The irrealis sector is an area of the grammar in which a number of linguistic forms tend to compete in fulfilling a given function. This often leads to conflict between normative prescription, as grammarians attempt to match form with function, and the evolution of the vernacular, which remains impervious to such pressures.

Adopting a variationist approach, which assumes that the same linguistic function may at times be realized in different forms, and making use of the Ottawa-Hull French Corpus (Poplack 1989), a massive compendium of spontaneous French conversations, this paper explores three heretofore undocumented shifts in semanticity among expressions of irrealis. I compare the variable expression of subjunctive mood, as in (1), future temporal reference, as in (2), and the formulation of conditions in hypothetical si- clauses (3) in spoken French.

'If they would do a little more, he would come back.'

Large-scale quantitative analysis of thousands of tokens of vernacular usage reveals not only that several forms compete in the expression of each of these areas, but that in each case, a non-standard option is preferred: indicative and conditional forms predominate in "subjunctive-selecting" contexts, the periphrastic future has largely replaced its synthetic counterpart in future temporal reference contexts, and the conditional is ousting the standard imperfect in hypothetical si- clauses.

These competing forms are far from evenly distributed across functions or linguistic contexts. Selection of subjunctive mood, though quantitatively the norm, is largely restricted to a few highly entrenched routines consisting of a single (morphologically irregular + highly frequent) matrix verb (falloir) and a small set of (morphologically irregular + highly frequent) embedded verbs. Choice of the synthetic future in -rai accounts for only a minority of future temporal reference expressions, and when frozen, formulaic and other coventionalized constructions are discounted, is observed to be retained virtually only in negated contexts.

Moreover, multivariate analysis of the relative contributions of syntactic, semantic, lexical and priming factors to choice of form reveals that despite a robust grammatical tradition imbuing specific meanings to the variability between subjunctive and indicative in the first case, and between inflected and periphrastic future in the second, these forms are not selected by speakers to perform any particular semantic or functional work. On the contrary, subjunctive morphology has no more than a 50:50 chance of being selected under matrix verbs other than falloir, where its occurrence is rather explicable in terms of priming effects and/or additional minor lexical associations. The inflected future is selected in under 20% of future temporal reference contexts. But aside from a few (weaker but still perceptible) lexical associations (many of which involve the very lexical items found to be associated with the otherwise non-productive selection of the subjunctive in the same corpus!), productive use has fallen below 6%, and is decreasing among the youngest speakers.

The third area, use of conditional for the standard imperfect in protases of hypothetical si- clauses, in contrast to its counterparts in the irrealis system, is fully productive within and outside of protases, and in no way lexically restricted. Here speakers have (re)"semanticized" the -rais form to express specifically potential or possible conditions in the present, while the (opaque) imperfect is relegated to the expression of past potential and counterfactual conditions.

Reexamination of these facts from a functionalist perspective shows that they exemplify well-documented usage effects associated with frequency, conventionalization and bleaching, among others. The difference in semanticity between the highly frequent but meaningless subjunctive and the frequent and (re)semanticized conditional suggests that, here at least, the determining factor is not so much frequency, but productivity. Productivity is inversely related to semanticity, explaining why the conventionalized subjunctive and inflected future forms have grammaticized (as marker of subordination and negative polarity item, respectively), while the highly productive alternation between conditional and imperfect in the protasis fulfills a semantic function.