July 11, 2004

"Losing" "the subjunctive"

Following up on Mark Liberman's posting about "subjunctive case" in if I were you (where the appropriate label is "mood", not "case"), Geoff Pullum (7/1/04) notes that "subjunctive" isn't a very good label, either, and suggests "irrealis" as an alternative. I'm going to use Geoff's brief comment as a springboard to combat two common misapprehensions about inflectional morphology and its relationship to syntax and semantics and to question a common analysis of inflectional forms (like irrealis mood) that can easily be seen only for a few lexemes -- what I'll call underbrush forms, because they lurk concealed in the morphological underbrush.

The first misapprehension -- which Geoff clearly doesn't hold, though a careless reader could miss this -- is that there is a simple relationship between syntactic properties of phrases or clauses, morphological properties of words, and semantics. The second misapprehension -- which Geoff also doesn't hold, though the terminological focus of his comment conceals this -- is that the primary issue in analyzing phenomena like the English so-called subjunctive is how to label the forms. The questionable analysis -- which Geoff actually proposes -- is that an underbrush form is simply lacking for most lexemes, with some other form standing in for it.

First, Geoff's comment, in its entirety:


It isn't actually the subjunctive. People often call the "were" of "I wish I were" subjunctive, but that term is much better used (as in The Cambridge Grammar of the English Language) for the construction with "be" seen in "I demand that it be done." The "were" form is often wrongly called a past subjunctive, but of course "it were done" is not a past tense of "it be done". The difference between the two is that the subjunctive construction occurs with any verb: "I demand that this cease" is a subjunctive (notice "this cease", not "this ceases"). The relic form in "I were" is only available for "be". For all other verbs you use the preterite: "I wish I went to New York more often." The Cambridge Grammar calls the "were" form the irrealis form. It is surviving robustly in expressions like "if I were you", but even there it has a universally accepted alternate "if I was you", and there is no semantic distinction there to preserve.


1. The very careful reader will see that Geoff sometimes talks about "constructions" and sometimes about "forms" (and that he makes no explicit reference to semantics). This is a distinction between syntactic properties of phrases or clauses (the expression "it be done" in "I demand that it be done" has, among others, the property Constr:285, for which CGEL uses the label "subjunctive") and morphological properties of words within these larger expressions (the verb word "be" in "it be done" has, among others, the property Form:U, for which CGEL uses the label "plain form", though others call it the bare, base, unmarked, infinitive, or unmarked infinitive form). Geoff provides no label for the construction of "I were you" (I'll call it the "plain counterfactual", Constr:286). For the property Form:I of "were" in "I were you", CGEL provides the label "irrealis".

Semantics gets into the act by being associated with phrase properties; strictly speaking, a construction is a pairing of (a) a package of conditions on the internal syntax of expressions, like being a finite clause with a head verb word of Form:U, with (b) conditions on the semantic interpretations for those expressions, like denoting an obligation.

In many, possibly most, situations the distinction between syntactic properties and morphological properties is too obvious to be missed. This is because so many inflectional forms are multifunctional, even flagrantly so: they occur in a variety of constructions, of otherwise varied syntax and semantics. Form:U, for example, is used in imperative clauses ("Be quiet!"), with infinitival "to" ("to be quiet"), with modals ("will be quiet"), in complements of certain verbs ("made them be quiet"), and in several other diverse constructions, including the plain counterfactual.

Every so often, though, an inflectional form is closely tied to one particular construction, and then it's tempting to identify the form with the construction (and with the semantics for the construction). This is the case with Form:I and the plain counterfactual construction. So maybe it's not entirely an accident that Geoff failed to give a label for Constr:286; it might not have seemed necessary.

Tempting, yes. But you should resist the temptation. This line-up of syntax, morphology, and semantics is a fluke, something that happens with new inflectional forms (which have not yet developed further uses) and moribund ones (which have lost their other uses, though replacement by innovative syntax).

Sometimes it turns out that there are other constructions that use an inflectional form. That's true here: there's Constr:287, which I'll call the inverted counterfactual, as in "were I your teacher". This is clearly different syntactically from the plain counterfactual (in addition to Subject-Auxiliary Inversion, the inverted counterfactual is also incompatible with a subordinator like "if"), though it shares with the plain counterfactual the use of Form:I and the contrary-to-fact semantics.

2. I've been providing arbitrary designations for both phrase properties (Constr:286) and word properties (Form:I), along with suggestive labels of my own devising or from CGEL (plain counterfactual, irrealis). It's important to realize that these suggestive labels play absolutely no role in the description of the language. If they're well chosen, they allude to some relevant aspect of syntax or semantics, but the labels are in no way descriptions, of either the syntax or the semantics.

So there's no substantive issue here. "Irrealis" is a much better name for Form:I than, say, "cislocative" or, for that matter, "elephant", but it's at best a hint at the semantics of the constructions in which it occurs.

3. Now we come to the Missing Form analysis of underbrush forms like Form:I. To describe the facts fairly neutrally: only one lexeme, BE, has a Form:I distinct from other forms; for all other verbs, in constructions that call for Form:I, a form identical to Form:T (variously called the past or the preterite) is used. Geoff's translation of this -- this isn't something he devised, it's a very common formulation -- is that only the lexeme BE has a Form:I, and that all other verbs lack a Form:I, using instead Form:T.

There are two parts to the Missing Form analysis. First, the specification of a lexical gap; almost all verbs have a Form:I gap, and in this respect they are like the modal verbs of English, which all lack Form:U, Form:N (the present participle, gerund participle, or -ing form), and Form:P (the past participle, or -en form). But the Form:I gap is unlike the gaps for the modals, in that, in a second piece of the analysis, a subsidiary principle fills the gap by supplying Form:T. There is some question about the details of the Missing Form analysis, but first let's look at the alternative, the Syncretism analysis, and ask why someone might reject it.

In the Syncretism analysis, every verb lexeme has a Form:I. For BE, this is stipulated to be "were"; otherwise, a rule of morphology (a "referral rule", in the terminology of an old paper of mine; the idea is developed at length by Greg Stump in his 2001 book Inflectional Morphology) says that Form:I for a verb is the same as Form:T for that verb.

So what's the problem? Well, lots of linguists think that there's something wasteful about having all those redundant Form:I's listed for every verb lexeme in the language except one. Frankly, I've never understood this objection. Nobody's claiming that people keep all this stuff stored in their heads as a big list, any more than anyone claims that people keep all those regular, perfectly predictable Form:T's in /d/ ("stored", "jumped", "batted") stored in their heads as a big list.

In any case, these objections normally arise only for a small set of morphological anomalies involving special forms, those in which three conditions are all satisfied: (1) the number of lexemes showing the special forms is very small (for Form:I, this figure is 1); (2) the number of (morphological) principles for alternatives to the special forms is very small (for Form:I, this figure is 1); and (3) the number of constructions that use the special forms is very small (for Form:I, this figure is approximately 2). This constellation of characteristics isn't rare: accusative (vs. nominative) forms for English Ns, partitive (vs. genitive) forms for Russian Ns, locative (vs. prepositional) forms for Russian Ns, vocative (vs. nominative) forms for Latin Ns, and many more. (I use more or less traditional names for the forms, rather than engaging in a gigantic meticulous renaming.)

The difficulty here is that each of these three conditions represents one pole of a cline that extends far in the other direction, and there's no motivated point at which you can say that the numbers are no longer small enough to justify the Missing Form analysis. Look at the far end of cline (1): Form:2 (usually called "plural") in English. For almost all English Ns, Form:2 is distinct from Form:1, but there are a few -- SHEEP is the classic example -- usually analyzed as "zero plurals" (as syncretic). No one says that SHEEP lacks a Form:2, and that constructions calling for a Form:2 use the Form:1 "sheep" instead. But that would be the parallel to the Missing Form analysis of Form:I for verbs. In between these two examples there are all sorts of things, some of them involving very complex patterns of identities (see the examples in Stump's book), that is, high numbers on cline (2), and some involving multifunctional forms, that is, high numbers on cline (3). A good framework for morphology would cover the whole territory, rather than carving out one small portion of the territory for a special Missing Form treatment, admitting syncretism everywhere else.

Now return to the details of the Missing Form analysis for Form:I. The issue is whether the gap-filling is done in the morphology or the syntax. If it's done by a morphological principle or principles, then there really is no difference between the two analyses; "use Form:T if there is no Form:I" is a referral rule under another name. If it's done in the syntax, then any construction calling for Form:I comes with the proviso that if Form:I is lacking, you use Form:T -- a kind of principle linking syntax to morphology. This version turns out to have empirical consequences, and for Form:I, those consequences are not nice.

Here's the thing... As Geoff observed in his comment, English changed, in a way that is usually described as the "loss" of Form:I: no verb lexeme has a Form:I any more, with the result that Form:I was replaced by Form:T. (Well, this isn't what happened, of course. Innovative forms always coexist for at least a while with the forms they replace. In fact, many speakers today still use both versions, and a great many people understand both.) The Missing Form analysis predicts that all constructions using Form:I would shift together, but this is not what happened: "if I were your friend" developed an alternative "if I was your friend", but "were I your friend" didn't develop an alternative "was I your friend (i would tell you the truth)".

On the Syncretism analysis, syntactic constructions behave separately; they can call for Form:I or Form:T or whatever. And they can change independently. Which is what happened in this case. The innovative construction, call it Constr:288, is just like Constr:286 but calls for Form:T. The three counterfactual constructions coexist for sometime -- in my variety of English, for example. When Constr:286 and Constr:287 no longer occur with sufficient frequency for new generations to learn them, then there will be no evidence left for a Form:I in the language, and it will genuinely be lost (as it surely is for some speakers already).

zwicky at-sign csli period stanford period edu

Posted by Arnold Zwicky at July 11, 2004 02:46 AM