We are all constructionists

Many linguists use ideological-sounding labels to identify themselves (or their colleagues), and I keep wondering about the purpose and content of these labels – what exactly is a cognitive linguist , for example? Is it someone who shares Lakoff’s (1991) “cognitive commitment” (to make their account of human language accord with what is generally known about the mind and brain from other disciplines”)? But why “commitment” – isn’t this simply a general and uncontroversial principle of science? I have attended quite a few “cognitive linguistics” conferences (and have published in, and reviewed for, the journal Cognitive Linguistics). Clearly, there is a real-world community of practice that is associated with this label, where interesting work is happening. But science is by definition universal and should avoid additional ideologies, so what, if anything, is behind this label, other than a tradition going back to George Lakoff?

In this blogpost, I address a related (and somewhat simpler) question, namely what distinguishes constructionism. While this label has been used by Adele Goldberg in a broader sense (Goldberg 2013; see below), I am simply asking here whether constructions play an important part in some or all approaches to grammar. And as in the case of structuralism (see my 2020 blogpost “We are all structuralists”), my preliminary conclusion at this stage is that we are all constructionists.

Not everyone uses the label “construction”, but it seems to me that the Main Idea of constructionism is just as uncontroversial as the Main Idea of structuralism. I formulate it as follows here:

The Main Idea of constructionism

The relationship between grammatical forms and meanings is not necessarily one-to-one, so that we need additional mechanisms (constructions) that go beyond the basic mechanisms of morph combination and composition.

Goldberg & Jackendoff (2004: 533-534) put it as follows: “Constructions are like traditional idioms: they are listed in the lexicon with a syntactic structure, a meaning, and (where there is a special morpheme) a partial phonology.” In other words: It is not generally sufficient to have a list of morphs with their meanings and to combine (or merge) these morphs.

Now in many simple cases, complex expressions do indeed consist of transparent morph combinations whose semantic composition is straightforward, e.g. [freshA waterN]NP, child-hood, tree-s, [theyPRO openV [theD windowN]NM]CS. Textbooks tend to present such cases first, so the phenomena that go beyond these clear cases may appear to be unusual (especially to beginners), and it may seem that they present a deep theoretical problem. However, we have always known that simple morph combination is not enough, and linguists have always used additional mechanisms.

Here is a reminder of a few well-known phenomena that go beyond simple morph combination and require constructions: idiomatic expressions (e.g. once in a blue moon, to hit the nail on the head); meaningful inflectional zeroes (e.g. she hit-s vs. she hit-Ø); semi-productive combinations (e.g. arriv-al, *depart-al); non-compositional derivation (e.g. sheep-ish ‘sheep-like; embarrased’); discontinuous inflection (e.g. German ge-geb-en ‘given’); cumulative inflections and portmanteau elements (e.g. Russian stol-ov [table-GEN.PL] ‘of the tables’; French aux ‘to the PL’, instead of à les); non-existent bases (e.g. gorge-ous); directionality mismatches (e.g. assassin ‘someone who assassinates’, linguist ‘expert in linguistics’); sister derivatives (e.g. ambi-tion/ambiti-ous; but there is no *ambit)); meaningful word orders (e.g. Russian pjat’ knig ‘five books’; knig pjat’ ‘about five books’); meaningful valency constructions (e.g. English double object construction [V NP NP] with transfer meaning; Goldberg 1992; English resultative construction V NP Adj, e.g. hammer the metal flat ‘cause the metal to be flat by hammering’); morphomic stems (e.g. Latin lauda-t- [passive participle], as used in lauda-t-urus ‘who will praise (active, future)’); suppletive inflectional forms (e.g. good/bett-er); transfixing inflection of the Semitic type (e.g. Arabic kataba ‘wrote’; kutiba ‘was written’); empty elements (pure stem-forming elements in inflection; periphrasis forms in syntax).

In Berkeley Construction Grammar (Fillmore 2013), the focus has mostly been on syntactic patterns of English, and in morphological approaches (especially Booij 2010; Audring & Jackendoff 2020), a few more languages have been brought into play (see also Booij (ed.) 2018). However, the kinds of phenomena mentioned in the preceding paragraph are rampant in the world’s languages, so there is no reason why the idea of constructions should be restricted to English, Dutch and German departments. And in fact, when we look at the earlier morphological literature (e.g. Nida 1949; Robins 1959; Matthews 1974), these phenomena are discussed and theorized prominently. At the time, the term “construction” played no role, but these earlier linguists were fully aware that simple morph combination is not sufficient. The term that was long used was paradigm, or paradigmatic relationship, which is not really different from construction.

A paradigm is a set of semantically and formally related expressions. The term paradigm comes from inflectional paradigms, but it has been extended to derivational paradigms (Štekauer 2014), and if one makes no strict distinction between morphology and syntax, there are also syntactic paradigms. In fact, the relationships between related inflected forms (writ-es/writ-ing), derivational forms (organiz-er/organiz-ation), and syntactic forms (the tree/this tree/our tree) are completely parallel. Some paradigms can be eliminated by reducing them to morph combinations, but many cannot. Even a simple paradigm like English they play/they play-ed involves a complication: The form without the Past-tense marker -ed is a Present-tense form (they play), with a distinct Present-tense function that cannot simply be described as “non-past”. One could describe this by means of a “zero sign”, but this would be an assimilation to the simplistic morph combination mechanism – it would thus hide the fact that what we have here is a paradigmatic – or constructional – phenomenon (I have argued against “zero markers” here).

Morphologists have a fairly long history of discussing paradigmatic/constructional phenomena. In the 1980s, two widely cited books were Plank’s (1981) Morphologische (Ir-)Regularitäten, and van Marle’s (1984) On the paradigmatic dimension of morphological creativity (see also van Marle 2000). Both discussed paradigmatic/constructional phenomena in derivational morphology in great detail, and these ideas were never forgotten. Geert Booij was evidently influenced by them, and as late as in 2008, he called his approach Paradigmatic morphology(Booij 2008). There is no difference between these Dutch/German ideas and what is called “constructionist approaches” nowadays. In Britain and the United States, this way of thinking has been called “word and paradigm” (made prominent by Robins 1959; and especially by Matthews 1974), and has been applied especially to inflectional paradigms. Since Anderson (1992), it became more and more prominent, and after Stump (2001), the need for constructional/paradigmatic descriptions has come to be known as “realizational”. Works such as Spencer (2001), Anderson (2015) and Blevins (2016) kept pointing to the “problems with the morpheme theory” (Spencer 2001: 285), advocating a paradigmatic/constructional (or “abstractive”, in Blevins’s terms) alternative. (Likewise, Aronoff’s 1976 idea of “word-based” morphology also argied against “morpheme-based” morphology.)

But in fact, the so-called “morpheme theory” was never adopted by anyone, and while many linguists since the 1930s felt that a “morph(eme)” concept was useful, it was always (or at least since Hockett 1947) clear that morphological patterns are more complex than is suggested by simplistic morph concatenation. The idea of morphology as the “syntax of words” was promoted by some in the 1980s (e.g. Selkirk 1982), but again: it has always been clear that syntax requires constructions. In simplistic cases, syntactic phrases can be described as concatenative, but we quickly reach more complex phenomena where this is no longer possible. For example, just as Russian expresses numerical approximation by word order (e.g. pjat’ knig ‘five books’; knig pjat’ ‘about five books’), English expresses polar questions by word order (e.g. She is here vs. Is she here?). Clearly, the question meaning is associated with a construction here, not with a particular morph. If one wants to maintain the fiction of simplistic morph combination, one can of course introduce a “zero question morpheme” (as was done by Katz & Postal 1964; see Baker 1970), but this does not make the analysis any less constructional. The essence of constructionism is the recognition that “we need additional mechanisms (constructions) that go beyond the basic mechanisms of morph combination”, so any zero element is a constructional feature. (And for that matter, any transformational rule is constructional, too, because transformations are additional mechanisms.) Zero elements and constructional schemas (as in Booij 2010) are notationally different, but there does not seem to be any difference in substance.

At this point, some readers may wonder whether I have emptied the notion of a construction (and of constructionist analyses) of all useful content. Well, this view of constructionist analyses certainly does not make it possible to contrast “constructionist” with “generative” approaches, but that was part of the point here. If there is a deep contrast, I would like to know what it is, in a way that I understand and that goes beyond notation. Just as “structuralism” does not contrast well with “generativism” or with “functionalism”, I do not see “constructionism” contrasting transparently with “generativism”. A rich discussion of the relationship between construction grammar and various generative approaches (especially HPSG) is provided by Müller (2021), who summarizes: “HPSG is a Construction Grammar” (and points to Riehemann 1998 as an important step in the development of “construction morphology”)

Goldberg (2013: 15) associates four features (or “tenets”) with her broader notion of “constructionist approaches”:

  1. Grammatical constructions: Phrasal constructions, like traditional lexical items, are learned pairings of form and function.
  2. Surface structure: Grammar does not involve any transformational or derivational component. Semantics is associated directly with surface form.
  3. A network of constructions: Phrasal constructions, words, and partially filled words (a.k.a. morphemes) are related in a network in which nodes are related by inheritance links.
  4. Crosslinguistic variability and generalization: Languages are acknowledged to vary in wide-ranging ways. The crosslinguistic generalizations that do exist are explained by domain-general cognitive processes or by the functions of the constructions involved.

On this list, No. 4 does indeed seem to be a major difference between generative and non-aprioristic linguistics, because generative linguists have traditionally assumed that there is a rich innate blueprint for grammar. But as I noted above, everyone must assume that “phrasal constructions are learned pairings of form and function” (No 1), in one way or another – notations vary widely, but the simplistic model of morph combination and composition is plainly insufficient. The “surface structure orientation” (No 2) is not a substantive difference either, because “underlying structures” and “derivational rules” are just one way of notating constructional patterns. All descriptions need abstractness because grammatical patterns are not simplistic, and unless one thinks that the rule types are innate (see the first sentence of this paragraph), transformations are merely one way of describing the facts.

It may be that the idea of an inheritance network (No 3) is truly different from generative grammar, but it is actually quite doubtful that inheritance networks are sufficient (see Müller & Wechsler 2014 for extensive discussion; also Müller 2021: §1.2.3). Booij & Masini (2015) provide some discussion of “second-order schemas”, and Audring (2019) discusses similar phenomena under the rubric of “sister constructions/schemas”. Basically, all inflectional paradigms are sets of sister constructions, and when Goldberg talks about “a network of constructions”, it is not clear whether this network is necessarily limited to pure inheritance relations. If not, then again the difference seems to be primarily notational.

(It is true that Goldberg’s fifth point about the usage-based nature of competence, that “knowledge of language includes both items and generalizations, at varying levels of specificity”, is never emphasized by generative grammarians, but it seems logically necessary to recognize that some complex forms are stored in memory, so again, the substantive difference is hard to pinpoint. But I leave this matter aside here.)

Thus, we must all be constructionists. I am saying “must” here, because in a sense, it would be desirable to have a simpler model of morphosyntax that makes do with morphs and their combinations. Constructions are powerful devices. If languages did not have constructions, and relied merely on morphs and simple combination patterns (e.g. if sentences were like web URLs, which consist of letters and letter combinations plus a few other symbols, combined according to a few simple rules), grammars would be much simpler. But with the power that constructions provide, languages could be much more complex than they are in reality. This then leads us to the question why many parts of languages seem to approximate the morph-combination model.

I find this a very challenging question, and I feel that we have not even asked it properly. Just arguing against the “morpheme theory” strawman does not seem to lead to significant understanding. If language systems consist of constructions, not of morphs and simple combination rules, then why do many sentences look like they largely consist of morph combinations?

References

Anderson, Stephen R. 1992. A-morphous morphology. Cambridge: Cambridge University Press.

Anderson, Stephen R. 2015. The morpheme: Its nature and use. In Baerman, Matthew (ed.), The Oxford handbook of inflection, 11–33. Oxford: Oxford University Press.

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

Audring, Jenny. 2019. Mothers or sisters? The encoding of morphological knowledge. Word Structure. Edinburgh University Press 12(3). 274–296. (doi:10.3366/word.2019.0150)

Baker, C. L. 1970. Notes on the description of English questions: The role of an abstract question morpheme. Foundations of Language. Springer 6(2). 197–219.

Blevins, James P. 2016. Word and paradigm morphology. Oxford: Oxford University Press.

Booij, Geert E. 2008. Paradigmatic morphology. In Fradin, Bernard (ed.), La raison morphologique. Amsterdam: Benjamins. (https://benjamins.com/catalog/lis.27)

Booij, Geert. 2010. Construction morphology. Oxford: Oxford University Press.

Booij, Geert & Masini, Francesca. 2015. The role of second order schemas in the construction of complex words. In Bauer, Laurie & Körtvélyessy, Lívia & Štekauer, Pavol (eds.), Semantics of complex words, 47–66. Cham: Springer.

Fillmore, Charles J. 2013. Berkeley Construction Grammar. In Hoffmann, Thomas & Trousdale, Graeme (eds.), The Oxford handbook of Construction Grammar. Oxford: Oxford University Press. (10.1093/oxfordhb/9780195396683.013.0007)

Goldberg, Adele E. 1992. The inherent semantics of argument structure: The case of the English ditransitive construction. Cognitive Linguistics 3(1). 37–74.

Goldberg, Adele E. 2013. Constructionist approaches. In Hoffmann, Thomas & Trousdale, Graeme (eds.), The Oxford handbook of Construction Grammar. Oxford: Oxford University Press. (10.1093/oxfordhb/9780195396683.013.0002)

Goldberg, Adele E. & Jackendoff, Ray S. 2004. The English resultative as a family of constructions. Language 80(3). 532–568.

Hockett, Charles F. 1947. Problems of morphemic analysis. Language 23(4). 321–343. (doi:10.2307/410295)

Lakoff, George. 1991. Cognitive versus generative linguistics: How commitments influence results. Language and Communication 11(1/2). 53–62.

Matthews, Peter H. 1974. Morphology. Cambridge: Cambridge University Press.

Müller, Stefan. 2021. HPSG and Construction Grammar. In Müller, Stefan & Abeillé, Anne & Borsley, Robert D. & Koenig, Jean-Pierre (eds.), Head-Driven Phrase Structure Grammar: The handbook (Empirically Oriented Theoretical Morphology and Syntax). Berlin: Language Science Press. (https://langsci-press.org/catalog/book/259)

Plank, Frans. 1981. Morphologische (Ir- )Regularitäten: Aspekte der Wortstrukturtheorie. Tübingen: Narr.

Riehemann, Susanne Z. 1998. Type-based derivational morphology. The Journal of Comparative Germanic Linguistics 2(1). 49–77. (doi:10.1023/A:1009746617055)

Robins, R. H. 1959. In defence of WP (word and paradigm). Transactions of the Philological Society 58(1). 116–144. (doi:https://doi.org/10.1111/j.1467-968X.1959.tb00301.x)

Selkirk, Elisabeth O. 1982. The syntax of words. Cambridge, MA: The MIT Press.

Spencer, Andrew. 2001. The paradigm-based model of morphosyntax. Transactions of the Philological Society 99(2). 279–314. (doi:10.1111/1467-968X.00083)

Štekauer, Pavol. 2014. Derivational paradigms. The Oxford handbook of derivational morphology. Oxford University Press Oxford 354.

Stump, Gregory T. 2001. Inflectional morphology: A theory of paradigm structure. Cambridge: Cambridge University Press.

van Marle, Jaap. 1984. On the paradigmatic dimension of morphological creativity. Dordrecht: Foris. (https://doi.org/10.1515/9783111558387)

van Marle, Jaap. 2000. Paradigmatic and syntagmatic relations. In Booij, Geert E. (ed.), Morphologie / Morphology, 225–234. Berlin: De Gruyter. (https://doi.org/10.1515/9783110111286.1)


9 thoughts on “We are all constructionists

  1. Focusing on this part of your post:

    “For example, just as Russian expresses numerical approximation by word order (e.g. pjat’ knig ‘five books’; knig pjat’ ‘about five books’), English expresses polar questions by word order (e.g. She is here vs. Is she here?). Clearly, the question meaning is associated with a construction here, not with a particular morph. If one wants to maintain the fiction of simplistic morph combination, one can of course introduce a “zero question morpheme” (as was done by Katz & Postal 1964; see Baker 1970), but this does not make the analysis any less constructional. The essence of constructionism is the recognition that “we need additional mechanisms (constructions) that go beyond the basic mechanisms of morph combination”, so any zero element is a constructional feature. (And for that matter, any transformational rule is constructional, too, because transformations are additional mechanisms.) Zero elements and constructional schemas (as in Booij 2010) are notationally different, but there does not seem to be any difference in substance.”

    ***

    I think you miss a basic and important point that distinguishes the construction grammar tradition from the generative grammar tradition and probably some others as well. The generative grammar tradition does *not* look at your Russian or English examples, say “ah, construction”, and stop there. While the generativist research culture will acknowledge, at least provisionally, that there are collocation-specific rules taking one from “kick the bucket” to the meaning “die” — that gets acknowledged only when we see no path forward to a deeper explanation. Positing a “construction”, in generative linguistics culture, is a defeat, not something to be satisfied with. Defeats happen, but when they do, they are — well — defeats.

 We are not all constructionists.

    Positing a “zero question morpheme”, to focus on your second example, is not “constructional”, not a defeat — because it implicitly hypothesizes important generalizations concerning yes/no questions: that the properties of the proposed null morpheme will fall together with other known properties of complementizers, of heads that attract other heads to them, of null elements in general, of elements with focus-related semantics, and more. And these hypotheses have proved generally proved fruitful in research carried out over the past several decade. Of course, at the end, we might have to concede some defeats, e.g. why should the morpheme be null in English, but not, say, in Japanese? Why do dialects and registers of English differ in the obligatoriness or possibility of subject-auxiliary inversion in matrix and embedded clauses? Maybe these will prove to be unanalyzable difference among languages motivating construction-grammar-ish statements like “this instance of C attracts the finite verb to it because, you know, it just does”. But it is also possible that energetic and inspired research will find hidden correlations that increase the depth of the theory in that domain as well. The work of my colleague Norvin Richards, tracing correlations between movement patterns and the overall prosody of a given language, exemplifies what efforts in this direction look like. My impression is that this difference in willingness to be satisfied with constructional defeat is a significant difference in academic culture. We are not all constructionists.

    

To focus for one second on your first example, here too I think we should not be satisfied with saying that Russian expresses “numerical approximation” by word order and calling it a day. Is that really where we should end the story? This Russian construction (using that term informally) has several weird and interesting properties, which I’m sure you know. For example, the noun may not be modified: “five books” has an approximative variant “books five”, but nothing comparable is possible for “five old books”. Here Loren Billings’ (1995) Princeton dissertation (https://rucore.libraries.rutgers.edu/rutgers-lib/37919/) is important: showing that this “no modifier” property pops up in a whole range of places within Russian (and additional work by Yadroff shows connections to the grammar of classifiers). I don’t know a good general explanation for this cluster of properties — a great topic to explore further (or perhaps it’s been solved already and I just haven’t read the right papers). But I am pretty sure that there is some complex structure to the problem not evident on the surface, and explanations of greater depth than “word order encodes approximation” to develop. I could be wrong, of course, but we won’t know that unless we try, and I think that is really our job as linguists. We are not all constructionists — nor should we be.

    To end this mostly contrarian but optimistic response to your article, I do want to say that I strongly agree with the final paragraph of your blog post. Yes indeed, your question there is a very good one, and central.

    • Many thanks, David! That’s a lot of food for further thought.

      So it seems that you are saying that a constructionist is someone who does not aim higher than treating everything as if it were an idiom – and that we gain from aiming higher: The need to posit an idiom is “defeat”, but each non-constructional analysis is a little victory.

      I certainly agree with that sentiment, but I’m still confused, because different constructionists adopt different rhetorics: While Fillmore et al. (1988) may have emphasized the “idiosyncrasies” rhetoric, Goldberg (1995) stressed the generality rhetoric, and is thus not so different from the usual generative approach: She felt that we “miss generalizations” when we posit many lexical rules. Many different rules would be “defeat” from this perspective.

      So yes: Linguists have different temperaments – some want maximal generality (at the expense of cutting some corners), and others want maximal attention to capricious detail (at the expense of fitting it into a big picture). But this does not need to be a difference that divides us (so that we have to attend different conferences, and publish in different journals). And it surely does not mean that there are different competing theories (“construction theory” vs. “generative theory”, or similar) that might be distinguishable empirically.

      • Well if one linguist’s theory posits that there is a null interrogative complementizer to which a tensed auxiliary moves in certain English questions — and another linguist’s theory denies the existence of that complementizer, the notion of movement, and who knows what else — I call those distinct theories. They will probably distinguishable empirically to the extent that notions like “complementizer” and “movement” (not to mention “construction”) have content that makes predictions about the other phenomena we expect to find and not find in individual languages and across the languages of the world.

        Obviously, we can imagine a better world in which linguists talk to each other about similar sets of phenomena in terms that make it clear where their ideas make distinct predictions — and then put those distinctions to the test. And equally obviously, we do not live in that world, unfortunately. Happy to help work towards it. But neither “attending different conferences and publishing in different journals” nor declaring in advance that differing approaches are probably notational variants strikes me as a productive path forward. Instead, we should figure out which scientific differences are real, and which are less than they seem, and how to actually decide what’s right. Then we can be a science like any other, which would be nice.

        Thanks for the reply.

        • Yes, but in the present context, what matters is not so much how our predictions can be tested, but whether there are larger (“metatheoretical”) incommensurabilities – such that, for example, there is no way to formulate the “movement to COMP” idea (with all its predictions) within a constructional notational framework. That is not clear to me. If the frameworks are not merely (i) notational practices, (ii) communities, or (iii) ideologies, but if they embody claims about Human Language, it would be important to clarify what claims are made. I don’t get the impression that there is any clarity about this.

    • It is indeed clear that there are still questions left over after we identify something as a construction. There must be some reason for why any construction is the way it is and not some other way. But what *kind* of an explanation do we seek?

      Generative linguistics, to me, seems to start from the assumption that explanations preferrably should be synchronic. On the other hand, as a historical linguist it seems clear to me that there in any case also exists a historical explanation (a “construction etymology”). Since you concede that idioms exist, this will be at least sometimes the complete, sufficient answer. So then why not also at some other times? This issue is at its most obvious in rule-based generative phonology, where deep-ranked transformation rules often end up simply duplicating sound changes already proposed in historical phonology. It seems very probable to me that the same is the case also within morphosyntax to at least some extent, even though we currently do not have similarly detailed reconstructions of historical morphosyntax to check with.

      You may wish to compare this position with other sciences. E.g. within biology, nothing is considered to be explained via “synchronic anatomy”. Positing that “the aorta is X wide because there is a rule (somewhere, where, how?) that it needs to carry Y amount of blood” would be teleology, not explanation. Any actual explanation of the aorta’s specific functions and properties will come instead from ontogeny (“these genes trigger the growth of an aorta of a particular shape in the embryo”) and phylogeny (“said genes were derived from these earlier genes”, “at these points in evolutionary history, individual with different aortas were culled by natural selection”).

      • Yes, this is the issue of functional/diachronic explanation, and you are absolutely right – it is often very insightful to look at diachronic sources. But one might also ask what explains the diachronic changes, and the directionality of the explanation is not always fully clear. See the papers by Cristofaro and Haspelmath in this open-access volume: https://langsci-press.org/catalog/book/220

  2. You end this post with an interesting question: “If language systems consist of constructions, not of morphs and simple combination rules, then why do many sentences look like they largely consist of morph combinations?”
    One possibility is that this is a result of the historical processes from which constructions originate. Constructions don’t just appear out of nowhere, with people randomly selecting which construction to use to express a given meaning out of all logically possible constructions; instead, constructions are created through language use. Thus, patterns in the way that language is typically used should be reflected in patterns for what constructions are actually found in the real world. Here are some thoughts regarding why morph-based analysis often works, and also how usage-based approaches can handle some of the other example you mentioned; I hope they will be of interest! (My answer here is heavily based on my understanding of Joan Bybee’s work; I strongly recommend anyone who finds my answer interesting to read her work! Especially her book “Phonology and Language Use”)
    As to why morph-based analysis often works: The basic mechanism of morph combination arises naturally through language use, so it is unsurprising that it is so commonly seen. Anytime some elements are commonly used together, people are likely to start running them together. Phonetically, this results in two previously distinct elements being pronounced as a single unit. In the earlier stages of this process, before there is a chance for them to be run together to the point of being inseparable, linguists can decompose these elements into separate “morphs”. If some elements are used especially often, they may run together to the point that it is difficult to separate the “morphs”; however, since this fusion necessarily requires an intervening period where the various elements aren’t as tightly bound, we would expect separable morphemes to occur more often than “fused” morphemes, as is indeed the case.
    Meaningful inflectional zeros: If there are only a few categories to be picked from, if all categories except for one are clearly marked, then there is no need to explicitly mark the last category. The lack of marking shows that the element in question doesn’t belong to any of the marked categories, so it must belong to the leftover category. This explanation is more problematic if more than one category is unmarked, however.
    cumulative inflections and portmanteau elements:
    At least as far as the French example, (aux ‘to the PL’, instead of à les), aux is just a contraction of à les. If à les occurred together very frequently in older forms of French, it would be very likely that they would fuse into a single word. To test whether this usage-based explanation makes sense, it would be necessary to look at a large corpus of older French documents. If à les occurred together more than other combinations of preposition + article, this would support the usage-based fusion analysis; otherwise, it would be surprising that the other combinations didn’t fuse.
    non-existent bases: These are often the result of borrowing whole words. “gorge-ous” is actually an example; English borrowed this word from French. Since the word as a whole already existed in another language, English didn’t need a pre-existing base like “gorge” to build this word from. (If there are examples which are not the result of borrowing, they might be more insightful).
    directionality mismatches: I’m not entirely sure what you mean here. In your example, you used the words “assassin” and “assassinate”; however, the noun “assassin” is a loanword from French; the word “assassinate” was created by adding the suffix “-ate” to the noun. This suffix is used to create verbs with the meaning “to act in the manner of the base noun”, so this seems to be a pretty straightforward correspondence of sound and meaning… Perhaps this is just a matter of how the words are defined? That is, does “assassinate” mean “do what an assassin does”, or does “assassin” mean “someone who assassinates”? It isn’t clear which word should define the other, which is a problem if “directionality mismatches” is based on which defines which. This also leaves open the question: are there any examples of seemingly morphologically complex words coming into use within a language before the corresponding morphologically simple word, where the seemingly complex word is *not* a loanword?
    sister derivatives: Similarly to the case of “gorgeous”, both “ambition” and “ambitious” are loanwords from French, so English had no need for a pre-existing base “ambit” from which to “build” these words.
    Meaningful word orders: This is trickier. It’s common for people to emphasize words by starting the sentence with the emphasized word. But I’m not sure how the Russian example of (pjat’ knig ‘five books’; knig pjat’ ‘about five books’) could be explained in terms of language use. This makes this example especially interesting!
    transfixing inflection of the Semitic type:
    I’ll start by making a detour to the similar inflection pattern seen in Germanic ablaut, which is better understood. The ablaut in “foot/feet, mouse/mice, tooth/teeth” is known to be a result of sound change. Originally, plural was marked by a suffix -i; the suffix caused an allophonic change of the vowel quality of the root vowel; later, the suffix was lost (a common sound change), leaving the change in vowel quality as the sole marker of plurality. In the case of Semitic transfixes, we don’t have enough data to be certain of when and how the transfix system formed; however, it is at least conceivable that it originated through similar processes as the Germanic ablaut. (Of course, after a construction has entered a language in this way, the speakers of a language might transfer it by analogy to other parts of the language, such as in the humorous “moose/meese”; but the point is that the construction itself entered the language through usage factors, rather than being invented at random.)
    I could go on; in particular, idioms and valency constructions are a core interest of the usage-based approach, but I think I’ve gone on long enough for now. 😊

      • Oh wow, thanks for the invitation!
        I’m just an undergraduate student of linguistics, so I’m not sure if I have all that of my own to contribute – I just enjoy reading the works of actual linguists, like yourself and Bybee, and occasionally have some thoughts about the things I read. One reason that I like this blog so much is that it’s rather informal, so I feel more comfortable sharing some of my thoughts. I’ve never really had a chance to discuss linguistics with anyone besides my linguistics teachers before, and I really enjoy the possibility of doing so here! 🙂

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.