Recall that Morris Halle had proposed to treat the phonological form of “abstract morphemes” (those with conditioned allomorphs) as (abstract) “Q.” Q would be replaced in the phonology by the actual phonological realizations of the morpheme, in some context. For DM, Morris and I assumed that, in place of Q, all morphemes had no phonological form at all in the syntax. Nevertheless, the insertion of phonological forms via “Vocabulary Insertion” was seen as a phonological process, subject to whatever principles we (thought we) knew were applicable in the phonology.
We imagined that a language contained a set of Vocabulary Insertion rules (VIn rules), that were rather like phonological rules in the Sound Patterns of English sense. When the grammar was ready to insert a Vocabulary Item (VIt) into a morpheme from the syntax, all the VIn rules would compete for use. The VIn rule specifying the largest subset of the features on the morpheme would win the competition. If two (or more rules) specified the same subset of morphosyntactic features, these rules would compete on the basis of any contextual features they had that restricted the environment of the morpheme into which the VIt could be inserted. For example, in the competition for insertion into a number morpheme with the feature [+plural], the VIn rule [+plural]<<—->> /-z/ would tie with the VIn rule [+plural] <<—->>Ø in terms of the subset of features that the rules spell out. The latter rule would have the context of the list of stems that take the zero plural (deer, fish, etc.) and thus would win the competition with the /-z/ rule when the [+plural] feature was on a number node sister to, say, deer. The /-z/ rule, having no contextual features, would be the default plural rule, used for [+plural] when the stem fell on no lists associated with the context for special VIn plural rules.
The “subset principle” here was supposed to be a version of a Pāninian elsewhere condition of the sort that governed the application of phonological rules. The nesting of features between VIn rules (say a rule spelling out +plural and +masc nesting a rule spelling out just +plural) was parallel to the rule abbreviation conventions of phonology (in the 1960’s at least) and captured the idea that the most specified VIt that fit into a morpheme would be inserted (blocking the insertion of a VIt whose features nested within the more specified VIt’s). Anderson’s principles for complementary distribution of the morphophonological rules realizing features in A-Morphous Morphology included a similar approach, again borrowed from phonology.
The overriding principle governing the complementary distribution of VIts was the one morpheme, one VIt principle. That is, if a morpheme had features A and B, and there were separate VIns spelling out A and B, only one VIt could be inserted. Such a situation, however, would not be covered by the subset principle, since A and B aren’t subsets of each other. To mediate competition in such cases, additional principles were necessary. Halle & Marantz propose stipulated ordering of VIs in such cases (so the one spelling out A might be ordered before the one spelling out B and thus bleed it – if the one spelling out A, however, had a contextual feature not met in a particular word, then B would show up). Other approaches to VI rule ordering were explored, including a universal hierarchy of features (see Noyer (1992), e.g., for early work on this idea).
But if there is just one VIt per morpheme, the “bundling” of features in single morphemes becomes crucial for explaining the distribution of VIs in a word. Halle & Marantz here supposed that “bundling” could be universal, e.g., if say Agreement in natural languages involved Agreement morphemes with a particular set of “phi-features” for person, number, and gender, then person, number and gender would be bundled universally into a single morpheme. Or bundling could be language specific, with each language determining how to package the universally available features into morphemes. A “bundling” parameter, for example, was explored in Pylkkänen (2008)) where languages were claimed to differ depending on whether or not they bundled the features of voice and those of v into a single morpheme, with consequences for both syntax and for morphophonology.
So, DM had bundles of features in terminal nodes of the syntax formed prior to the use of these bundles in the syntax, a hierarchical organization of these terminal nodes (= morphemes) produced by the syntax, and a principle of Vocabulary Insertion via VIn rules that provided, in the morphophonology, a single VIt for each morpheme from the syntax. An additional set of assumptions was necessary to account for word formation – how syntactically distributed morphemes end up in a sequence of phonological words. Marantz (1984) included an elaborate theory of “morphological merger” – the process by which morphemes were put together into words. DM did not adopt the analyses of Marantz (1984), instead buying into what was turning into a consensus account of syntactic word formation: a morpheme that headed a lower phrase raised and adjoined to the head of the phrase that took its maximal projection as its complement (if XP were the complement to YP, X could raise and adjoin to Y). Halle & Marantz more or less presuppose the viability of this analysis, despite the fact that mainstream generative theory at the time, influenced by Chomsky, included a lexicalist assumption that words were the basic units of syntax, constructed in the lexicon (see the final section of Halle & Marantz (1993) for a comparison with Chomsky’s approach). In addition to head movement (and adjunction), Halle & Marantz supposed that some morphemes could be inserted (and adjoined to morphemes already in the syntactic tree) after the syntax proper, e.g., agreement and perhaps case morphemes. And cliticization (of the sort found in English possessive constructions like “the queen of England’s hat”) was assumed to involve yet another post-syntactic process, (morphological) merger under adjacency, adjoining two morphemes that were adjacent in the morphophonology.
Bundling, head-movement, morphological merger, and the one morpheme, one VI principle served as the scaffolding for a piece-based realizational approach to morphology and phonology, contrasting with, e.g., Lieber-style (1992) lexicalist theories with their phonology-laden morphemes and Anderson-style realization theories without Vocabulary Items as pieces. However, empirical issues required additional mechanisms for early DM, one central to the theory and two place-holders for a better theory to come. The subset principle of Vocabulary Insertion along with assumptions about how context resolves ties between VIns that spell out the same features lead to this generalization about contextual allomorphy: the more specific VIts go in the more specified environments, while the more general VIts go in the less specified environments (are relative defaults). However, against this generalization, there seemed to be situations in which a more general VIt is inserted in a more specific environment. Bonet (1991) provided a set of cases of this sort from Romance pronominal clitics and suggested a principle of Impoverishment could explain the facts. For example, the Spanish dative clitic le (for third person masculine nouns) occurs in most environments with other clitics. However, in some dialects, before the third person accusative lo, le surfaces as se, apparently the third person reflexive clitic. Bonet argues that se is actually a default (third person) clitic, and its distribution motivates the deletion of features from the 3rdperson dative clitic in the environment of the 3rd accusative clitic before Vocabulary Insertion. This Impoverishment of features causes the insertion of a more general clitic, se, in a more specified environment (before lo) over le, which is the default third person dative clitic. An understudied claim of Marantz’s (e.g., 2010) is that the locality constraints on the relationship between the trigger of Impoverishment (in our example, the accusative clitic) and the target of Impoverishment (here the dative clitic) is looser than that between the locus of VIn and any environment that might trigger a contextual allomorph to be inserted. In Halle & Marantz, the analysis of Potawatomi involved a longish distance relationship between an Impoverishment trigger and its target, a relationship that was too distant to have triggered contextual allomorphy at the target. If Marantz’s observation is correct, Impoverishment and contextual allomorphy would properly be separated in the theory, as in standard DM.
Given the one morpheme, one VI principle, certain patterns of morpheme distribution become difficult to describe. For example, data motivating portmanteaux morphemes (where a single VIt looks as if it is spelling out two morphemes) and circumfixes (where two VIts look to be spelling out a single morpheme, on opposite sides of a stem) challenge straightforward accounts of Vocabulary Insertion under one morpheme, one VI. Halle & Marantz endorse two brute force operations to provide rather standard accounts of these phenomena. In Fusion, two morphemes join into one before Vocabulary Insertion, allowing the features of both to contribute to the choice of the VIt for this fused morpheme and predicting complementary distribution between a portmanteau VIt and any other VIts that spell out only the features of one or of the other of the pre-fused morphemes. For Georgian, we supposed that the subject and object agreement morphemes fused prior to vocabulary insertion, explaining the fact that only a single prefix reflecting the person and number of the subject or the object occurs with any verb, even though prefixes exist in the language to separately spell out a subject and an object agreement.
In Fission, some of the features of a morpheme are split off from the morpheme into a separate terminal node, allowing for a VI to be inserted into the original morpheme and an additional VI to be inserted into the new terminal node. For Georgian, we accounted for the appearance of number suffixes in the same verbs as agreement prefixes by Fissioning off the number feature when it occurred with a certain set of other features in the Fused subject and object agreement morpheme. (See Blix (to appear) for a recent non-Fusion, non-Fission analysis of the Georgian data within a realizational morphology.)
After the appearance of Halle & Marantz, Jochen Trommer (e.g., 1999) pointed out that if we abandon the one morpheme, one VIt principle, allowing multiple VIn into a single morpheme, we could derive (at least most of) the same forms as the H & M version of DM does without Impoverishment, Fusion, or Fission. Impoverishment would be replaced by “consuming zeros”: phonologically null VIts that eat up features in the environment of other morphemes, effectively Impoverishing them prior to VIn of a phonologically contentful VIt. Fission would simply involve multiple VIn into a single morpheme, and Fusion could be replaced by contextual allomorphy at one of the apparently Fusing morphemes in the context of the other, followed by either non-insertion of a VIt into the other morpheme or insertion of a consuming zero VIt.
The question of which of these mechanisms to retain in the theory is not a simple matter of simplicity or redundancy. Before we abandon, say, Fusion, we need to ask whether or not there is a theory of Fusion that makes different predictions than DM with, say, Trommer’s changes. Of particular interest are locality domains for the operations, specifically the relation between targets and triggers, as well as any interactions among operations that result from their ordering. For example, we have already seen that the environment for the trigger of Impoverishment may be at a longer distance from the target than that between the environment for Vocabulary Insertion and the morpheme at which we’re inserting the VIt. If this is correct, it argues against Trommer’s collapsing of Impoverishment with VIn (of a consuming zero). Similarly, Matthew Hewett has recently argued for the autonomy of Fission as an operation within DM based on the interaction of Fission with other processes (Hewett (2020)).
Blix, H. (to appear). Spans in South Caucasian agreement. NLLT. https://doi.org/10.1007/s11049-020-09475-x
Bonet, E. (1991). Morphology after syntax: Pronominal clitics in Romance. MIT.
Halle, M., & Marantz, A. (1993). Distributed morphology and the pieces of inflection. Hale, K. & SJ Keyser (eds.), The View from Building 20.
Hewett, M. (2020). On the autonomy of Fission: Evidence from discontinuous agreement in Semitic. NYU MorphBeer handout.
Lieber, R. (1992). Deconstructing morphology: Word formation in syntactic theory. University of Chicago Press.
Marantz, A. (1984). On the Nature of Grammatical Relations. Linguistic Inquiry Monographs Cambridge, Mass, (10), 1-339.
Marantz, A. (2010). Locality domains for contextual allosemy in words. Handout of a talk given at the University of California, Santa Cruz, 30.
Noyer, R. R. (1992). Features, positions and affixes in autonomous morphological structure (Doctoral dissertation, Massachusetts Institute of Technology).
Trommer, J. (1999, August). Morphology consuming syntax’resources: Generation and parsing in a minimalist version of Distributed Morphology. In Proceedings of the ESSLI Workshop on Resource Logics and Minimalist Grammars (pp. 469-80).
I have long thought that some morphological operations ought to come for free in a model of the grammar where morphology and phonology are (a) on the same “branch”, e.g., both part of PF realisation, (b) where both trade in the same currency, viz., features, and (c) where phonology itself has operations. In such a set up, phonological operations are just PF-branch operations on features. Consequently, they should apply just as readily to morphosyntactic features as they do to phonological ones (unless, obviously, an operation depends on the kind of structure that features occupy and structures are not equally distributed across morphology and phonology). (I started writing something about this once but got distracted by other stuff. The nearest thing I’m aware of, though it pushes a slightly different angle, is Nevins’ 2008 chapter on “phon–phi” parallelisms.)
With regard to Trommer’s reduction of operations to vocabulary insertion: another argument against this comes from feature insertion, which Noyer and I have thought necessary for Nimboran and Kiowa (though Jane Middleton is busy reanalysing the facts; I’m less compelled by Hewett’s case but no point commenting until it is published.)