Lecturette #28: Laying the Initial Stratum: the Unaccusative Hypothesis In general, nominal arguments of verbs are assigned initial GRs in (hope- fully) non-controversial fashion according to how many of them there are. A single argument is assumed to be a subject. If a verb has two nominal arguments one is assumed to be its subject and the other its direct object. Which is which may present some difficulties which we shall sidestep for the moment. You will note that this assumption in- volves a strong bias toward subjects: the subject `slot', so to speak, must be filled first. There are various reasons for this, some better than others, which we shall discuss when we get to the `Final 1 Law'. But fairly early in the development of RG it was discovered that certain data presented problems with the assumption that nominal arguments could be routinely assigned GRs in a sequence consistently beginning with the sub- ject. [There's an interesting essay surveying the history of this investi- gation, along with some discussion of questions of scientific priority, in G. K. Pullum's book The Great Eskimo Vocabulary Hoax. Unfortunately, at the moment i can't find my copy in order to get the exact reference. If anybody else can, please pass it on.] Some of this evidence turns on so-called `pseudo-passives' of VPs headed by verbs collocating with pre- positions. Cf. (1-4): (1) a. Charles II slept in that bed. b. That bed was slept in by Charles II. (2) a. A camel stepped on Ralph's package of Camel cigarettes. b. Ralph's package of Camel cigarettes was stepped on by a camel. (3) a. The contestants ran under the bridge. b. The bridge was run under by the contestants. (4) a. The children were jumping on the bed. b. The bed was being jumped on by the children. Not all such verbs, however, are passivizable, even when their superfi- cial syntax seems similar. Witness (5-7): (5) a. Something silly happened in that room. b. *That room was happened in by something silly. (6) a. The ice cubes melted in the oven. b. *The oven was melted in by the ice cubes. (7) a. Little Red Riding Hood disappeared in the forest. b. *The forest was disappeared in by Little Red Riding Hood. (8) a. A troll exists under that bridge. b. *That bridge is existed under by a troll. {Those of you familiar with the history of transformational-generative grammar will remember how the difference in syntactic behaviour between the superficially similar predicates `easy/eager to please' prompted some interesting theoretical developments with regard to abstract levels of structure.} RG's solution to this seeming paradox is as follows. First of all, the post-verbal phrases in the (a) sentences in (1-8) are assumed to be true PPs bearing in the initial stratum oblique relations to the verbs. The `pseudo-passive' is then assumed to be made possible by a reinterpreta- tion of the preposition as part of the verb, as is the case with such relexified, idiomatic verbs as `put up with' meaning `tolerate' or `turn on' meaning `arouse'. The oblique nominal that is the preposition's complement is consequently raised to the status of a direct object, from which it can then advance to subjecthood in a passive-like revaluation. Thus, in each `active' sentence there is only one nominal that could bear a Term relation. In (1)-(4), that bare nominal is assumed to be a 1 (sub- ject) in the initial stratum, which can be bumped into chomage by the subsequent advance of the (derived) 2 to subject status. But in (5)-(8), it was hypothesized, the bare nominal is assumed to be an initial 2 which is presumed to advance to 1 in a subsequent stratum, in compliance with the `Final 1 Law' which requires that the final stratum include a 1-arc. {Since in these clauses as in those in (1)-(4) the PP-complement is pre- sumed to rise to 2, we have here examples of clauses in each of which two different nominals bear the 2-relation at different stages of the deriva- tion. Note that this is quite incompatible with the corollaries of REST's theta-criterion. Note also that the `Chomeur Advancement Ban' (which condemns a chomeur in a given derivation to remain a chomeur throughout the rest of the derivation) imposes an essential order on these revaluations. If the raising of the PP-complement to 2 were to occur first, it would result in the chomage of the initial 2, which would then be unable to advance to 1.} The unacceptability of the `passive' versions of (5)-(8) is seen as evidence for a universal constraint on revaluations, the `1-Advancement Exclusiveness Law', the gist of which is that, in the derivation of any given clause, advancement of an object to subject status may happen at most once. But the 1-Advancement Exclusiveness Law and its repercussions are not what i want to focus on here. Here i am more interested in discussing the notion that a unique bare nominal, a unique Term, is not necessarily a subject in the initial stratum but may be an object instead. This idea is generally called the `Unaccusative Hypothesis'. Clauses like those in (1-4), in which there is a 1 but no 2, are called *unergative*; clauses like with a 2 but no 1 (in the initial stratum) are called *unaccusative*. Obviously, both are intransitive, since (in the relevant stratum) they fail to have simultaneously both a 1 and a 2. The most basic assertion of the Unaccusative Hypothesis is that intransitive clauses are not all alike, but rather can be subdivided into unergative and unaccusative clauses. The distinction to depend totally on the nature of the (verb or) predicate; some predicates can take (initial) subjects but no objects while others can only (initially) take objects but no subjects. {The easiest way i've found to keeping straight which label goes with which type of clause is to remember that, in the garden-variety language with overt ergative case, a subject can get ergative marking only in the simultaneous presence of a direct object. Thus, a subject without a direct object is *not ergative*. Yeah, I know, it doesn't work very well for me either. To add to the confusion, the distinction that lies at the root of the Unaccusative Hypothesis has been gradually recognized in other frameworks; unfortunately, what in RG is called an *unaccusative* clause is e.g. in REST called an *ergative* clause. Sometimes it helps to write all this stuff down and pin it on one's bulletin board. Along with the sequence of definitions in lecturette #15.} I would now draw your attention to a semantic distinction between the unergative verbs in (1-4) and the unaccusative verbs in (5-8). The unergative verbs -- sleep, step, run, jump -- are all to some extent agentive, that is, they predicate some behaviour on the part of their arguments. {Whether this is as true of `sleep' as it is of the others is open to question. However, it can be argued that sleep is something one *does*, that one can *choose to do*.} The unaccusative verbs -- happen, melt, disappear, exist -- differ in that their arguments are more patient-like. Just as an example, *melting* is obviously not so much something ice *does* as something that *happens to* it, an experience it *undergoes*. And there are reasonable, if not always absolutely solid, semantic grounds for viewing the other unaccusative verbs similarly. So, in distinguishing *syntactically* between unergative and unaccusative verbs, are we merely distinguishing between different *semantic* classes of verbs? Indeed, precisely this apparent congruence between semantic and syntactic facts was among the principal original motivations of the Universal Alignment Hypothesis discussed in the previous lecturette. But Carol Rosen [`The Interface between Semantic Roles and Initial Grammati- cal Relations', Perlmutter & Rosen, eds., Studies in Relational Grammar II (1984), pp. 38-80.] demonstrated that, attractive as the congruence is and useful as it is as a first approximation, it doesn't hold as perfect- ly as the UAH would require, thereby undermining the UAH as a reliable theory about the laying of initial strata. Apart from the acceptability or lack thereof of the `pseudo-passive' construction, there is little if anything in English grammar that turns reliably upon the unergative/unaccusative distinction. In the Romance languages such as Italian and French on which Rosen has lavished much of her research energies, this is not the case; the unergative/unaccusative distinction is quite robust in these languages. Most importantly, auxi- liary selection is sensitive to it. Unergative verbs pattern with tran- sitives in selecting *habere as their auxiliary for compound tenses, while unaccusatives collocate with *essere. {I'm giving here the Latin originals of the auxiliaries, since what is said here for Italian is equally true of French and, as far as I know, all Romance languages.} Given the great importance of compound tenses in these languages, espe- cially in colloquial usage, and the relevance of auxiliary selection for various other issues such as verb-agreement, the unergative/unaccusative distinction is therefore an essential part of the grammar of these lan- guages. Rosen demonstrates that it is not entirely predictable on the basis of semantics. First she demonstrates that, although in general agentive-type intransi- tives in Italian such as `nuotare' (swim), `viaggiare' (travel), and `scherzare' (joke) are unergatives and take `avere' as their auxiliary, `fuggire' (flee) as an intransitive takes `essere', as shown in (9) (9) a. Aldo ha fuggito ogni tentazione. `Aldo has fled every temptation.' b. Alda `e fuggito. `Aldo fled.' (lit., `Aldo is fled') Contrariwise, while non-agentive intransitives such as `cadere' (fall) or `morire' (die) typically take `essere' as their auxiliary, `deviare' (go awry, be misdirected) in its intransitive usage takes `avere'; cf. (10). (10) a. Bertini ha deviato il colpo. `Bertini deflected the blow.' b. Il colpo ha deviato. `The blow went awry.' (lit., `The blow has gone awry') Nor are `fuggire' and `deviare' simply anomalies. Rosen indicates that they are representative of a small but respectable number of Italian verbs whose auxiliary selection is contrary to what the Universal Alignment Hypothesis, in conjunction with the Unaccusative Hypothesis, would pre- dict. Of the two, Rosen argues that the Unaccusative Hypothesis is the better substantiated, especially in Romance as noted above. Therefore, it is impossible to define any simple identification between semantic roles and initial GRs that holds strong predictive power in the Romance languages; hence the Universal Alignment Hypothesis cannot be truly uni- versal. Rosen then nails the coffin lid firmly closed by demonstrating that a se- mantic predicate that one language treats as an unergative verb is treat- ed by another as unaccusative, or vice versa. Just as the unergative/ unaccusative distinction is critical in Romance for auxiliary selection, so in Choctaw it is critical for the selection of subject-agreement mark- ers. (To oversimplify somewhat, Choctaw verbs agree with Nuclear Term relations on the basis of their GRs in the initial stratum; hence unaccu- sative subjects pattern with transitive objects while unergative subjects pattern with transitive subjects.) Rosen shows that some intransitive predicates, such as `die', are unaccustive in Italian but unergative in Choctaw, while some others, such as `sweat', are unergative in Italian but unaccusative in Choctaw. After surveying several such examples, she concludes that it is unfeasible to establish some completely reliable semantic ground for either the assignment in any given language or for the cross-linguistic discrepancy. Thus, there can be no `universal alignment' between semantic roles and initial GRs. The upshot of all this is that, as indicated at the end of the previous lecturette, the `Universal Alignment Hypothesis' is (1) useful as a guide to a first-order approximation of the organization of the initial stratum but (2) not reliable on the fine details thereof, and certainly not able to bear much in the weight of argument, i.e., don't try too hard to build a theoretical argument using it as a basis. In the process of discussing this, I have introduced the Unaccusative Hypothesis, which is *very impor- tant*, not only in RG but in other frameworks as well; as already men- tioned, the understanding behind it has been recognized by a wide range of syntactic theorists as generally valid, and as a result it has been adopted in some form or other into other frameworks. I have also, in the course of this lecturette, mentioned some other `Laws' of RG (e.g., the Final 1 Law, the Chomeur Advancement Ban), about which I'm going to have to talk at greater length in the near future.