Desktop version

Home arrow Language & Literature

Implicational structure of inflectional series

German declensions illustrate a simple type of multiple-part system in which each of the principal parts is of high diagnostic value within a sub-network and a weaker implicational relation holds across networks and between principal parts. Conjugational systems are often structured in this way, comprising multiple subparadigms, organized around larger inventories of principal parts.[1] If a system is extremely simple or highly uniform, a single principal part may still suffice. For example, Bonami and Boye (2003) propose that regular verbs in French are based on a single principal part. In a subsequent study (Bonami and Boye 2007), they argue that irregular verbs can also be based on a single principal part, provided that the part may be chosen ‘dynamically’ (Finkel and Stump 2007), that is, not necessarily the same principal part for each verb.

Languages in which regular items can be represented by a single diagnostic form conform to a pedagogical ideal of inflectional economy. A similar conception underlies more theoretical principles, such as the Single Base Hypothesis of Albright (2002). However, the variation within conjugational systems is often not predictable from a single form. Instead, the paradigms of a verb cluster into groups, usually termed ‘series, based on a common or similar stem. Descriptions of these types of systems specify a principal part for each sub-paradigm or series. Latin conjugations provide a familiar illustration:

Certain forms of the verbs are known as the Principal Parts, because they furnish the key to the inflection of any given verb, showing, as they do, the Present Stem and thereby the Conjugation, and the Perfect and Participial Stems. These are: 1. The Present Indicative Active, cited in the First Person Singular. 2. The Present Infinitive. 3. The Perfect Indicative Active, cited in the First Person Singular. 4. The Perfect Passive Participle, cited in the Nominative Singular Neuter. (Hale and Buck 1903:77)

Principal parts representing the four basic Latin conjugations are given in Table 4.12. Each class is represented by four forms of an exemplary verb. The first pair, the indicative active and infinitive, are drawn from the present series. Standard grammars represent the present indicative by the 1sg form, given in parentheses in Table 4.12. However, the uniform ending of this form makes it an unsuitable principal part. In particular, the 1sg present indicative neutralizes the contrasts between classes I and II and between classes IITh and IV. These contrasts are preserved in the 2sg forms in Table 4.12.20 The second pair, the indicative active and passive participle, are drawn from the perfect series. The need for four forms of each verb reflects the fact that no two rows of forms exhibit congruent inflectional patterns. For example, inspection of the present series shows why the classes in Table 4.12 cannot be represented by a single principal part. Selecting just present indicative forms would not distinguish third conjugation capio from fourth conjugation audio. Selecting present infinitive forms would distinguish capere from audire but at the cost of losing the contrast within the third conjugation between tego and capio. Similar considerations justify each of the standard principal parts. But even an inventory of four principal parts achieves considerable economy, given that Latin verbs may have up to 150 inflected forms and periphrases.

Although the present forms in Table 4.12 “show... the Present Stem and thereby the Conjugation”, the present Stem is not itself a principal part, but rather a ‘recurrent partial’ that is exhibited by the word forms that do serve as principal

Table 4.12 Principal parts of Latin verbs (Hale and Buck 1903: 82ff.)



Indic Act


Indic Act

Pass Prtl


amas (amo)




‘to love’


monas (moneo)




‘to advise’


tegis (tego)




‘to cover’






‘to take’


audis (audio)




‘to hear’


I am grateful to Olivier Bonami for drawing these points to my attention.

parts. A classical WP model is consistently word-based, at the level of grammar and lexicon. Like the exemplary paradigms that instantiate the morphological patterns of a language, the principal part inventories that represent the lexicon consist of whole word forms that realize specific paradigm cells.

As this description of Latin illustrates, a highly diagnostic principal part may become established as a citation form within a grammatical tradition. However, it is important to bear in mind that the value of a principal part is determined by its informativeness with respect to other forms, not by its status in terms of notions like morphosyntactic or morphotactic ‘markedness’. Forms whose unmarked status is due to the neutralization of within-item variation are often among the least useful principal parts, since they tend to neutralize class-defining variation. At the other extreme, highly infrequent or even archaic forms may preserve diagnostic phonetic characteristics that have been reduced or otherwise obscured in more frequent forms. This type of diagnostic value is of limited relevance to any analysis that is meant to underlie a model of naturalistic language acquisition or use. The essential prerequisites for principal parts are that they occur above a nominal frequency threshold (i.e., be in active circulation) and that they exhibit class-identifying variation.

The first of these conditions is distributional and the second is implicational. Both are independent of morphosyntactic or morphotactic considerations. Hence principal parts maybe found in nearly any cell of an inflectional system. This point is illustrated by the principal parts of the Russian conjugational system. Russian verbs can be assigned to two basic conjugations based on the stem-final vowels in present paradigms. In all present paradigms, the isg ends in -u. In the first conjugation (or ‘e-conjugation’), the vowel -u also occurs in the 3pl, and -e occurs elsewhere. The second conjugation (or ‘/-conjugation’) exhibits a contrast between -a in the 3pl, and -i elsewhere. These patterns are displayed by the present paradigms in Table 4.13.

The lexemes delat’ and govorit’ exhibit regular inflectional endings and show no stem allomorphy, as illustrated by the first rows of forms in Table 4.14. Hence a single principal part could identify both the stem and conjugation vowel of these items, if all verbs were equally uniform. Yet they are not, as the forms in the second row in Table 4.14 show. Hence one needs to know that the stems of a verb do

Table 4.13 Exemplary present paradigms in Russian (Wade 1992:23off.)






















‘to do’

‘to speak’

And further subdivided based on stem patterns (Timberlake 2004:99®).

Table 4.14 Standard conjugational series in Russian (Unbegaun 1957:166)

Present Series

Past/Infinitive Series





Past Masc

Past Prtl

I delaju






II govorju






I iscu






II smotrju






II vizu






not alternate, information that is most efficiently expressed by a second principal part. So a single principal part does not suffice, even for fully regular verbs. The general patterns exhibited by verbs with alternating stems are illustrated by the forms of iskat’ ‘to find, smotret’ ‘to look at’ and videt’ ‘to see’ in Table 4.14. The lexeme iskat’ shows the characteristic first conjugation pattern, in which stems in the present series exhibit consonant mutation. The lexeme smotret’ ‘to look at’ exhibits a similar series-level split, in which the stem-final vowel -i in the present series alternates with -e in the past series. The lexeme videt’ combines both patterns. Consonant mutation occurs just in the isg, as is characteristic of alternating second conjugation verbs, and there is a series-level contrast between -i and -e.

The Russian conjugational system exhibits much additional variation, though the patterns in Table 4.14 suffice to clarify the usual motivation for distinguishing a present stem, which underlies the present series, from an infinitive stem, which underlies the past series. A fundamentally similar analysis is preserved in more theoretical accounts, such as Brown (1998):

It follows therefore that there are two main variants of the stem, or two allostems. These are the stem used in the infinitive and the past tense and the stem used in the present tense. (Brown 1998:199)

Since most forms of an item tend to be at least partially informative about class, selecting a specific form again reflects the essentially pedagogical goal of assigning a maximally economical and uniform description. Forms in the past series are mutually predictable, so the infinitive is as good a choice as any. However, there is greater variation in the relative informativeness of present forms. As noted already, the 1sg is the least informative, so it is a poor choice of principal part (much as in the Latin paradigms in Table 4.12). For the purposes of identifying class, the stem vowels in 3pl forms are as informative as the stem vowel in the remaining present forms. However, some verbs exhibit further consonant mutations in all forms but the 1sg and 3pl. This pattern is illustrated by the forms of moc’ ‘be able to’ in Table 4.15.

Hence the most informative present form is the 3pl form, as it preserves a class-identifying vowel and unmutated stem. The diagnostic value of this form is of course frequently noted in descriptive and pedagogical grammars. Wade

Table 4.15 Present and Past Series forms of moc’ ‘be able to’



3pl Infinitive

Past Masc

Past Prtl

I mogu


mogut moc’



(1992:228) uses the 3pl as the reference form for the present series in proposing that “[t]he present-future stem of a verb is derived by removing the last two letters [i.e. segments] of the third-person plural of the verb”. Levin et al. (1979:195) goes even further in suggesting that “[t]he 3rd Person plural implies the entire conjugation membership”, at least for regular verbs.[2]

In sum, the variation in the Russian conjugation system motivates multiple principal parts. The past series can be represented by the infinitive citation form or, equivalently, by a past form or past participle. But the best representative of the present series is the 3pl present form (or any form, including the present Active participle, which predicts the 3pl present form). This choice reflects the fact that the value of a principal part correlates with its relative informativeness, not with its status as a citation or reference form. In one sense this is unsurprising. Given that all forms are at least partially informative, there is no obvious reason why a specific form should invariably be the most informative. As with other properties of linguistic systems, informativeness will tend to reflect various interacting and even counteracting factors. For example, any general tendency to align inflectional patterns with highly frequent forms maybe disrupted by the influence that frequency appears to exert in inducing or accelerating sound changes. As Bybee (2010) notes:

A robust finding that has emerged recently in quantitative studies of phonetic reduction is that high-frequency words undergo more change or change at a faster rate than low- frequency words. (Bybee 2010:20)

Individual forms may likewise exhibit characteristic distributional profiles across languages, reflecting differences in paradigm size, syntactic construction inventories, and general patterns of use. Hence the principal parts of a morphological system will, in effect, emerge from the language model that a speaker constructs based on frequency and patterns of co-occurrence. As discussed in greater detail in Chapter 7, the task of ‘choosing’ principal parts is one that arises in describing languages, rather than in acquiring and using them. A formal reconstruction of a classical WP model should provide an effective means of measuring the mutual information between forms. This measure should identify fully diagnostic forms (and sets of forms). But the task of ‘justifying’ the choice of ‘correct’ principal parts is essentially a pseudo-problem that reflects pedagogical idealizations about ‘compact’ grammatical descriptions.

From a classical WP perspective, one can also see stem-based descriptions of inflection systems as something of a pseudo-solution to this pseudo-problem. Representing lexical items by stems or stem sets avoids the need to select particular forms that display the stems. In the simpler Russian nouns in Section 4.1, the isolation of stems removes diagnostic endings. In verbs, the distinctive stem vowels and consonantal variation occur inside inflections, so stems retain information about class membership. For single-stem verbs like delat’ and govorit’, one stem form can both identify class and also underlie the full set of inflected forms.[3] However, the stems of multiple-stem verbs must be associated with appropriate sets of forms. Hence a stem-based description replaces the problem of principal part selection by a problem of stem selection. As in the description of Russian conjugations, the class-specific (or subclass-specific) distribution stems can be expressed by indexical features that associate stems and endings with paradigm cells or inflected forms.

Stem selection principles might appear to be less stipulative than the diacritic class features introduced to associate nouns and case endings. In particular, the fact that variation occurs within present and past series might suggest the possibility of keying selection to tense features of a stem. However, this idea rests on a pervasive overinterpretation of morphological terms. Although series are conventionally assigned morphosyntactic labels, such as ‘past’, ‘aorist’, ‘perfect’, etc., the forms in a series often share a common base rather than a set of grammatical properties. In Russian, the past and present series both contain nonfinite as well as finite members. The past series in Table 4.14 contains the infinitive and past participle; the extension of the present series in Table 5.1 adds the present participle. Hence the familiar designations ‘past’ or ‘present’ participle register the fact that these forms are based on the same stems that underlie the past and present paradigms, and do not reflect tense features that could regulate the distribution of the stems.

  • [1] Items with a purely etymological connection can also function as models for the deductionof inflected forms. Like the ‘pseudo-plural’ compound bases in German in Section 3.1.4 and thegrammatical case form bases in Estonian in Section 4.2, this is a ‘morphomic’ (Aronoff 1994) relationbetween items at the level of form. As Spencer (2013) notes, this type of pattern is characteristic ofprefixed Indo-European verbs.
  • [2] The diagnostic value of the 3pl is reinforced by the fact that it also predicts other present forms,notably the present Active participle, discussed in Section 5.1.
  • [3] Though as Timberlake (2004:98) notes, even regular verbs have contrasting present and paststems. For example, the past stem of delat’, dela, ends in a vowel, whereas the present stem, delaj,ends in a glide that triggers palatalization in Table 4.13.
< Prev   CONTENTS   Source   Next >

Related topics