True Truncation in Colloquial Hebrew Imperatives

There are two types of truncation that yield shortening of a morphological constituent, fake truncation (templatic) and true truncation (a-templatic, subtractive). This article provides an analysis of true truncation in colloquial Hebrew imperatives. It is shown that true truncation cannot target a designated phonological unit, since in some forms CV is truncated and in others only V. In addition, there are cases where truncation is blocked. The framework of optimality theory adopted here allows a unified account of the data in terms of constraint interaction. It is argued that an antifaithfulness truncation constraint, which must be morphological, interacts with both faithfulness and markedness constraints. Truncation is minimized to one segment by a general antideletion faithfulness constraint, but markedness constraints may impose truncation of more than one segment. There are cases where truncation is blocked, which suggests that the truncation constraint is violable. The discussion includes regular and irregular verbs and instances of free variation.^*

1. The issues

The term true truncation refers here to a direct requirement of truncation in the derivation of one lexical category from another. It differs from fake truncation, whereby truncation is a byproduct of the imposition of templatic constraints. This article is concerned with truncated imperatives in colloquial Hebrew (hereafter TIs), which are a case of true truncation.¹

A TI is shorter than its future base in that one or two segments are truncated from the 2nd person future prefix. The truncated material is not a morphological unit since in some cases it consists of only part of the prefix. In ti-∫mor/∫mor ‘to guard.fut/TI’ the entire prefix is truncated, but in te-xapes/txapes ‘to search.fut/TI’ only V is truncated out of the CV prefix. These examples also show that the truncated material is not a designated phonological unit, as in some cases V is truncated and in others CV. These facts are rather surprising considering documented cases of true (nontemplatic) truncation, such as Tohono O’odham perfectives, Danish imperatives, and Koasati plurals, where the truncated material seems to be a designated phonological unit. In addition to the variable truncated material, there are cases where truncation is blocked, as in tirkod ‘to dance.fut’ and tagdil ‘to enlarge.fut’, which are used for both future and the imperative.²

It is necessary to question the motivation behind truncation. Clearly, truncation cannot be expressed in a purely phonological fashion, as it affects only a specific morphological category and the size of the truncated material varies. Truncation does not act like the familiar morphological processes, which usually add segmental material (with the exception of conversion, where nothing is added or deleted).

I argue that truncation is imposed by a morphological constraint (Truncation), and thus deletion is an inherent requirement of the constraint. Truncation does not limit the truncated material to a designated phonological unit; it just states that ‘not every segment in the input has a correspondent in the output’. This statement is the negative expression of the faithfulness constraint MaxSeg, thus reflecting the inherent antifaithfulness of the constraint. Antifaithfulness constraints have been proposed in Alderete 1998 and 2001 to account for morphophonological alternations, and the analysis of true truncation provides further empirical support for the theory developed in these studies (see also Horwood 2001). I also suggest that Truncation is a universal constraint, active in every language exhibiting true truncation (regardless of the size of the truncated material). The phonological shape of the output, and thus of the truncated material, is determined by the interaction of Truncation with other constraints and thus can vary. MaxSeg, the faithfulness counterpart of Truncation, minimizes truncation to one segment, as truncation of two segments satisfies Truncation as much as truncation of one segment but has more violations of MaxSeg. Therefore the TI of ti∫ava ‘to swear.fut’ is t∫ava (V truncation) rather than *∫ava (CV truncation). Markedness constraints, when ranked above MaxSeg, may force truncation of two segments in cases where truncation of one segment violates some restrictions on...