectional classes. Cambridge, MA: MIT Press.

Aronoff, Mark. 2018. Morphology and words: A memoir. In Bonami, Olivier & Boyé, Gilles & Dal, Georgette & Giraudo, Hélène & Namer, Fiammetta (eds.), The lexeme in descriptive and theoretical morphology, 3–17. Berlin: Language Science Press.

Bloomfield, Leonard. 1933. Language. New York: H. Holt and Company.

Bybee, Joan L. 1985. Morphology: A study of the relation between meaning and form. Amsterdam: Benjamins.

Gaeta, Livio & Ricca, Davide. 2009. Composita solvantur: Compounds as lexical units or morphological objects? Rivista di Linguistica 21(1). (

Haspelmath, Martin. 2022. Against “lexicalization” (and what to replace it with). London. (

Jackendoff, Ray & Audring, Jenny. 2020. Relational Morphology: A cousin of Construction Grammar. Frontiers in Psychology 11. (

Lehmann, Christian. 2020. Univerbation. Folia Linguistica 54(s41–s1). 205–252. (doi:10.1515/flih-2020-0007)

Pepper, Steve. 2020. The typology and semantics of binominal lexemes: Noun-noun compounds and their functional equivalents. Oslo: University of Oslo. (PhD dissertation.) (

Cite this blog post
Martin Haspelmath (2022, September 6). Two senses of “lexicon”: The inventorium and the lexemicon. Diversity Linguistics Comment. Retrieved June 13, 2024, from

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.