Schachter (1976:506) presents a generalization that if an imperative lacks an overtly indicated subject, a second person subject is understood. I would not have found this claim particularly noteworthy, other than for its mysterious attribution (Schachter cites an unpublished manuscript by Bell (1974), who allegedly cites Kenneth Hale), were it not for an (unpublished) counter-observation made by Ivan Sag. He noted that the old Indo-European languages such as Latin and Sanskrit had forms that were considered third person imperatives, even without overt subjects; and expressions such as French allons-y are considered to be first person, even without an expressed nous. In this squib I should like to investigate whether Sag's examples actually constitute a valid counterexample, and if so, how much damage is done. Is the hypothesis of Hale (or Bell's interpretation of it, or Schachter's interpretation of that; henceforth, `the HBS generalization') a statistical universal, for which we must be content with a certain amount of material on the lighter side of the scale? Or can it be tightened up so that it makes less sweeping but more valid claims?
For this study, I looked at a sample of about fifty languages in varying amounts of detail, ranging from a rather careful look at some of the Indo-European languages, especially those mentioned by Sag, down to a more or less unexamined recording of cursory statements recorded in basic grammars. Results must be considered preliminary, but it appears at this point that the HBS generalization about 2p imp. subjects needs to be revised and reformulated under the following three heads.
There is a cross-linguistic general relationship between the imperative mood (imp.) on the one hand and number and person on the other. If a language has any distinctively marked imperative form, it will minimally be the second person singular (2p sing.). If it has another, it will be the 2p plural (pl.). If another, 1p pl. inclusive or 3p. The 1p sing. is the form least likely to be found in an imperative system. Such patterns are rarely found in other moods, such as the subjunctive, which is often pulled in to fill in gaps in the imperative paradigm.
When the imperative system covers more than one person, these categories of person and number may be conflated, so that the person or number of a particular form may be underdetermined. When person is left ambiguous, because a verb form is so underdetermined and there is no explicit subject, then there is a strong conversational implicature that a command addressed to a party is meant for that party to carry out, i.e., has an implicit 2p subject. However, at least in the case of pro-drop languages, languages seem in general to leave open the possibility that the command is meant for a third party, that is, that a 3p subject is to be supplied. The languages of Latin and Sanskrit are particularly clear examples of this, and reveal a serious weakness in Schachter's formulation of the HBS generalization.
The unmarked nature of the 2p imperative is further indicated by the fact that cross-linguistically, in languages that mark person, the 2p sing. is very often left unexpressed. If any form has no personal affix in the imperative, its scope will include the second person. If any imperative can be used without a subject, the 2p pronoun can be omitted. This tendency to leave the 2p or 2p sing. unexpressed may be related to a more general tendency to make the imperative be the shortest form of the verb.
The body of this paper is available as a PostScript file.