Desktop version

Home arrow Language & Literature arrow Word and paradigm morphology

No segmentation without representation?

To make this issue more concrete, let us return to the proportion discussed by Matthews (1991:192f), repeated in (5.2), with the deduced form in italics.

(5.2) Analogical deduction of Latin case forms dominus : domini = servus : servi

It is easy to see why the pattern matching and pattern extension involved in this deduction can be interpreted as tacitly introducing the same elements that would be combined in a rule-based analysis. The analogical principles that yield the solution servi implicitly recognize the noun stems domin and serv, along with the case inflections -us and -i. Analogical extensions of the Russian noun forms in Table 4.1 likewise operate over the stems and endings isolated and cross-indexed in Table 4.2. Hence in these cases, analogical deductions appear to achieve much the same result that would be obtained if word forms were built by rules that combined lexical stems and inflectional exponents.

Yet there is an essential difference between the status of sub-word forms in analogies and in analyses that build words from smaller parts. A pair of exemplary cells a and b may consist of a common stem and distinct inflections, as in (5.2). But a and b may exhibit any pattern that can be extended to another pair of forms. Proportional analogies thus apply to cases in which a principled stem-exponent segmentation is difficult to motivate, to cases in which segments do not correspond to general units in a language or in which segments cannot be assigned a consistent morphosyntactic analysis.

This flexibility derives from the morphotactic agnosticism of a classical WP implicational model, which in turn reflects an emphasis on relations rather than on units of analysis. As Morpurgo Davies (1998: 258b) puts it, proportional analogies “offered an algorithm for a structurally based form of morphological segmentation, without making any claims about the segments in question”. Because analogical principles deduce rather than build novel forms, they can exploit any predictive patterns, without attaching grammatical significance to the segmentations that are of predictive value. Distinct proportions that extend different patterns need not even impose consistent segmentations. The structural agnosticism of proportional analogies addresses the concerns expressed by Lounsbury (1953) on p. 4 in cases where “comparison of forms suggests one placement, while another comparison suggests another”.

The inflectional systems in preceding sections exhibit a variety of patterns that bring out the differences between deductive analogy and constructive rule application. One correspondence involving Russian participles is stated below:

A convenient way of forming this [present active JPB] participle is by replacing the final [-t] of the 3rd pers. pl. by the ending [-scij] (Unbegaun 1957:169)

The present active participle is formed by replacing the final [-f] of the third-person plural of the present tense by the endings [-scij] (Wade 1992:361)

Another frequently-noted correspondence between present and participial forms holds between the 1pl present and the imperfect passive participle:

A convenient way of forming this [imperfective passive] participle is by adding the adjectival ending [-yj] to the 1st pers. pl... (Unbegaun 1957:170)

The imperfective passive participle is formed by adding adjectival endings to the first-person plural of an imperfective transitive verb ... (Wade 1992:364)

Like many of the principles stated in descriptive grammars, these correspondences are explicitly‘Priscianic’ (Matthews 1972), directly relating two surface word forms. Since the patterns apply across conjugation classes, nearly any verb can define the antecedents for a valid analogy. The proportions in (5.3) use the relation between forms of the exemplary verb govorit’ ‘speak’ as a model for the deduction of the italicized forms of videt’ ‘see’. The correspondence between 3pl forms and present active participles is expressed in (5.3a); the relation between 1pl forms and imperfective passive participles in (5.3b).

(5.3) Analogical deduction of Russian participles

a. govorjat: govorjascij = vidjat: vidjascij

b. govorim : govorimyj = vidim : vidimyj

As with Matthews’ original example, these proportions adopt a convenient shorthand that suppresses the features associated with terms. Each term in a proportion is a ‘word’ in the sense of “a form which may be uttered alone (with meaning)” (Bloomfield 1926:156). Hence govorjat is not a simple form but is implicitly associated with the 3pl cell of the present indicative paradigm of the verb govorit’. The other terms have similar grammatical meanings.

The simple patterns described by the deductions in (5.3) can of course be expressed by word-building rules that combine stems and exponents. However, when we examine the rules we find that they largely mimic analogical deductions in ways that are alien to the normal operation of rule systems.

It might seem that the present active participle forms in (5.3a) could be built by adding a participial ending to the consonant-final “present-future stem” (Wade 1992:228) that underlies the present conjugational series. On this analysis, the ending-ascij would be added to the stems govor, vid and smotr in Table 5.1. But this analysis would fail to apply correctly to the forms delajuscij, iscuscij and moguscij, which contain the vowel u (instead of a). A general rule-based analysis must then add the consonant-initial ending -scij to vocalic stems which preserve the vowel of the 3PL present form. Hence, extending a rule-based description to the patterns in (5.3a) is possible, but only at the cost of recognizing an extra vocalic variant of the “present-future” stem.

Table 5.1 Present active participle formation in Russian (Unbegaun 1957:169b)














Table 5.2 Imperfect passive participle formation in Russian (Unbegaun 1957:170)














A rule-based strategy can, again in principle, be applied to the pattern in Table 5.2 but here the choice of ‘stem’ is even more problematic. The base for the imperfect passive participle differs from the present-future stem in forms like mozemyj, which exhibit consonant mutations that distinguish them from the basic stem mog. The base also differs from the “infinitive-preterite” stem (Wade 1992:228) in forms like iscem, which contrast with past-infinitive forms based on the unmutated isk(a). The final vowel also varies across verbs, and contrasts with the vowel in the present active participles. In short, there is no form that provides a better base for the imperfect passive participle than “the first-person plural of an imperfective transitive verb” (Wade 1992:364).

A rule-based analysis can avoid a direct relation between the present forms and participles in Tables 5.1 and 5.2 by positing abstract stems that underlie both pairs of forms. But it is unclear that these stems serve any real purpose other than avoiding a direct relation between word forms. Within a model of morphological analysis that is committed to constructing words from smaller units, the avoidance of word-level correspondences may be regarded as a goal worth pursuing for its own sake. However, this strategy has a range of consequences for a rule-based account. As discussed above, one immediate effect is the expansion of the stem inventory and/or an increase in affixal ambiguity.

More subversive is the effect on the interpretation of these components. A rule-based analysis of delaemyj must either treat delaem as realizing 1pl in some contexts and as an uninflected stem in others, or it must assign this context- dependent interpretation to a subsequence that includes at least -em. Parallel remarks apply to forms like delajuscij. Each of the choices available to a rule- based analysis represents a retreat from a model in which forms are assigned stable segmentations into recurrent units with fixed grammatical interpretations. Instead, the use of stem allomorphy and affixal ambiguity in rule-based account introduces, to paraphrase Morpurgo Davies (1998:258E), ‘a structurally based form of morphological segmentation, without making any context-independent claims about the segments in question’.

In sum, to the extent that “analogies... are indistinguishable from rules” (Kiparsky 1975:75), this merely reflects the fact that generative rule systems are powerful enough to mimic the effects of proportional analogies. Reencoding analogies by rules that invoke stem allomorphy and affixal ambiguity does not of course show that the original analogies tacitly incorporate “the rules and principles of sound structures” (Chomsky 1988: 26). Instead, this analysis illustrates how morphotactically agnostic proportions can be recast in terms of rules if there are no constraints on the abstractness or ambiguity of units.

< Prev   CONTENTS   Source   Next >

Related topics