Problems of segmentation

By associating meaning with formatives, the IA model enhances the significance of word-internal structure. The issues that arise in segmenting utterances into words had been instrumental in Bloomfield’s (1914) arguments against classical word-based conceptions. Yet segmenting utterances into morphs did not solve any outstanding problems that arose in identifying word-sized units. On the contrary, segmenting utterances into formatives created a range of new issues, as the analyst was now faced with the challenge of arriving at a principled basis for factoring complex forms into stems and inflections, and isolating the root from derivational exponents and stem formatives. The treatment of ‘thematic vowels’ illustrated the types of choices: should these vowels be treated as part of a stem, as part of an ending, or as separate from both? And what segmentation should be assigned in cases where a thematic vowel is fused or absent in specific forms or, as in Spanish preterites described by Green (1997) below, where segmentation leaves a ‘residue’?

The order of morphemes is fixed: (derivational prefix(es)) + lexical stem + theme vowel + tense marker (sometimes including an empty morph) + person marker. Some forms, however, have fused in the course of history and a neat segmentation is not always possible. The preterit is the most difficult paradigm to analyse, since the theme vowel is sometimes indistinguishable, and segmentingthe second and third person plural markers in the regular way, /-is, -n/, leaves an awkward residue that occurs nowhere else in the system. (Green 1997:99)

Problems of this sort arose even in very simple systems. For example, neither Harris (1942) nor Hockett (1947) was able to decide on the segmentation of English children. Harris (1942:113) contrasts the alternatives child + ren and childr + en and concludes only that “each of the points of division has advantages and disadvantages”. Returning to this “recalcitrant” problem, Hockett (1947:240) considers three further analyses: child + r + en, child with vowel change + en, and no division, and again reaches no firm conclusions, suggesting that “this is one of the cases in which all of our preferential criteria... fail and nothing remains but a resort to convenience”. As these examples illustrate, the challenge that faced the IA model model was not one of merely deciding on a consistent segmentation but of arriving at a principled segmentation.

Non-contiguous arrangements of morphs presented a separate problem, as they required a departure from the simplest procedures of segmentation. The ‘long components’ introduced in Harris (1941, 1951) provided the most general description of infixed and circumfixed morphs. Infixation is illustrated by the Tagalog ‘actor focus’ marker um in (2.1a) and by the Lezgian repetitive marker x in (2.1b). Circumfixation is illustrated by the Georgian superlative marker u. ..esi in (2.2a) and the Chuckchee negative marker in (2.2b).

(2.1) Infixation in Tagalog and Lezgian

a. basa ~ bumasa ‘to read’ (Aspillera 1981:46)

b. akun ‘see’ ~ axkun ‘see again’ (Haspelmath 1993:175)

(2.2) Circumfixation in Georgian and Chuckchee

a. qru ‘deaf’ ~ u-qru-esi ‘the most deaf’ (Tschenkeli 1958: 225)

b. tejkev-эк ‘fight’ ~ e-tejkev-ke-it-эк ‘not to fight’ (Comrie 1981:247)

The principal difficulty presented by thematic vowels and discontinuous arrangements was, in a sense, methodological. In the case of theme vowels, it was possible to cut forms into contiguous parts and even to impose a consistent segmentation. The challenge arose in justifying one split over another. Infixes and circumfixes could also be isolated, though at the cost of departing from procedures that divided utterances into successively smaller sequences.

Yet as acknowledged in the quotation from Lounsbury (1953) on p. 4 above, the search for a principled basis for assigning (or evaluating) segmentations may turn

Table 2.1 Competing motivation for genitive plural forms in -te































out to be misconceived in cases where cuts are motivated by separate, mutually incompatible, patterns. Matthews (1972) returns to this point in his discussion of the theoretical implications of Latin conjugation. Matthews stresses at the outset of his study that the main challenge faced by segmentation-based models of analysis does not arise in imposing cuts but in evaluating alternatives. In the simple case of the infinitive ferre ‘to bring, as he notes, there is language-internal motivation for each of the segmentations fer-r-e and fer-re. Selection of either of these choices disrupts the other pattern.

Genitive plural forms in Estonian present an instructive case in which “one comparison of forms suggests one placement, while another comparison suggests another”. There are two genitive plural patterns in Estonian, partly conditioned by the metrical structure of the stem. Nouns with non-trochaic stems form genitive plurals in -te; nouns with trochaic stems, along with loans and other members of a defective ‘fourth’ declension, form genitive plurals in the default-de.7 This variation is illustrated in Table 2.1 by the grammatical case forms of the basic Estonian nouns sikk and raamat and the loan auto.

A comparison of the forms in Table 2.1 determines an unambiguous segmentation of the forms sikkude and autode into the vowel-final stems sikku and auto and the genitive plural ending -de. Assigning a similar analysis to raamatute, as proposed by Tuldava (1994:195), reinforces the cross-class parallel between the stems raamatu, sikku and auto, and establishes a paradigmatic contrast between -te and -de. Yet, as noted by Murk (1997:13), nouns with genitive plurals in -te also exhibit a distinctive class-internal pattern. Comparison of the genitive plural with the partitive singular motivates a segmentation into a consonant-final base raamatut and a genitive plural marker -e.

The choice between the cuts of raamatute illustrate two problems for a segmentation-based analysis. The first is that the competing motivation comes from different parts of the morphological system: raamatu/te is motivated by cross-class comparisons and raamatut/e by a class-internal comparison. It may be possible to assign priority to one source by fiat, but there is no principled basis for this decision. The second, and more fundamental, problem is that segmentation forces a choice between two convergent analyses. In fact, raamatute contrasts simultaneously with the partitive singular in its own class and with the genitive plurals in other classes. Yet expressing either pattern via segmentation treats one pattern as significant and disrupts the other. [1]

Conundrums of this type call into question the usefulness of segmentation even as a tool for describing morphological structure. From a classical WP perspective, the problems that arise in segmenting forms reflect the intrinsically destructive character of decompositional analysis. The procedure that Lounsbury (1953) outlines starts from a comparison of a set of forms, which identifies patterns of overlap and alternation. Abstractive generalizations stated over these forms can capture the full range of variation, including any convergent patterns. In contrast, the use of decompositional strategies tends to disrupt at least some of the patterns, and isolates ‘recurrent’ units, which, on closer analysis, often exhibit patterns of subphonemic variation.[2]

  • [1] Where d is a short voiceless stop and t a long counterpart. See also Erelt et al. (1995, 2013) andBlevins (2008a) for more discussion of noun classes in Estonian.
  • [2] As discussed in Kemps et al. (2005); Plag et al. (2016) and Chapter 8.3.
< Prev   CONTENTS   Source   Next >