Desktop version

Home arrow Language & Literature

Syntagmatic uncertainty

The relationship between unit size and grammatical uncertainty guides the discussion of examples in previous chapters, so a brief summary of some of the relevant patterns will suffice here. Consider first the Georgian verb paradigm in Table 6.5, repeated below in Table 7.1, with just plural suffixes in bold.

At the word level, 16 of the 18 forms in Table 7.1 can be associated with a unique cell. In some forms, it is the presence of agreement markers that is distinctive, while in others it is the absence of markers that is contrastive, as Anderson (1992) remarks in the passage on p. 56 above. A form with no prefix can only encode a 3p object. Forms encoding 1sg objects are marked by m-, those encoding 1pl objects are marked by gv- and those encoding 2p objects are marked by g-. A form with no suffixal marker can only encode a singular subject, since all forms encoding 3pl subjects are marked by -en and all forms encoding 1pl or 2pl subjects are marked by -t. The most extreme example of distinctive absence is provided by daxatav,

Table 7.1 Future indicative paradigm of xat’va ‘paint’ (Tschenkeli 1958: §31)




































1 An abstract paradigm ‘cell’ corresponds to a set of distinctive properties for a given word or inflection class, and the paradigm itself consists of a set of cells associated with an item. The forms that express these properties are described as ‘realizations’ of the cell—in the traditional sense of ‘formal expressions, not in the more recent sense of ‘outputs of spell-out rules’. An ‘instantiated paradigm’ consists of a set of cell-form pairs. The intended sense of cells, paradigms and realizations should be clear from context.

Table 7.2 Contrasts marked by-t




i 2p object

pl vs sg

dagxatavt vs dagxatav

ii 2p subject

pl vs sg

damxatavt vs damxatav

iii 1p subject

pl vs sg

davxatavt vs davxatav

in which the lack of prefixal and suffixal markers unambiguously encodes a 2sg subject and a 3p object.

The two ambiguous forms, dagxataven and dagxatavt, both occur in the subparadigm with 2p objects. Ambiguity arises in this part of the paradigm in Table 7.1 due to the fact that 2p g- occupies the prefixal agreement position, leaving only the suffixal slot to encode the number of the object and the person and number of the subject. Thus dagxataven realizes the two cells with a 3pl subject, while dagxatavt realizes four of the remaining cells with 2p objects.

In sum, the markers in Table 7.1 define 18 distinct word forms, distributed over 22 paradigm cells. Significantly, the word-level ambiguity in this paradigm cannot be resolved by decomposing forms into sub-word units, even in cases where the units can themselves be assigned unambiguous analyses. In the case of dagxataven, the prefix g- unambiguously marks 2p object features and the suffix -en 3pl subject features. However, the use of -en to mark 3pl subject features prevents the marking of object number. In cases where sub-word units are also ambiguous, decomposition can lead to an analytical conundrum.

Unlike the unambiguous -en, the plural suffix -t in dagxatavt marks a contrast between plural and singular in the three ‘contexts’ in Table 7.2. Contrasts (i) and (ii) share 2p features but differ in grammatical relation, whereas (ii) and (iii) share a common grammatical relation but differ in person. Contrasts (i) and (iii) differ both in person and grammatical relation. Hence, there are essentially three options available for a model that seeks to assign a feature analysis to -t. One analysis would introduce separate analyses for each of the contexts in Table 7.2, a second would subsume (i) and (ii) under a general 2p marker, and a third would subsume (ii) and (iii) under a general subject marker.

No coherent analysis could subsume just (i) and (iii), given the contrast in person and grammatical relation. Yet the cell in Table 7.1 that specifies a 2pl object and a 1pl subject is occupied by dagxatavt, which is marked by a single occurrence of -t. A model that associates properties with sub-word units must determine ‘which -t’ marks this form. But there is no basis for choosing either (i) or (iii), and no plausible means of consolidating them into a single marker.

This is the conundrum that Anderson (1992) attempts to evade by proposing a ‘general plural’ analysis on which there is a single -t marker. However, as noted in Chapter 6.3.2, the analysis that Anderson formulates does not state a rule so much as describe a disjunctive set of rules. A different strategy is pursued by Stump (2001:70), who consolidates (ii) and (iii) into a common marker -t1 and represents (i) by a separate marker -t2, which is restricted so that it “appears only in the context of singular subject agreement”. At one level, this is a fairly unprincipled

Table 7.3 Representative case forms of pukk ‘trestle’

Grammatical Cases

Semantic Cases























stipulation; one could just as well consolidate (i) and (ii) into a common 2p marker -tu and associate (iii) with a marker -t2 that is restricted so that it ‘appears only in the context of singular object agreement’. At a more basic level, this analysis represents a tacit acknowledgment that the distribution of -t cannot be specified in terms of a single set of agreement properties. Instead, as suggested in Chapter 6.3.3 and elaborated in Chapter 8.3, markers function in the context of larger systems of property-form contrasts.

Although the analyses proposed by Anderson (1992) and Stump (2001) are couched in a realizational idiom, they confront a version of the general ‘overextraction’ problem that arises in Post-Bloomfieldian approaches. The association of properties with individual markers tends to increase uncertainty by disassembling words into parts that are more ambiguous and/or more resistant to analysis than the wholes from which they are obtained. This uncertainty is not inherent to the language itself. Instead, it is symptomatic of analyses that apportion properties to units of form that do not function as signs, but as markers that serve to discriminate larger meaningful units. The issues that these analyses create are largely independent of whether the analyses are formulated in terms of morphemes or realization rules.

The uncertainty that arises in associating grammatical properties with inflectional markers is not a peculiarity of Georgian, which exhibits a pseudoagglutinative structure that is nearly as favourable as possible to decomposition. The uncertainty created by disassembling more fusional patterns is even greater, as discussed in previous chapters. Consider the forms of the Estonian noun pukk ‘trestle’ in Table 7.3, repeated from Tables 4.7 and 4.8.

The forms in each of these columns are fully discriminable and, with the exception ofpukki (which realizes the short illative singular as well as the partitive singular), they can all be assigned a single grammatical analysis. However, the attempt to associate grammatical properties with sub-word units of form only increases uncertainty. In the singular grammatical cases, the units pukk, puk, and -i combine to express the three case contrasts. However, none of these units can be assigned grammatical properties in isolation that ‘add up’ to the properties of the word forms. The same pattern recurs in the plural grammatical cases. The forms puki and pukki combine with -d, -de and -sid to express a three-way contrast. But isolating the bases creates ambiguity between their uses as case forms in the singular and as case- and number-neutral stems in the plural. The same pattern recurs yet again in the semantic cases, where the bases are ambiguous between genitive forms and case-neutral ‘oblique’ stems.

Table 7.4 Priscianic syncretism in Latin

AMO ‘to love’

MONEO ‘to advise’

tego ‘to cover’









Past Pass Prtl




Fut Act Prtl




Stem syncretisms in Latin conjugations exhibit a similar pattern, differing only in that the recurrent stems function as bound forms and never occur in isolation. The forms in Table 7.4 (repeated from Tables 4.12 and 5.3) contrast the present stem, which occurs in the infinitive, with the supine stem which occurs in the supine, past and future participles. As in previous examples, the full word forms can be associated with determinate grammatical properties. However, as discussed at length by Matthews (1972), Aronoff (1994) and Maiden (2005), syncretic stems in Romance cannot be assigned a determinate analysis in isolation. The supine stems in Table 7.4 can either be assigned different features in each context in which they occur, or they can be treated as grammatically indeterminate or ‘underspecified’ in those contexts.

In each of these examples, disassembling a word into smaller recurrent parts disrupts an initially stable association between form and grammatical properties, increasing ambiguity and uncertainty. The locus of the uncertainty depends on general characteristics of the language. In Georgian, disassembly isolates agreement markers that cannot be assigned a determinate analysis. In Estonian and Latin, the analysis of words into recurrent partials creates classes of grammatically indeterminate stems. In Latin, the disassembly of fusional forms creates further indeterminacy in the division of stems and inflectional endings. Technical strategies can be devised to manage ambiguity, and policies can be imposed to guide segmentation, as discussed in Spencer (2012). But these proposals all serve as correctives, addressing challenges that arise when, as Matthews (1972:74) remarks in connection with principal parts,“the theory creates a problem which it is then unable, or only partly able, to resolve”.

< Prev   CONTENTS   Source   Next >

Related topics