'A remarkable tribute to the inertia of ideas'
In short, for all its familiarity, the structuralist morpheme represents a particular and, in many respects, extreme view of the way that grammatical information is organized and packaged within a language. The morpheme is not something that is ‘in’ a language or which can reasonably be said to have been discovered, or otherwise established, except with reference to methods that themselves stand in need of validation. Unlike words, morphemes do not in general occur in isolation and are rarely demarcated in the speech stream by phonetic cues or other language- internal evidence. Hence, the procedures of analysis developed for identifying morphemes do not refine some ‘pretheoretical notion of morpheme’, for there is no such notion. Instead, like the morphophoneme, the morpheme is a purely theoretical construct within the specific model of distributional analysis developed by the Post-Bloomfieldians.
Given that the careful formulation of principles of morphemic analysis led many of its initial proponents to abandon a morpheme-based model, one might have expected the structuralist morpheme to suffer the same fate as the structuralist phoneme. But even though the notion of the morpheme had been modelled on and supported by the phoneme in Post-Bloomfieldian approaches, their paths diverged with the eclipse of these approaches. Early generative accounts (Halle 1959; Chomsky and Halle 1965) led a determined attack against the phoneme, or at least against a level of phonemic representation. In contrast, the morpheme successfully jumped hosts and enjoyed a new lease of life within the emerging generative paradigm. This was not because any of the outstanding problems with morphemic analysis had been solved in the meantime. The procedures and definitions in Harris (1942), Hockett (1947) and Nida (1948) remained the most explicit statements of morphemic analysis, as indeed they do today. Later versions of the IA model simply took morphemes for granted and focused on technical refinements of morphemic analysis.
Many of the problems and solutions associated with the original IA model persist in its modern variants and derivatives. For example, the analysis of verbal inflection in Halle and Marantz (1993: i26f.) is essentially an amalgam of the analyses of Harris (1942) and Bloch (1947). A strong preterite form such as took is analyzed as consisting of a zero tense marker preceded by a stem in which the effect of a replacive morph is shifted onto an item-specific ‘readjustment rule’. This type of analysis preserves the most problematic features of Post-Bloomfieldian accounts. The tense features of took are ‘marked’ by a covert element, while the vowel change that actually distinguishes the preterite form is treated as a case of ‘stem allomorphy’ that serves no signalling function. The problems posed by zeros are, if anything, even more acute in modern accounts, given the expanded role that zeros play in generative approaches, as Pullum and Zwicky (1992) point out. The morphotactic problems posed by non-segmental alternations are intrinsic to the IA model because they derive from the basic agglutinative bias of this model.
As Chomsky (1965) acknowledges in one of the early generative treatments of inflectional morphology, the problems with morphemic analysis identified in the 1940s survive intact in transformational accounts that attempt to impose an agglutinative IA analysis onto fusional patterns. After enumerating the complications that arise in representing German declensions in terms of discrete case, number and gender morphemes, he concludes:
I know of no compensating advantage for the modern descriptivist reanalysis of traditional paradigmatic formulations in terms of morpheme sequences. This seems, therefore to be an ill-advised theoretical innovation. (Chomsky 1965:174)
As Chomsky clearly appreciated at the time, there was no principled reason why standard transformational models should be morpheme-based rather than word- based. Much the same remains true into the contemporary period, as the modern WP approaches of Aronoff (1994), Beard (1995), Stump (2001) and, especially, Anderson (1992) are all broadly compatible with subsequent transformational models. Nevertheless, in what Matthews (1993:92f.) characterizes as “a remarkable tribute to the inertia of ideas”, the adoption of a morpheme-based model in Chomsky and Halle (1968) set a decisive precedent. The influence exercised by ‘analytical precedents’ and ‘exemplary analyses’ within the transformational paradigm (Blevins 2008b) discouraged the reevaluation of the status of the morpheme, so that morphemic analysis became part of the transformationalist toolkit. The alternative approaches that split off like tributaries from the transformational tradition were largely preoccupied with syntax and simply carried the morpheme- based model with them. There is no obvious internal motivation for morpheme- based analysis within models such as Lexical Functional Grammar (Kaplan and Bresnan 1982; Dalrymple 2001), Generalized Phrase Structure Grammar (Gazdar
et al. 1985) or Head-driven Phrase Structure Grammar (Pollard and Sag 1994). Yet each of these models adopted, at least initially, a mainly morpheme-based model of morphology, in practice if not necessarily in principle. In time, morpheme- based analyses became established in textbook presentations of these models and in general introductions to the study of language and linguistics.
Yet this period of consolidation coincided with a retreat from the substantive claims of morphemic analysis within descendants of morphemic models. Although models such as Distributed] M[orphology] (Halle and Marantz 1993) retain the term ‘morpheme’, the notion in DM has at most a historical connection to the original ‘morpheme-as-sign’ conception. The term ‘morpheme’ is applied to various constructs that mostly lack a specific label in other approaches but are in any case devoid of semiotic content. In initial models of DM, the term ‘morpheme’ is used to refer ambiguously to (i) a grammatical feature bundle, which occurs as the terminal element of a syntactic tree, or (ii) a grammatical feature bundle, together with a phonological feature bundle:
The terminal elements of [syntax] trees consist of complexes of grammatical features. These terminal elements are supplied with phonological features only after Vocabulary insertion at MS [morphological structure]. Although nothing hinges on this terminology in what follows, we have chosen to call the terminal elements “morphemes” both before and after Vocabulary insertion, that is, both before and after they are supplied with phonological features. (Halle and Marantz 1993:114)
On the first of these conceptions, the DM ‘morpheme’ is closest to the Bloomfiel- dian sememe, or the signifie of a Saussurean sign. On the second, it appears to encompass the whole sign. In neither case does it correspond to the morphemes defined by Bloomfield (1933), Harris (1942) or Hockett (1947).
The break from the Bloomfieldian tradition of morphemic analysis is complete when later versions of DM apply ‘morpheme’ to an “abstract unit”:
A morpheme is an abstract syntactic unit that finds an interpretation in form and in meaning... The notion of “abstract” when applied to morphemes characterizes not only the independence of a particular morpheme from a specific realization in form but also its independence from a specific semantic value. For example, the “plural” morpheme in English is the same morpheme, appearing in the same syntactic position with respect to the noun stem, whether it conveys a meaning associated with “more than one” (as in “cats” and “oxen”) or whether it does not (as in “(eye-)glasses” or “pants”). More generally, the features associated with morphemes by linguists, while connected to their syntactic and semantic properties, find their grammatical import (their role in the syntax in particular) independent of their possible semantic (or phonological) interpretations. (Marantz 2013:1f.)
On this view, the “plural” morpheme need not be associated with a constant ‘unit of form’, since it the ‘same’ morpheme in cats and oxen. It also need not be associated with any stable ‘unit of meaning’ since it may or may not convey a notion of plurality. Whatever the function of such an “abstract” unit within the theoretical frame of reference assumed by DM, it preserves no part of the original conception of the morpheme as a Saussurian sign linking minimal ‘units of form’ and ‘units of meaning’. On the contrary, DM adopts an extreme anti-morphemic position, given that it eliminates all of the components of the traditional Bloomfieldian morpheme (meaning, form and linkage) and simply reassigns the term ‘morpheme’ to an abstract ‘unit’ dictated by the syntax.
In sum, to paraphrase Matthews (1993:92), current models of morphemic analysis, at least as represented by DM, provide a remarkable tribute to the inertia of nomenclature, insofar as they preserve the term ‘morpheme’ while abandoning all of the substantive claims originally associated with the notion.
-  In accounts such as Lieber (1992) morphemes are regarded as so firmly established that there isno need for an index entry for ‘morpheme, much less a definition.