[This post is a draft of a section of a chapter that I’m supposed to be writing with Neil Myler on contextual allosemy in Distributed Morphology for the forthcoming Handbook of Distributed Morphology. Comments and corrections solicited!]
In Distributed Morphology, idioms have been characterized as contextual allosemy. “Allosemy” is the property of having multiple meanings, either related or unrelated; “contextual allosemy” involves the choice of one of these meanings in a particular context or environment. The discussion of idioms as contextual allosemy starts with observations from the literature about locality restrictions on idioms. Notably, “subject idioms” with an idiomatic subject and verb, but an open object position, are observed to be rare or nonexistent, as are passive idioms that are not idiomatic in the active. These observations suggest a VP-sized locality domain for idioms that excludes transitive subjects. As part of the general project to dissolve the word/phrase distinction, Marantz (1997) relates these observations to Miyagawa’s (1980, 1984) claim that lexical causatives can be idiomatic, while syntactic causatives in Japanese are never idiomatic. The leading idea here is that the locality domain for idiom formation might fall inside of a phonological word. The Japanese syntactic causative, although a word-sized unit, nevertheless cannot take on an idiomatic meaning as a whole because the relationship between the causative affix and the verb stem would cross a locality barrier, much like the way that the relationship between a transitive subject and the verb does.
(25) VP idioms with V-(s)ase (Miyagawa (1984: 190)
a. hana o sak-ase
flower ACC bloom-CAUS
‘to succeed’b. hara o her-ase
stomach ACC decrease-CAUS
‘to be hungry’
Marantz (1997) links these observations to the active voice node, which is a putative phase head in Minimalist Grammar. Such a node would separate the subject and a transitive verb, as well as the syntactic causative affix and the verb stem, which arguably contains active voice as well. The idea is that a phase boundary may occur within a word, as well as in phrases, and that phases mark the boundaries of lexical influence, such that idioms must be fully contained within a phase.
From the origins of this proposal within Distributed Morphology, any possible contrast between idioms and polysemy was blurred – “polysemy” referring to the property of having multiple related meanings, such as book the physical object and the intellectual property. Consider the title of a widely distributed paper addressing issues of stem suppletion, “‘Cat’ as a phrasal idiom” (Marantz 1995). The leading idea of the “Cat” paper was that all words, even apparently morphologically simply words like cat, decompose into at least a root and a category determining affix. The meaning of roots would always be determined contextually, within the domain of the first phase head up from the root. Calling cat a phrasal idiom, then, emphasizes both the dissolving of the word/phrase distinction and the assumption that idioms involve the same contextual calculation of root meaning as is involved in polysemy. That is, the connection between (fill the) bucket and (kick the) bucket might be parallel to that between (physical) book and (intellectual property) book.
What was missing from this early work, then, was any clear delineation among at least three possible relations between different meanings of the same phonological form: (accidental) homophonic, allosemic, and idiomatic. The two meanings of bat would, for the synchronic grammar at any rate, involve accidental homophony, the two meanings of book would involve polysemy and the two meanings of bucket an idiomatic connection. If these distinctions are real, linguistic theory should provide means beyond meaning intuitions to classify cases as one or the other. For the distinction between accidental homophony and polysemy, what’s needed is a theory of polysemy – what kinds of meaning relations are made available by grammars that might connect related meanings of the same root. Two apparent meanings of the same phonological form would involve polysemy to the extent that the relation between the meanings is analyzable within the theory of polysemy. What about the distinction between polysemy and idioms? Here, two possible generalizations emerge from the literature. First, idioms always involve a reading in addition to a possible literal reading, whereas in contextual allosemy, one reading may be forced. If there’s a bucket on the ground, you can always kick the bucket and spill its contents. However, although globe has related meanings of a sphere and the earth,global has only the earth reading. Second, when we’re considering the locality domain for interpretation, idioms seem to involve the relation between (at least) two roots, while allosemy may involve a root and a functional morpheme. Even in the case of the Japanese syntactic causatives, recent work by Yining Nie (2020) suggests that a root is involved for the causative affix in such constructions.
In an important and illuminating paper, Anagnostopoulou and Simiati (2013) present evidence from Greek adjectival participles to argue that the locality domains for idiom formation and contextual allosemy are indeed distinct, and that the domains correspond to the conflicting analyses of “special meanings” in Marantz (2001) and Marantz (2013). While active voice defines the domain for idioms, semantically relevant category determining nodes (n, v, a) create barriers for contextual allosemy. Anagnostopoulou and Samioti consider different types of deverbal adjectives formed with the suffixes –t(os) and –men(os). They argue that many –tos formations involve an adjective formed directly from a (semantic) root. With no barrier between the –tos affix and the root, these words can involve contextual allosemy on the verbal stem, with a special meaning triggered by the –tos not available for the verbal root as a verb.
(39) Stative –tos participles showing direct attachment of –tos to Rootevent
a. Verb sfing-o ‘tighten’
Participle sfix-tos ‘tight, careful with money’b. Verb ftin-o ‘spit’
Participle ftis-tos ‘spitted, spitting image’
Meanwhile, canonical –men(os) adjectives are built on semantically eventive v stems. Since the v intervenes between –menos and the root, –menos may not trigger special meanings not available for the use of the root as a verb. On the other hand, –menos adjectives may have additional, idiomatic readings not present for the verbal stem. For example, in (41), the –menos participle has either the literal meaning or the idiomatic reading, while the idiomatic reading is not present for the verb in its other forms.
(41) Eventive –menos participles
Verb trav-a-o ‘pull’
Participle trav-is-menos ‘pulled, far-fetched’
Finally, there are –t(os) adjectives with “ability/possibility” readings parallel to –able adjectives in English. The semantics of ability adjectives implicate active voice, and Anagnostopoulou and Samioti (2013) argue that there is additional morphological evidence indicating that the ability –tos adjectives are formed from a larger stem than the perfective –tos adjectives and –men(os) adjectives. For the ability –tos adjectives, no idiomatic reading is possible; any ability reading of the adjective is paralleled by a reading of the (transitive) verb.
(51) Ability –tos adjectives have no idiomatic readings
a. –menos trav-ig-menos ‘pulled, far-fetched’
–tos aksi-o-travix-tos ‘worth pulling’b. –menos stri-menos ‘twisted, crotchety’
–tos aksi-o-strif-tos ‘worth twisting’
The Distributed Morphology literature has presented evidence that idiom formation respects boundaries that may appear inside or outside of phonological words. This evidence supports the general thesis of DM that the phonological word is not itself a relevant constituent for syntactic and semantic principles that govern syntactic structure and meaning. However, the literature does not yet provide a comprehensive analysis of idioms or of polysemy, relying on generalizations in these areas that stand outside a general theory.
References
Anagnostopoulou, E., & Samioti, Y. (2013). Allosemy, idioms, and their domains: Evidence from adjectival participles. In Folli, R., Sevdali, C., & Truswell, R. (eds.), Syntax and its Limits, 218-250. Oxford: OUP.
Marantz, A. (1995). “Cat” as a phrasal idiom: Consequences of late insertion in Distributed Morphology. MIT: Ms.
Marantz, A. (1997). No escape from syntax: Don’t try morphological analysis in the privacy of your own lexicon. University of Pennsylvania Working Papers in Linguistics, 4(2), 14.
Marantz, A. (2001). Words. WCCFL XX handout, University of Southern California.
Marantz, A. (2013). Locality domains for contextual allomorphy across the interfaces. In Matushansky, O., & Marantz, A. (eds.), Distributed Morphology today: Morphemes for Morris Halle, 95-115. Cambridge, MA: MIT Press.
Miyagawa, S. (1984). Blocking and Japanese causatives. Lingua, 64(2-3): 177-207.
Miyagawa, S. (1980). Complex Verbs and the Lexicon. University of Arizona: PhD dissertation.
Nie, Y. (2020). Licensing arguments. New York University: PhD dissertation.
Two questions (this is becoming a habit…):
1. Are we supposed to intuit a difference between (39a) on the one hand and (41) on the other, such that the former is “polysemy”/”contextual allosemy”/”special meaning” while the latter is “idiomaticity”? I ask because, to my untrained eye, the relation between “pull” and “far-fetched” is not qualitatively different from the relation between “tight” ad “careful with money”. And if there is no semantically independent criterion for “polysemy”/”contextual allosemy”/”special meaning” vs. “idiomaticity”, isn’t the claim that the two respect different locality boundaries vacuous…? To put this all another way: in the paragraph right above (41) you say, “‑menos may not trigger special meanings not available for the use of the root as a verb”; but you then proceed to show trav-…‑menos, that has a meaning not available for the verb formed of trav‑. That seems contradictory, so I’m probably missing something; what is it?
2. Is it true that the existence of a compositional meaning, which is accessible alongside the non-compositional one, is going to slice the pie the right way? Alongside its non-compositional meaning, “kick the bucket” allows a compositional reading, relating to making contact between one’s foot and the bucket. Similarly, “transmission” has a non-compositional reading (relating to cars, and crucially, extant even for those speakers who have no idea how cars work), but it co-exists with a compositional reading (the transparent meaning(s) one would expect the nominalization of “transmit” to have; e.g. denoting the act of transmitting). “Terrific”, on the other hand, does not seem to allow access to its compositional meaning (which would roughly mirror that of “horrific”). What is the relevant structural property that groups “kick the bucket” and “transmission” together to the exclusion of “terrific”?
Omer as usual asks the right questions and notices slipshod presentation. In any case, he comes to the same conclusion I thought I was pushing in the post — that a convincing division between idioms and polysemy awaits a constrained theory of polysemy.
Note that, within a phase, both polysemy and idioms may be possible and, moreover, contextual allosemy that provides an alloseme consistent with a theory of polysemy doesn’t have to rule out another alloseme (the forms can be ambiguous). The upshot, as possibly for cases like (39a), is that the theory might allow either a polysemic or idiomatic analysis of the same examples.
Crucially, but very hard to show, is that heads falling outside a phase containing a root can’t condition contextual allosemy but CAN be involved in idiom formation. So Omer’s quote about -menos, probably badly written on my part, is meant to claim that -menos can only be involved in idiom formation with the root, not contextual allosemy.
Omer’s 2. brings out the need to examine some assumptions hidden in the background. I personally have assumed that “contextual allosemy” could be behind the fixing of a root meaning in context even when no polysemy is involved. So, for example, if the root “ami-” occurs only in the word, “amiable,” the suffix -able could be seen as triggering the meaning of “ami-” as its sister, even though it’s not contrasting the meaning of “ami-” in this environment with its meaning in another environment (by hypothesis, there is no other environment to consider). I would treat “terrific” the same way, claiming that the connection between the root in “terrific” and that in “terrify” is not one of polysemy (more akin to accidental homophony). As for “transmission,” I think this example has been under analyzed in the literature. The “transmission” in a car is not a part, like an axle — it’s an assembly (a system). My strong intuition is that it couldn’t be a part — transmission is the modifier of a null head (transmission system). As such, the word itself is transparent in meaning.
I left out examples from Anagnostopoulou and Samioti that for them, and for me, were more convincing as to the claim that contextual allosemy rather than idioms were involved. These were like “amiable,” where the stem embedded under -tos has no meaning/use without the suffix. For Omer, however, words that only occur in idioms are canonical in idioms, so these examples would not convince him without more argument that a tighter definition of “idiom” would rule these out of the class.
I learned about the generalization that idioms do not have an open object position in my undergrad syntax class, and I’ve been noting counterexamples to it ever since. Russian (of course) has quite a few, for example “меня жаба душит” [menʲa ʐaba duʂɨt] 1p.sg.acc toad-nom.sg chokes-3p.sg,, which means “I am feeling very stingy”. Here the open position is the object of “choke”. There are more examples that have the open position that’s an adjunct, like у него душа ушла в пятки” [u menʲa duʂa uʂla v pʲatki] by 1p.sg.acc soul-nom.sg went-3p.sg into heels, which means “I am frightened”. These are fairly numerous, although, in line with the generalization, there are many more examples of standard idioms where the nominative subject is the open position.
I am sure other counterexamples have been noted to this generalization, but I’ve sometimes wondered whether it has to do with Russian’s richer case system, and in particular the fact that subjects in Russian do not have to be nominative. So I wonder whether there are similar counterexamples from languages like Icelandic or Tagalog. And if there are, why would this be?
Oops, my Cyrillic does not match my IPA in the second example–it should be у меня душа ушла в пятки.
In my post, I did not do justice to the long literature on these issues about idioms, including literature explicitly responding to my own claims, e.g., in my 1984 book. The upshot of this literature is that any generalization about the lack of “subject idioms” has to be narrowed to subjects of eventive, not stative, transitive constructions. Maria’s examples are stative in meaning (I am feeling very stingy, I am frightened), and so fall outside the generalization. In addition, there are idioms with idiomatic subjects and free possessive slots, like “the cat got X’s tongue,” that also fall outside the generalization and seek explanation within any constrained theory of idioms.
Introductory courses often oversimplify issues. In this case, of course, one can ask whether the numerous qualifications necessary for the generalization about subject idioms undermine the conclusion that there’s any there there.