I continued my argument for the two domains approach, in the following way: I identified various features which tend to get bundled together in things that people roughly call "verbs" (largely coextensive with things that can combine with tense, but excluding modals and possibly also auxiliaries and light verbs, and including verb-particle collocations and various idioms that include a verb). These include:

Specification of manner (e.g. run vs. walk)
Specification of result (e.g. clean vs. wash)
Argument structure (e.g. eat vs. dine vs. devour)
Aktionsart (e.g. stative vs. achievement vs. accomplishment vs. activity)
Animacy of arguments (e.g. in Algonquian languages)
Category of arguments (e.g. PP vs. CP vs. DP)
Case of arguments (e.g. dative vs. accusative)
Conjugation class

We can find these features bundled up in a single verb without necessarily expecting other verbs to have them, and without necessarily expecting to find minimal pairs in the language for every combination of such features. In this way I want to contrast these features with "paradigmatic" features like tense: if you have a past/present distinction in one verb, then you expect to find it in all verbs, or at least all verbs in a semantically coherent class.

I am suggesting that all of the above features belong to the lower domain, the one that is contained within the first "phase" above the verb root, and that this is a unit that tends to be lexicalized or listed, with unpredictable properties. Outside that domain are such features as

Force (Interrogative vs. Declarative)
Topic
Focus
Epistemic modality
Tense
Deontic modality
various kinds of outer aspect, including the perfect, the progressive, etc.

As you will recall from last time, the claim is that things in this latter group might form words or idioms, e.g. modals, but no verbs will form that span the boundary (certain questions remain regarding whether this boundary is relevant for adjectives; my suspicion that it is not has led to my talking too much about "events").

Then I argued that the features responsible for determining gross clausal word order are generally also in this latter group, rather than the first group. For example, we see V-X and X-V alternations depending on:

Finiteness of V (German)
whether X is a pronoun or not (French)
whether X is a directional or locative PP (Dutch, Chinese)
whether X is DP or PP (Chinese)
whether X is definite, focused, topical, specific, etc.

We also see verbs varying according to all the different features in the first cluster. But, I contend, we do not find verbs varying according to whether their complement is on the left or the right. Someone, I believe it was Patrycja, pointed out the classic case of Italian in which the unaccusatives allow postverbal subjects more freely than do so-called unergatives, e.g. you can say "arrived Gianni" in contexts which do not allow *"telephoned Gianni" (i.e. the absence of subject focus, I think). This is an interesting counterexample, of a sort, though it really involves an option that is available depending on whether an argument is external or not, so in one sense is not about complements; but the point is a fair one since I think subjects of all types originate inside the lower domain.

My contention is actually that the heads which fix word order tend to be higher than the domain of lexicalization, and if this were not true then there should be languages in which one says "I eat snails" and "I eat baguettes" but "I wine drink" and "I cognac drink," i.e. languages in which verbs choose seemingly arbitrarily whether their complements will be to the left or the right. No theory since the days of the headedness parameter makes the correct prediction about this. Of course, it is simple enough to stipulate, in our theories, but it does not follow from anything at all, if verbs are decomposed and there are multiple object positions (as we all now believe), that some objects are not to the left and others to the right of their thematically introducing heads.

There are languages like West Greenlandic, in which only some verbs incorporate: this is another counterexample, of sorts; it means that *some* of the heads which determine gross clausal word order are relatively low. But I think that in general, an even stronger claim can be made: not only do we not find "I eat bread" vs. "I beer drink" in the same language (apart from the incorporation case), we don't even find "I eat bread" vs. "I bread like," i.e. even verbs from quite different semantic classes tend to show the same basic word order options. I think this is because the accusative case licenser is outside the first phase. That has the important result that even your typical VO language like English probably has the object outside the verb phrase (identified as the part of fseq that includes the stuff in the first cluster of features I mentioned).

We discussed some details of Slavic morphology in this context; my contention was that the regularity of tense (including the past -l forms) and agreement, compared to the irregularity of the thematic vowels and the secondary imperfective, suggested that the boundary was right between the two. My attempt to further pinpoint the boundary by decomposing the secondary imperfective into a part below and a part above it ran into some trouble, and it turned out that details of motion verbs were crucial. Fortunately Zhenja will work that all out for us.

Peter Svenonius