From a Post-Bloomfieldian perspective, the challenges raised by the participial patterns in Russian, Latin and Sanskrit might still be regarded as consistent with the view that analogy is relevant to marginal constructions occupying the ‘periphery’ of a language but that the ‘core’ remains governed by rules of a more conventional sort. It is also possible to deny that morphomic patterns have any real status in the grammar and reanalyze morphomic patterns as instances of simple homophony. In the case of the Latin third stem pattern, the present stem could be treated as a common base for past passive participles in -t and future active participles in -tur. On this analysis, there would be no direct relation involving these participles (or the supine).7 The correspondence between the present infinitive and the imperfect subjunctive could also be regarded as fortuitous, on the grounds that it does not characterize the (regular) alternation between the short -re in the infinitive and the long -re in ipl imperfect subjective floreremus and the corresponding 2pl form floreretis.8
Yet the treatment of putatively morphomic patterns as homophony raises a number of general issues. As discussed in Chapter 8.3, genuine homophony appears to be much less common than previously assumed, and is often an artifact of imprecise descriptions. Although it is not possible to investigate sub-phonemic variation in Latin conjugations, it is at least clear that homophony enjoys no distinguished status as a null hypothesis. Moreover, as Maiden (2005:169) notes, morphomic conjugational patterns in the modern Romance languages appear to be highly resilient, to the point that “speakers can actually pass up golden opportunities to align allomorphs with morphosyntactic properties... in favour of the ‘morphomic’ distribution”. More significantly, the question of whether the -t in amatus is the ‘same element’ as the -t in amatUrus or whether the -re in florere is the ‘same’ as the -re in florerem betrays a residually derivational perspective. For a morphomic analysis, it is the predictability of form variation that is of central importance; whether the variants can be derived from a common historical source is largely immaterial.
Thus these alternatives highlight a latent ambiguity in the description of morphomic ‘units’. The reuse of familiar terms like ‘stems’ may imply a notion akin to a ‘shared derivational base’. However, if morphomes are, as suggested in Section 5.2.1, interpreted as units of predictive value, then stems will merely be interpredictable sequences larger than individual formatives. Derivational relations between stems will play no role in a synchronic description but will instead characterize shared historical origins. This ‘abstractive’ conception of stems is expressed by the definition offered by Robins (1989).
The term stem is often specifically to refer to that part of an inflected word less its inflections; stems may therefore be the same in form as roots, or they may consist of root morphemes together with one or more derivational affixes. (Robins 1989:244)  
The availability of homophony-based alternatives also reflects the simplicity of stem syncretisms in Latin. Although the Latin third stem has become a standard example of a morphomic correspondence, more intricate patterns of this type may play a fundamental role within an inflectional system. As recognized in everything from elementary school textbooks to philological studies, the patterns of basic noun and verb inflection in Estonian are irreducibly analogical. Consider the patterns within grammatical case forms, described first in Chapter 4.2. The proportions in (5.5) use the exemplary noun pukk ‘trestle’ and principal parts of lukk ‘lock’ to exhibit two class-specific patterns.
(5.5) Analogical deduction of Estonian case forms
a. pukki: pukkide = lukku : lukkude
b. pukkide : pukkides = lukkude : lukkudes
The proportion in (5.5a) expresses a general correspondence between the strong vowel-final partitive singulars pukki and lukku and the corresponding genitive plurals pukkide and lukkude. The proportion in (5.5b) in turn expresses an exceptionless correspondence between the genitive plural and the inessive plural, here pukkides and lukkudes. These patterns raise precisely the same issues as the relation between ipl forms and imperfect passive participles in Russian. The form pukki functions as the partitive singular in isolation and as a case- and number-neutral base in the genitive plural pukkide. The form pukkide functions likewise as the genitive plural in isolation and as the case-neutral base in inessive pukides. Similar analogical patterns apply to other case forms, determining the tight implicational structure of Estonian declensions.
In the Estonian examples, the segmentations are stable, but the interpretation of segments is context-dependent. There is transparent historical motivation for these patterns if, as generally assumed, the semantic case suffixes derive from postpositions that governed the genitive (Grunthal 2003). Once established, the analogical correspondences remain transparent. But a cascade of analogical correspondences in which the inessive plural contains the form of the genitive plural, which in turn contains the form of the partitive singular, is not conducive to analysis in a rule system that builds complex forms out of individually meaningful parts. The letter of a rule-based system can be preserved by treating forms like pukki and pukkide as abstract stems that take zero partitive singular and genitive plural markers, respectively. But, like the infamous analysis of English verb morphology in Bloch (1947) this analysis ignores the salient morphological variation that identifies forms, and invests all grammatical meaning in nondiscriminative zeros.
The analogies discussed above are typical in deducing novel words, even though the proportional format applies to units of any size. The privileged role of words again reflects the fact that, at least in many languages,“[t]he word is a more stable and solid focus of grammatical relations than the component morpheme by itself” (Robins 1959:128). Proportional analogies exploit two dimensions of this stability. First, words can be assigned a determinate grammatical interpretation that is unaffected by uncertainty about the interpretation of their parts. Second, the word is usually of greater deductive value than any sub-word unit by itself. The reason for this is that the tightest implicational relations hold between paradigm cells, or betweencells realized by formsofaparticular shape, anditiswords (orlargerunits) that realize cells.
This point can be illustrated by the deduction of infinitive forms in Estonian. As summarized in Table 4.18, Estonian conjugations can be organized into three basic classes, which are identified by a number of minimally contrastive forms. One diagnostic pair is the 2sg familiar and 2pl formal imperative forms. In the first conjugation, the 2sg is weak and the 2pl is strong. The second conjugation inverts this pattern. In the third conjugation there is no alternation. Since a contrast is required to identify class, the deduction of conjugational forms takes the form of the six-part proportions in (5.6). In each triple, the first form is the 2sg imperative, the second is the 2pl imperative and the third is the infinitive. The proportion in (5.6a) uses exemplary forms of first conjugation oppima ‘to study’ to deduce the infinitive of leppima ‘to accept’. The proportion in (5.6b) uses second conjugation huppama ‘to jump’ to deduce the infinitive of lippama ‘to scamper’. The proportion in (5.6c) uses prosodically light third conjugation elama ‘to live’ to deduce the infinitive of pilama ‘to laugh at’ and (5.6d) uses prosodically heavy third conjugation tarbima ‘to consume’ to deduce the infinitive of muutuma ‘to change’.
(5.6) Deduction of da-infinitive forms in Estonian
a. opi: oppige : oppida = lepi : leppige : leppida
b. huppa : hupake : hupata = lippa : lipake : lipata
c. ela : elage : elada = pila : pilage : pilada
d. ‘tarbi: ‘tarbige : ‘tarbida = ‘muutu : ‘muutuge : muutuda
As in the Estonian declensions in Tables 4.3 and 4.5, the forms in (5.6a) and (5.6b) exhibit quantitative grade alternations (which are marked by the contrast between single and double consonants). Also as in the declensions, first and second conjugation verbs have both strong and weak stems: opi^oppi in first declension oppima and htipa^htippa for second conjugation huppa. Given this parallelism, no grammatical meaning can be assigned to strong and weak stems in isolation. Moreover, no deductions can be drawn from these stems without knowing their distribution in the paradigm.
As this example shows, the role of words in proportional analogies derives in part from the fact that contrasting pairs of words (strong and weak in this case) may sanction reliable deductions about other forms of an item that cannot be deduced from the parts of those words that exhibit the alternation. The same logic underlies the applicability of proportional analogies to other types of word- based alternations in which the alternating parts cannot be classified in isolation. Exchange patterns involve oppositions between segments or suprasegmental properties in which the contrasting elements may express either value of the contrast. As Matthews (1991:199) notes, the contrast between 3sg indicative and subjunctive verbs in Spanish is marked by a process of “vowel reversal”. A first conjugation verb such as comprar ‘to buy’ has the 3sg indicative form compra and the subjunctive compre. Second conjugation verbs like comer ‘to come’ and third conjugation verbs like vivir ‘to live’ invert this pattern, combining a 3sg indicative come or vive with a subjunctive coma or viva. Both patterns provide a basis for analogical extension, as the deduction of the 3sg subjunctive forms hable ‘speak’ and meta ‘put’ in (5.7) show.
(5.7) Exchange patterns in Spanish (Matthews 1991:199)
a. compra : compre = habla : hable
b. come : coma = mete : meta
It is possible to divide the forms in (5.7) into stems and theme vowels, but not possible to assign a fixed mood value to either part in isolation. Instead, the opposition between forms in -a and -e is intrinsically relational, mediated by their place in a larger set of forms. As Matthews (1991:199) concludes:
In reality, it is not the vowels as such that are important. A form in e is Subjunctive only if it belongs, as a whole, to the paradigm of a Verb like comprar. A form in a is Subjunctive only if it belongs as a whole to the paradigm of a Verb like comer or vivir. (Matthews 1991:199)
Further examples of exchange patterns have already been encountered in the paradigms described in previous chapters. The alternation between the partitive singular pukki and stem partitive plural pukke in Table 4.7 illustrates the most regular exchange pattern in Estonian. The proportion in (5.8a) uses exemplary forms of pukk ‘trestle’ to deduce the partitive plural of hekk ‘hedge’. Proportion (5.8b) exhibits the opposite deduction, in which forms of exemplary kukk ‘rooster’ identify the partitive plural of lill ‘flower. As in Spanish conjugations, neither -e nor -i have number features in isolation. One is plural in paradigms where the other is the theme vowel, and vice versa.
(5.8) Partititive patterns in Estonian (Erelt et al. 1995)
a. pukki : pukke = hekki : hekke
b. kukke : kukki = lille : lilli
-  See Blevins (2003) for discussion of a similar choice in West Germanic.
-  I am grateful to Hans-Heinrich Lieb for pointing out these alternatives.
-  One reflex of this development is that adjectival modifiers of the most recently grammaticalizedcases—the last four in in Table 4.8—still occur in the genitive. Hence in vanas raamatus ‘in the/an oldbook’ both the noun raamatus and the adjective vanas occur in the inessive, whereas in vana raamatuga‘without the/an old book’ the noun raamatuga occurs in the comitative but the adjective vana retainsthe historical genitive.
-  Recall from Chapter 4.3.3 that Estonian contains two infinitival forms: the supine (or ma-infinitive) and the infinitive (or da-infinitive). The supine functions as the conventional citation formand is strong in both the first and second conjugations.