Wednesday 13 April 2011

Unaccusative Effects in Vietnamese and English: Further evidence for Inner Aspect

[This is a slightly edited version of a paper submitted to a volume out of the 2007 Forces in Grammatical Theory Conference (Paris). The paper brings together material from two recently published articles, a NELS paper from 2005, and an article recently published in an 2011 OUP volume edited by Rafaella Folli and Christiane Ulbrich Interfaces in Linguistics. Considerably more work is necessary to satisfactorily unify these ideas—in particular, the second section needs reworking to convince the sceptics, but it's a start. All comments welcomed!]

Inadvertent Causes and the Unergative-Unaccusative Split in Vietnamese and English

This paper draws together several strands of evidence in support of the claim that two kinds of cause relations are independently represented in phrase-structure. The first of these kinds is the familiar, intentional/volitional cause associated with the thematic relation AGENT, typically represented in the current generative literature as the argument licensed by ‘little v’: in recent years, it has once again become commonplace to assume that this intentional CAUSE is abstractly represented in phrase structure, either as a primitive predicate, or as a relational notion: see Hale and Keyser (1993); Baker (1997); also Pustejovsky (1991), Tenny and Pustejovsky (2000). This paper however focuses on the structural representation of the second type of cause: a less studied relation that that I’ll term INADVERTENT CAUSE (IC), and which—in contrast to its more robust cousin—has generally escaped detailed scrutiny until quite recently.2 The analysis presented here develops a proposal originally articulated by (Travis 2010, 2000, 1991), which associates the IC thematic relation with the specifier position of a VP-internal functional category, namely, Inner Aspect (IAspP). Travis’ proposal is originally motivated by facts from a completely different range of (Western Malayo-Polynesian) languages: to the extent that it extends naturally to the phenomena discussed here, the present work provides confirmation of the profitability of a syntactic approach to inadvertent cause.
Travis’ proposal incorporates two distinct empirical claims. The first of these is that the long-standing bi-partite division of subject arguments—as (underlying) Agents or Themes—should be reinterpreted as a tri-partite division, whereby some cause arguments are merged to an intermediate functional specifier position, situated between ‘little v’ and and the core thematic verb-phrase [Spec, VP]). The second, associated, claim is that this functional projection involves aspectual features of some kind, which predicts that its syntactic behavior should co-vary with the aspectual type of the root predicate. This paper pursues both of these claims independently: section 2 below offers evidence from Vietnamese causative constructions in support of an intermediate specifier position for IC arguments, while section 3 provides evidence from English participial constructions concerning the aspectual nature of this intermediate projection.

1 Preliminaries
Before presenting new data, it is worth drawing attention to some cases that have already received attention in the generative literature, and which draw out the distinction between the more familiar intentional/volitional cause and the IC relation: in each case, the distinction is syntactic in so far as one or other thematic relation is unavailable in a particular structural configuration.
Consider first the Binding contrasts in (1) and (2), which privilege DP subjects bearing the IC relation over agentive subjects. As discussed by (Pesetsky 1995, Harley 1995, and Fujita 1996 amongst others, backwards binding is permitted in the (a) examples, where the subject expresses inadvertent cause, but blocked in the (b) examples, where the surface subject anaphor must be interpreted agentively. Assuming with these previous authors that Binding contrasts are to be explained configurationally—that is, in terms of c-command—such contrasts offer prima facie evidence that inadvertent causes are initially merged prior to the merger of intentional causes (Agents), and that experiencer DPs intervene between the two positions (at whatever point in the derivation Principle A applies).
1. a. ?Each other’s remarks made Bill and Mary laugh.
    b. *Each other’s friends (intentionally/deliberately) made Bill and Mary laugh.
2. a. ?Each other’s pictures annoyed Sue and Mary.
    b. *Each other’s friends (intentionally) annoyed Sue and Mary.
As Fujita and others have pointed out, the paradigm in (1) and (2) extends to the double object constructions shown in (3), where binding facts once again suggest a lower underlying position for inadvertent causes than for Agents. The interesting twist here is that in English more generally—that is to say, in non-binding contexts—intentional agents show a wider distribution than inadvertent causes: evidence for this is offered by the fact that prepositional datives disallow IC subjects. To see this, compare the sentences in (3) and (4):
3. a. ?Each other’s pictures gave Bill and Mary (an idea for) a book.
    b. *Each other’s friends (intentionally) gave Bill and Mary a book.

4. a. Interviewing Nixon gave Mailer a book/a headache.
    b. *Interviewing Nixon gave a book/a headache to Mailer.
Postponing the question of precisely how these English contrasts should be analyzed, let us turn to a syntactic reflex of this thematic distinction in another language variety. The case in question is Travis’ (1994, 2000, 2010) treatment of the verbal prefix (ma)ha in Malagasy. Travis observes that the addition of this aspectual morpheme to certain initially intransitive predicates has a causativizing function. Compare (5a) vs. (5b) below:
5. a. Tsara ny trano. [Malagasy, from Travis (2000)]
        beautiful the house
        ‘The house is beautiful.’

b.  Maha-tsara ny trano ny voninkano.
     PRES.a.ha.beautiful the house the flowers
     ‘The flowers make the house beautiful.’ (literally, ‘..beautified the house’)
c. *Maha-tsara ny trano Rabe.
      PRES.a.ha.beautiful the house Rabe
      ‘Rabe make the house beautiful.’
Though apparently similar to other inchoative-causative alternations, the crucial difference here is that only IC arguments are licensed: the contrast between the acceptable (5b) and the unacceptable examples in (5c) shows that (ma)ha is incompatible with agentive cause arguments. Two points should be borne in mind. The first is that there is no general syntactic constraint blocking the introduction of agentive arguments in Malagasy—this constraint is specific to a particular transitivizing prefix. Second, the IC relation is linked here to a prefix that is properly treated as expressing an aspectual function: with other predicates, the addition of the same prefix serves to change the (Vendlerian) aspectual class of the base predicate, converting activities into achievements, as shown by the alternations in (6). In other words, there is a direct association in Malagasy between the IC thematic relation and a particular kind of aspectual semantic function, which Travis glosses as ‘[+telic]’:
6. a. mijery 'to look at' ~ mahajery 'to notice' [Travis (2000)(Phillips 2001)]
b. mandinika 'to examine' ~ mahadinika 'to remark'
In Travis’ analysis, this association is cashed out syntactically: ha is analyzed as heading a VP-internal Aspect projection, with the argument interpreted as the clausal subject in (5b) initially merged as the specifier of this head. This is schematized in (7):

7.


Two aspects of this analysis are especially relevant here: first, the articulation of a VP-internal argument position intermediate between that associated with prototypical Agents and that of Themes, thus presenting a three-way contrast in transitivity in place of the standard dichotomy; second, the explicit association of this intermediate position with aspectual features that are logically independent of thematic relations. With this in mind, let us turn to the new data. In the next section I present some new evidence from Vietnamese consistent with this intermediate specifier position: just as in Malagasy, Vietnamese causatives exhibit a thematic restriction that systematically excludes embedded subjects from being interpreted agentively. The section that follows then presents new data from English bearing on the syntactic relationship between transitivity and Inner Aspect: on the one hand, these data clearly support the idea that such a relationship exists; on the other, they challenge the view assumed by almost all commentators—see also also Folli and Harley (2005), Schäfer (2008)—that telicity is the key aspectual property underlying such alternations.

2 Vietnamese Causatives
In this section, the concern is with constraints on causativization in Vietnamese. In line with other morphologically isolating languages, Vietnamese has no synthetic causatives. Instead, causativization is invariably expressed periphrastically: the introduction of an additional subject argument (DP1) must be licensed by a higher causative predicate (V1) làm (which otherwise functions as a lexical light verb meaning ‘do, make’).
The most immediately significant fact about simple làm causatives in Vietnamese is their incompatibility with (already) transitive or clearly unergative V2s: so, for example, làm cannot be added to a base predicate to derive the equivalent of ‘John made [the girls help him]’ or ‘John made [the child sing]’, as shown by the unacceptability of the examples in (8) below. Instead, làm combines exclusively with monovalent predicates whose arguments are non-agentive, as is the case for the core unaccusatives exemplified in (9) and (10) below. Notice that with such predicates the (apparently) inverted order [DP1 V1 V2 DP2] is clearly preferred over the canonical [DP V] order, though both orders are grammatically acceptable:
8. a. ?*Tôi làm [đứa con gái giúp anh ấy].
            I make cls. cls. girl help prn dem2
            ‘I make the girl help him.’
b. ?*Tôi làm [đứa con gái nhảy/hát/ngủ].
        I make cls. cls. girl dance/sing/sleep
        ‘I make the girl dance/sing/sleep.’

9. a. Tôi làm gẫy cái que.
        I make break cls stick
       ‘I broke the stick.’
b. Tôi làm rách tờ giấy.
     I make torn cls paper
    ‘I tore the sheet of paper.’

10. a. (?)Tôi làm cái que gẫy.
             I make cls. stick break
            ‘I broke the stick.’

 b. (?)Tôi làm tờ giấy rách.
         I make sheet paper torn
        ‘I tore the sheet of paper.’
A less obvious interpretive fact about the examples in (9) and (10) is that grammatically acceptable làm causatives receive by default an indirect interpretation: that is to say, the matrix subject is normally interpreted as the inadvertent cause of the event: a ‘salient participant’, rather than an Agent. Notice that in contrast to the English periphrastic causative, this does not imply that the subject has any less involvement in the core event, only that there is less intentionality on the subject’s part. For this reason, a better translation of (10a), for example, might be through the ‘Ethical Dative' construction: ‘The stick broke on me.’ The interpretive parallels with the Malagasy paradigm in (5) above should be clear.
Matters become interesting when one considers the paradigm more closely: in particular, when one considers V2 predicates that are neither the excluded ‘core unergatives’ in (8), nor the core unaccusatives in (9), which show apparently inverted [DP1 V1 V2 DP2] word-order. The structural representation in (7), with causative làm projected under V1, directly predicts the existence of an intermediate set of grammatical V2 predicates: those whose sole argument is involved but non-volitional (in other words IC); these should obligatorily appear following làm but preceding V2. The sentences in (11)-(14) directly bear out this prediction: the examples in (11) show that predicates that are typically classed as unergative, but which—in contrast to example (8b) above—are uncontrolled, may be causativized; the examples in (12) show that even typically agentive predicates such as those in (8b) may be causativized if it is clear that the action is non-volitional/uncontrolled by the participant—compare the ungrammatical (8b) with that in (12), where ‘girl’ is replaced by ‘puppet’. The examples in (13) demonstrate that inverted order is strongly dispreferred in all of these cases, in clear contrast to the core unaccusative examples in (9) and (10):
11. a. Tôi làm đứa con trai khóc. ([DP V1 DP V2])
          I make cls. cls. male cry
        ‘I made the boy cry.’

b. Tôi làm đứa con trai cười.
      I make cls. cls. male laugh
     ‘I made the boy laugh.’

12. ?Tôi làm con búp-bê nhảy/hát.
         I make cls puppet dance/sing
        ‘I make the puppet dance/sing.’

13. a. *Tôi làm khóc đứa con trai. ([*DP V1 V2 DP])
           I make cry CLS. CLS. male
          ‘I made the boy cry.’

b. *Tôi làm cười đứa con trai.
       I make laugh cls cls. male
       ‘I made the boy laugh.’

c. *Tôi làm nhảy/hát con búp-bê.
      I make dance/sing cls. puppet
     ‘I make the puppet dance/sing.’
Pursuing the issue further, the examples in (14) below reveal a similar split among the set of predicates normally classed as unaccusative: even though—as we saw in (9) above—inversion is strongly preferred in cases where the causee undergoes a radical and permanent change of state, it is dispreferred where the causee is merely the ‘involved participant’ in the event:
14. a. Tôi làm (?ngã) thang-be (ngã).
           I make (fall ) boy (fall)
           ‘I made the boy fall.’

b. Tôi làm (?biến-mất) thang-be (biến-mất).
     I make (disappear) boy (disappear)
     ‘I made the boy disappear.’
This paradigm demonstrates that the traditional dichotomy between unergative and unaccusative predicates is unhelpful, at least for Vietnamese: moreover—given the contrasts between examples (8) and (12)—it suggests that the terms ‘unergative’ and ‘unaccusative’ cannot denote inherent and immutable properties of lexical stems/roots, but must refer instead to different patterns of syntactic projection, with thematic relations being read off different Specifier-Head configurations, as proposed by Hale and Keyser (1993), Baker (1997) amongst others.
Travis’ tripartite structure, by contrast, captures the paradigm extremely well: the interpretation of arguments projected to the intermediate [Spec, Asp] position as ICs directly explains the structural conflation—or rather, unification—of “more unaccusative unergatives” (12) with “more unergative unaccusatives” (14), both being realized as preverbal DP2s. Furthermore, if one assumes that làm is projected under V1 within a monoclausal structure, the structure in (15) explains not only the preverbal vs. postverbal distribution of DP2 arguments—as participants (ICs) vs. Themes—but also the impossibility of transitive/unergative complements *(8): làm cannot take a uncontrolled unergative predicate as a complement since the sole argument of such a predicate (prior to causativization) is normally projected to a position above that of làm, i.e., to [Spec,VP1]: that is, it competes for the initial merge position of the clausal subject:



Thus, these Vietnamese data seem to offer clear, independent support for the first part of Travis’ proposal, namely, that IC arguments are projected to an intermediate specifier position below that of intentional Causers/Agents but above that of Themes.
For evidence in support of the second part of the proposal—viz., that this thematic position is linked to a projection with aspectual properties—we turn now to English, this time to a previously unremarked constraint on adjective formation with English present participles.

3 Dancing Girls and Flying Squirrels: Constraints on Adjective Formation
There is a long tradition of work, dating back at least to Jespersen (1940), that documents and/or attempts to explain an asymmetry in the distribution of English past participles, such that typically unergative participles are prevented from functioning as pre-nominal modifiers: see, for example, Jespersen (1940), Lakoff (1965/1970), Bresnan (1982, 1985, 2001, 1982), Levin & Rappaport (1986), Langacker (1991), Haspelmath (1993), Ackerman and Goldberg (1996). This constraint is illustrated by the contrasts in (16):
16. a. the frozen river/ a fallen leaf/ a broken spoke
      b. *the run man/*a coughed patient/*a swum contestant
In previous work, I have sought to draw attention to a more subtle interpretive contrast among present participles running in the opposite direction. In Duffield (2005), it is claimed that typically unergative present participles are able to form adjectives with dispositional (property/atemporal) readings, and thus may enter into lexical compounds, whereas typically unaccusative participles may not, being forced to retain their verbal (or temporally-bound) status. Before articulating the analysis of this constraint with reference to the Inner Aspect projection, let us consider some relevant data, beginning with the sentences in (17) and (18):
17. a. She wants to buy a burning candle. [*DR/okBR ]
      b. They didn’t want to have a crying baby. [okDR/okBR]

18. a. He found the burning candle. [*DR/okBR/okIB]
      b. They found the crying baby. [okDR/okBR/okIB].6
The observational claim is that the (a) and (b) examples in (17) and (18) crucially differ with respect to TEMPORAL ANCHORING, in the sense of Klein (1994, 1998, 2006): whereas the temporal value of unaccusative predicates such as burning is obligatorily linked to some Topic Time in the immediate discourse, unergative participles such as crying may also be interpreted dispositionally, as properties, temporally independent of any Topic Time. In this respect, crying is ambiguous in a way that burning is not: for (17a) to be true the candle must be burning at the time of purchase, but this is not the case in (17b), where the baby must only have the habit of crying more than is usual for babies. To better appreciate this contrast, compare the examples in (19) and (20):
19. a. I'd like to buy a *melting/soft cheese. (cf. a cheese that melts easily).
      b. Don't buy lenses with *breaking glass; only buy specially toughenecd glass, or plastic ones. (cf. brittle glass, also “breaking saddle” — see below)
      c. Do you have *burning material in that waste-paper basket? (cf. flammable material)

20. a. I'd like to buy a rocking chair, a whistling kettle.
      b. Hire non-singing (i.e., instrumental) bands for your event.
      c. Do you have any chatting room-mates in your house?
Example (19a) is perfectly acceptable with a temporally-bound reading; that is, if it is my wish to purchase a cheese that is melting at the time. What this example does not mean is that it is my wish is to buy a type of cheese, in whatever state at time of purchase, that has the property of melting easily: Raclette, as it might be, as opposed to Monterey Jack. Likewise, were it acceptable, breaking glass could refer to those types of glass that break easily—compare the acceptable pre-nominal adjectives fragile or brittle, or the equally acceptable post-nominal relative. Again, (19b) and (19c) are fully acceptable with a temporally-bound reading: although it may be strange to buy a product that is breaking at the time of purchase, (19d) is perfectly acceptable if the speaker sees smoke emanating from the waste-paper basket. No similar constraint applies to the examples in (20).
A point to stress is that the failure of predicates of this type to form dispositional adjectives is not due to pragmatics: this is shown by the fact that for every instance in (19) where the dispositional reading for an unaccusative participle is blocked, an acceptable paraphrase or equivalent bare adjective is available.
The interpretive difference between core unergative vs. core unaccusative participles is reflected in contrasting patterns of lexicalization (or perhaps is reflective of such patterns, depending on the grammatical theory one assumes). For present purposes, a participle is operationally defined as lexicalized just in case:
  • (i) it has an entry as an adjectival participle (ppl.a.) in the OED (online edition) that is independent of the entry for the verb stem; 
  • (ii) at least one sub-entry is not listed as obsolete; 
  • (iii) at least one sub-entry can be directly paraphrased by a relative clause (...that X’s).
A survey of the 68 monovalent predicates examined in Sorace (2000), plus a number of others, reveals that unergative participles with dispositional readings are lexicalized significantly more often than unaccusatives.9 The distinction is not absolute, since there are isolated collocations with unaccusative predicates ('BE' predicates, in Sorace’s terms): these include Dying God, Falling Leaf, and Burning Bush. Nevertheless, the distribution is heavily skewed in favor of HAVE participles.
Furthermore, most of the cited collocations with BE participles involve the ‘other’ thematic relationships discussed in more detail below: whereas a weeping ash is one that figuratively weeps, a wilting coefficient doesn't itself wilt, nor does a descending letter descend (rather, part of the letter descends below a fixed height).
Finally, listed collocations formed from BE participles tend to be of very low frequency and restricted to specific registers; Falling Leaf is a good example of this, referring as it does to a particular aerobatic trick. By contrast, collocations formed from HAVE participles show up in a much wider range of registers and have markedly higher token frequencies.
Notice once again that there is no pragmatic or logical reason why many of these present
participles should not allow a dispositional reading. In principle, for example, one could have coined the term sinking ship for submarine, or subsisting farmer instead of subsistence farmer; persisting headaches might compete with persistent headaches, and so forth, yet the former term of each pair only admits the temporally-bound reading.
Also, even where unaccusative participles are listed as a sub-entry of the verb, their interpretation is invariably temporally-bound (verbal), rather than dispositional (adjectival). This is illustrated by the examples in (21):
21. a. 1848 MACAULAY Hist. Eng. I. 182 Indications of a coming storm.
b. 1848 MILL Pol. Econ. III. xxiv. §3 The speculative holders are unwilling to sell in a falling market.
c. 1876 FREEMAN Norm. Conq. IV. 73 Norwich, with its newly rising castle, was put under his special care.
d. 1884 Century Mag. Jan. 356/2 Wilting flowers are hardly appropriate to a steamship.
e. 1704 RAY in Lett. Lit. Men (Camden) 206, I look upon my self as a dying man.
f. 1853 R. S. SURTEES Sponge's Sp. Tour xli. (1893) 217 The staying guests could not do much for the good things set out.
g. 1859 MILL Liberty i. (1865) 5 The still subsisting habit of looking on the government as representing an opposite interest to the public.
h. 1980 G. M. FRASER Mr American II. xvii. 322 Mr Asquith...would find himself out of office, and the ticking bomb of Ireland could be hastily passed to his successor.
As just mentioned, unaccusatives are not absolutely barred from forming dispositional adjectives. The constraint is more subtle: viz., unaccusatives cannot form dispositional adjectives that are transparent in terms of their thematic relations: Whereas unergative participles are typically interpreted as bearing the same thematic relationship to the modified head noun as the base verb does to its sole argument (X-ing Y = Y that X’s), the head nouns in collocations involving unaccusative participles are either interpreted as instrumental arguments, or as arguments bearing some ‘other’ thematic relation, as in (22) below; alternatively, as in (23) and (24), they are coerced into ‘inadvertent cause’ readings with a separate object (implicit in (23), explicit in (24). In all cases, the directly corresponding inchoative reading is blocked.
22. a. I'd like to get a melting iron/knife. (= an iron used for melting sth.)
      b. He drove her to breaking point. (= point at which s.o. breaks)
      c. The conjuror performed the usual vanishing tricks. (the trick doesn’t
vanish).
     d. America is the Melting Pot of cultures. (the pot doesn’t melt)

23. a. sinking-verbal ships (= temporally-bound = ships that are themselves sinking)
      b. ?sinking-adjectival ships (= dispositional = ships that cause others to sink: e.g., battleships, not submarines)
In (22a), a melting iron is not one that itself melts, but one that serves to melt something else; similarly, in (22b)-(22c) it is not the point that breaks or trick that vanishes.
Alternatively, unaccusative participles can evade the thematic restriction through overt causativization, that is to say, by incorporating a Theme nominal into the derived adjective. This process is illustrated in (24):
24. heart-breaking stories/mind-bending drugs/bulb-growing countries
Aside from usage statistics, the split between unergative and unaccusative participles is reinforced by three other kinds of distributional evidence. First, where a dispositional reading is intended the participle in unergative A-N collocations attracts lexical (compound) stress : as the contrasts in (25) and (26) show, this is not available to the few unaccusative participle-N collocations listed in the OED.
25. a. 'Rocky the Flying Squirrel' wasn't in fact a Flying Squirrel.
      b. Those dancing girls aren't dancing girls: the dancing girls are sitting over there!
      c. Don't confuse that running back with the running back: they're different players (in different sports).
26. a. The Falling Leaf is not a falling leaf; it's an aerobatic stunt.
      b. A blooming letter is not the same thing as a blooming (‘bloomin’) letter.
      c. In some cases, it’s not the staying horse that wins, but the staying horse.
       d. On one side of the parapet was a disappearing gun; on the other, a Disappearing gun, which happened not to be disappearing that day.
The exclusive ability of more unergative participles to enter into lexical compounding and thus to attract lexical stress is directly reflected in an obvious distributional difference: viz., only (dispositional) unergatives can appear to the right of prenominal adjectives denoting nationality, which is normally taken to be that class of adjectives positioned closest to the N head: see Sproat and Shih (1991), Cinque (1994, 2005), amongst others). The examples in (29) show that certain unergative participles can in fact appear twice in the same phrase: to the left of the nationality adjective with a temporally-bound reading; to the right with a dispositional reading (attracting compound stress):
27. a. The falling British [ inflation-rate/?*The British [falling inflation-rate
      b. The rising Japanese [ yen/?*The Japanese [ rising yen
      c. The disappearing Vulgarian [ diplomats/?*The Vulgarian [disappearing diplomats.

28. a.  The singing English nuns (BR only)/The English singing nuns (DR only)
      b. *The weeping Irish willow/The Irish weeping willow
29. a. The Canadian running back/The running Canadian back
       b. The running Canadian running back
The examples in (30) and (31) below highlight an additional difference between unergative and unaccusative participles, namely, that unergatives, in contrast to unaccusatives, show no contradiction under sentential negation. This follows from their ambiguous status: the (verbal) temporal reading is not in conflict with the adjectival property reading:
30. a. #This burning candle isn't burning (now).
      b. #He watched a burning candle, but it wasn’t burning that night.
      c. #He waited for an arriving plane that never arrived.

31. a. Those crying children aren't crying (now).
      b. He watched the Singing Nuns, but they weren’t singing that night.
      c. This Snapping Turtle isn't snapping (at the moment).
Summarizing the discussion thus far, various kinds of evidence show that more unergative present participles may function as prenominal adjectives—and thus enter nominal compounds—whereas more unaccusative ones may not (unless they incorporate a separate Theme argument). Prima facie, this constraint is puzzling, not least because, as was noted earlier, adjectival past participles show precisely the opposite constraint: compare again the examples in (16) above, repeated here for convenience:
16. a. the frozen river/ a fallen leaf/ a broken spoke
      b. *the run man/*a coughed patient/*a swum contestant
Under the traditional dichotomous view of the unergative-unaccusative distinction, this contrast between present and past participles is paradoxical since, if the explanation for the effects discussed here is a structural/thematic one—as I assume—it cannot simply be the standard explanation for adjectival past participles run backwards, that is, ‘before passivization’. This is because, as Haspelmath (1993) points out, most structural/thematic approaches account for the contrast between (16a) and (16b) above by claiming that only THEME arguments—alternatively, only the underlying objects of telic predicates—are accessible for this type of modification, with the sole arguments of unergatives being either of the wrong lexical type or projected too high in the thematic structure. If this explanation carried over to active participles, we would expect to see either the same thematic restrictions applying here—that is, incorrectly excluding unergatives—or conceivably no restriction, with unpassivized unergatives remaining low enough to be accessible for modification. Thus, the solution to the present participles problem also forces a reconsideration of previous analyses of the past participle alternation in (16).
By contrast, Travis’ tri-partite phrase-structure proposal in (15), supplemented by a number of additional assumptions, provides a solution to the opposing restrictions on both present and past participles that resolves this paradox whilst simultaneously tying the IC thematic relation directly to a syntactic position that is aspectual (in the most obvious sense of hosting aspectual morphology).
A basic assumption underlying the present analysis is that that the interpretive ambiguity between temporally-bound vs. dispositional readings for prenominal participles stems directly from a categorial structural ambiguity between pre-nominal verbal participles and prenominal bare adjectives. In the case of the (unrestricted) temporally bound reading, I assume that pre-nominal participles project exactly the same verbal structure as they do in predicative position; by contrast, the dispositional reading arises whenever participles are converted to and projected as bare adjectives (where this is permitted). Notice that I assume that this conversion process is a syntactic one, albeit covert and within the lexicon: in other words, it is a piece of l-syntax, in the sense of Hale & Keyser (1993).
To a first approximation, let us assume, following Reuland (1983), that the representation of verbal present participles involves a functional head containing the formal features of the –ing affix, as well as all the phrase structure governed by this head. Reuland’s original structure is diagrammed in (32a) below: for present purposes, ‘Infl’ may be re-interpreted as corresponding to Travis’ Outer Aspect (OAsp) projection, as in (32b).



As for the projection of bare attributive adjectives, I follow Higginbotham (1985), in which it is claimed that the representation of adjectival modifiers involve at an open argument position with which the modified head noun must be identified, as shown in (33)—see Higginbotham (1985: example [45]). Note that in contrast to (32), participles realized as bare adjectives project no functional structure. The claim is that unergative participles functioning as pre-nominal modifiers are structurally ambiguous between these two modes of projection (s-syntactic vs. l-syntactic projections, respectively):

Of course, much hangs here on the correct interpretation of the anachronistically labeled ‘VP’ node, an issue addressed directly below. Nevertheless, if the more general assumption is correct, then the right analytic question is why some participles may undergo conversion to adjectives, while others cannot. The answer I suggest is that the sole arguments of active unergatives are in the right structural position to undergo l-syntactic conversion, whereas those of “core unaccusatives” are not.
With the previous discussion of Vietnamese causatives in mind, assume now that a three-way distinction in argument projection is likewise implicated in the unergative-unaccusative split in English, such that in verbal projections:
  • (i) arguments of V1 are merged to [Spec,V1], interpreted as AGENTS
  • (ii) arguments of (Inner) Asp are merged to [Spec, Asp], interpreted as INADVERTENT CAUSE/KEY PARTICIPANTS
  • (iii) arguments of the root verb are merged as sisters to V2, interpreted as THEMES
As was the case in Vietnamese, suppose that for core unaccusatives, only the third type of projection is possible: by contrast, for “unergatives”, either of the former modes of projection are available: they may initially project either as canonical transitives (option 1) or as Inadvertent Causes (option 2), as in (34) cf. (15) above:

Given this structural representation, the various constraints on participial adjectives observed above follow fairly directly if one assumes the following lexical constraint on participle to adjective conversion:
35. Unique Mapping Constraint on Adjective Formation
The argument mapped to the argument position of the adjective template must be projected into the [Spec, Asp’] position of the participle at the point of lexical conversion.
Combining this constraint with the phrase-structure in (33) has the following consequences. In the first place we can immediately derive the core unaccusative-unergative contrasts in (17-20) above: unless there is some reason to raise ‘Theme’ arguments through [Spec, Asp]—such arguments will not occur in this position in l-syntax, and adjective formation will be blocked by the UMC. This is schematized in (36):
The two exceptional contexts where arguments of unaccusative predicates may be converted to adjectives are precisely those where there is evidence of some kind of argument raising: either the regular case of passivization—this is the case of adjectival past participles in (16)—or the (apparently more lexical) instances of Theme-incorporation, as in (24) above.
Taking the passive participles first, a possible derivation is given in (37) below. As should be clear, this is no more than a modification of the standard generative analysis of passive (Baker, Johnson, and Roberts 1989), such that the abstract passive/perfective -EN morpheme is initially associated with Inner Aspect rather than with V1/‘little v’, as is more commonly assumed:

With this revision, Baker, Johnson, and Roberts (1989)’s analysis of verbal passives extends to the analysis of prenominal participles: unergative predicates (16b) are correctly excluded—the ‘external’ theta-role normally being assigned to the argument in [Spec, ASP] is assigned to the passive morpheme, and thus unable to license any ‘external’ argument, also blocking the adjectival conversion; by contrast, unaccusative predicates are permitted in both their verbal and adjectival forms in virtue of covert raising to/through [Spec, Asp] in l-syntax. This verbal vs. adjectival contrast is exemplified in (38) and (39), respectively, where once again the two forms are disambiguated by their distribution with respect to nationality adjectives.
38. a. the frozen Norwegian lakes
      b. the burnt French toast
      c. the broken Japanese videocamera
39. a. the Norwegian frozen yogurt
      b. the French burnt ochre is superior to the Spanish tint
      c. the Japanese broken hearts club
Notice that this analysis reconciles two traditionally diametrically opposed perspectives on passivization: passive as a lexical vs. syntactic operation. On this analysis, passivization is lexical in the sense that it takes place in the lexicon, early enough to feed adjective conversion; at the same time, however, it is syntactic in the sense that it operates over the same structures and disposes of the same mechanisms that constitute the overt syntax of more isolating languages, such as Vietnamese (as we saw in section 2 above).
Turning to the examples illustrated in (24) above involving Theme incorporation, these are now reanalyzed as instances of overt (l-syntactic) raising, as in (40) (‘mind-bending drugs’):


Finally, this analysis explains why it is that those marginal cases where unaccusative predicates may function as prenominal adjectives involve transitivization of the predicate such that the head noun is interpreted as an IC of the process denoted by that predicate, rather than as the Theme: that is, for example, why a breaking saddle is one used for breaking (in) horses, not one that breaks:


Summarizing this section: the distribution and interpretation of English prenominal participles—especially, the opposing constraints applying to present vs. perfective participles—offers direct supports for the tri-partite structural representation proposed by Travis (op cit.), insofar as the availability in English of an intermediate specifier position within the extended VP offers a straightforward explanation for what would otherwise be a paradoxical flip in the possibilities for prenominal modification, depending on the aspectual properties of the predicate. The additional value of the English data is that the licensing properties of this projection are directly linked to an uncontroversially aspectual element, namely, the perfective morpheme –EN. Thus, the English data provide support not only for the existence of such a position, but also the justification of its syntactic label (Inner Aspect).
Before closing this section, one feature of the analysis requires some additional comment. Previously, in setting out Travis’ analysis of the Malagasy prefix ha-, it was noted that this morpheme was taken to be an exponent of the semantic feature [+telic]; subsequently, it was also observed in passing that other researchers of this have concluded that telicity is the determining feature of relevant alternations of this type.
The data presented here, however, cast doubt on this conclusion (assuming of course that the phenomena are to be handled by a parallel treatment). This is because the contrast analyzed here involves an ‘Anti-telicity’ effect: lexically, the participles that permit adjective formation are unergative, a class of predicate normally assumed to be atelic; by contrast, unaccusative predicates, which are generally classed as telic, resist adjective formation (unless passivized).
The force of this evidence is compounded by the fact that experiencer predicates display a similar contrast with respect to adjective formation: participles formed from Object Experiencer verbs, such as annoy, exciting, interest, and so forth, permit dispositional readings (42), whereas Subject Experiencer participles are grammatically unacceptable (43ab), unless—like unaccusatives—they are passivized or contain incorporated Theme arguments (43cd):
42. a. an interesting fact
      b. an amusing story
      c. a frightening incident
      d. an entertaining stunt

43. a. She is a *(god-)fearing woman (cf. fearful)
      b. *He was an envying man (cf. envious)
      c. *Loathing or hating people should be avoided if possible.
      d. an *(all)-knowing God
As discussed at greater length in Duffield (2005, esp. in prep), this contrast is not fully parallel to the unaccusative vs. unergative split outlined above, in that object experiencer modifiers are still restricted to the verbal (left-hand) side of nationality adjectives, and bare subject experiencer predicates are excluded entirely (by whatever constraint prevents stative predicates from combining with -ing in English). Nevertheless, the fact that two subclasses of lexically atelic predicates are implicated with the IAsP projection argues against the idea that telicity is the relevant semantic feature. Though more work is necessary to establish this, it is possible that a lexically more neutral feature, such as perfectivity, or ‘boundedness’, may be at work here (boundedness being an intrinsic feature of telicity, but not the other way around).

Conclusion
This article has focused attention on a less scrutinized causal relation, that of Inadvertent Cause (IC), and on its representation in phrase-structure. The investigations summarized here provide some independent empirical evidence in support of the specific phrasal architecture proposed in Travis (2000, 2010), involving an Inner Aspect projection below the position of intentional Cause (V1, v), suggesting that the conclusions previously arrived at on the basis of Western Malayo-Polynesian have quite general application. The data presented here also speak to the question of the unaccusative-unergative distinction, challenging the more standard assumption of an inherently lexical dichotomy between two kinds of predicate—for example, in terms of inherent telicity: see (Chierchia 1989), (Levin and Rappaport Hovav 1995), amongst others—replacing this binary with a tripartite structural account of unaccusativity, where thematic relations are read off syntactic representation; see also (Haertl 2003). Overall then, the article reaffirms the profitability of syntactic, as opposed to semantic or pragmatic, accounts of such thematic alternations: cf. Narasimhan, Di Tomaso, and Verspoor (2007).

References
Ackerman, Farrell, and Adele E. Goldberg. 1996. Constraints on adjectival past participles. In Conceptual Structure, Discourse and Language, edited by A. E. Goldberg. Stanford, CA: CSLI Publications.
Baker, Mark. 1997. Thematic Roles and Syntactic Structure. In Elements of Grammar: a Handbook in Generative Syntax, edited by L. Haegeman. Dordrecht: Kluwer Academic Publishers.
Baker, Mark, Kyle Johnson, and Ian Roberts. 1989. Passive Arguments Raised. Linguistic Inquiry 20:219-251.
Borer, Hagit. 1995. Deconstructing the construct.
Botha, Rudolf P. 1983. Morphological mechanisms. Oxford: Pergamon Press.
Bresnan, Joan. 1995. Lexicality and argument structure. Paper read at Paris Syntax and Semantics Conference, at Paris.
———. 2001. Lexical-Functional Syntax: Blackwell.
———, ed. 1982. The mental representation of grammatical relations. Cambridge, Massachusetts: MIT Press.
Chierchia, Gennaro. 1989. A semantics for unaccusatives and its syntactic consequences. ms, cited in Tenny & Pustejosky (2000).
Cinque, Guglielmo. 1994. On the evidence for partial N movement in the Romance DP. In Paths Toward Universal Grammar, edited by G. Cinque, J. Koster, J.-Y. Pollock, L. Rizzi and R. Zanuttini. Georgetown: Georgetown University Press.
———. 2005. Deriving Greenberg’s Universal 20 and Its Exceptions. Linguistic Inquiry 36:315–332.
Dowty, David. 1979. Word Meaning and Montague Grammar.
Duffield, Nigel. 2005. Flying Squirrels and Dancing Girls: Events, Inadvertent Causes and Unaccusativity in English. In Proceedings of the thirty fifth annual meeting of the North East Linguistic Society, edited by L. Bateman and C. Ussery: Booksurge Publishing.
———. in press. On Unaccusativity in Vietnamese and the Representation of Inadvertent Cause. In Researching interfaces in linguistics, edited by R. Folli and C. Ulbrich. Oxford and Cambridge: Oxford University Press.
Embick, David. 2004. On the structure of resultative participles in English. Linguistic Inquiry 35.
Folli, Rafaella, and Heidi Harley. 2005. Flavours of v: Consuming Results in Italian and English. In Aspectual Enquiries, edited by P. Kempchinsky and R. Slabakova. Dordrecht: Springer.
Fujita, Koji. 1996. Double objects, causatives and derivational economy. Linguistic Inquiry 27 (1):146-173.
Haertl, Holden. 2003. Conceptual and grammatical characteristics of argument alternations: the case of decausative verbs. Linguistics 41 (5):883-917.
Hale, Kenneth, and Samuel J Keyser. 1993. On argument structure and the lexical expression of syntactic relations. In The view from building 20: essays in linguistics in honor of Sylvain Bromberger, edited by K. Hale and S. J. Keyser. Cambridge, Massachusetts: MIT Press.
Haspelmath, Martin. 1993. Passive participles across languages. In Voice: form and function, edited by B. Fox and P. J. Hopper. Amsterdam: John Benjamins Publishing Company.
Higginbotham, James. 1985. On Semantics. Linguistic Inquiry 16:547-593.
Jespersen, Otto. 1940. A modern English grammar. Part V. Syntax. London: Allen and Unwin; Copenhagen: Munksgaard.
Klein, Wolfgang. 1994. Time in Language. London: Routledge.
———. 1998. Assertion and Finiteness. In Issues in the Theory of Language Acquisition. Essays in Honor of Jürgen Weissenborn, edited by N. Dittmar and Z. Penner. Bern: Lang.
———. 2006. On finiteness. In Semantics in acquisition, edited by V. Van Geenhoeven. Dordrecht: Springer.
Kwon, Nayoung. 2004. A semantic and syntactic analysis of Vietnamese causatives. Paper read at Western Conference on Linguistics, at UC, San Diego.
Lakoff, George. 1965/1970. On the Nature of Syntactic Irregularity. Indiana University dissertation. Published by New York, NY: Holt, Rinehart and Winston as Irregularity in Syntax.
Langacker, Ron. 1991. Foundations of cognitive grammar. Stanford, CA: Stanford University Press.
Levin, Beth, and Malka Rappaport Hovav. 1995. Unaccusativity: at the Syntax-Lexical Semantics Interface. Edited by S. J. Keyser. Vol. 26, Linguistic Inquiry Monograph Series. Cambridge, Massachusetts: MIT Press.
Levin, Beth, and Malka Rappaport. 1986. The formation of adjectival passives. Linguistic Inquiry 17:623-61.
Lieber, Rochelle. 1988. Phrasal compounds in English and the morphology-syntax interface. Paper read at Proceedings from the Parasession on Agreement in Grammatical Theory, at CLS.
———. 1992. Deconstructing Morphology: word-formation in syntactic theory. Chicago: University of Chicago Press.
McCawley, James D. 1968. Lexical insertion in a transformational grammar without deep structure. In Fourth Regional Meeting of the Chicago Linguistic Society, edited by B. J. Darden, C.-J. N. Bailey and A. Davison. Chicago: University of Chicago.
Narasimhan, Bhuvana, Vittorio Di Tomaso, and Cornelia M. Verspoor. 2007. Unaccusative or Unergative? Verbs of Manner of Motion.
Parsons, Terence. 1990. Events in the Semantics of English: a study in subatomic semantics. Cambridge, MA: MIT Press.
Pesetsky, David Michael. 1995. Zero Syntax. Cambridge, Massachusetts: MIT Press.
Phillips, Vivianne. 2001. The interactions between prefix and root: the case of maha in Malagasy. In Formal issues in Austronesian linguistics, edited by V. Phillips, I. Paul and L. Travis. Dordrecht: Kluwer Academic Publishers.
Pustejovsky, J. 1991. The syntax of event structure. Cognition 41 (1):47-81.
Reuland, Eric. 1983. Governing -ing. Linguistic Inquiry 14:101-136.
Schäfer, Florian. 2008. Two types of external argument licensing. GLOW Newsletter 31.
Sorace, Antonella. 2000. Gradients in auxiliary selection with intransitive verbs. Language 76 (4):859-890.
Spencer, Andrew. 1991. Morphological Theory. Oxford and Cambridge, MA: Basil Blackwell.
Sproat, Richard, and C Shih. 1991. The Crosslinguistic Distribution of Adjective Ordering Restrictions. In Interdisciplinary Approaches to Language: Essays in honor of S.-Y. Kuroda, edited by C. Georgopoulos and R. Ishihara. Dordrecht: Kluwer.
Tenny, Carol, and James Pustejovsky, eds. 2000. Events as grammatical objects: the converging perspectives of lexical semantics and syntax. Stanford CA: CSLI Publications.
Travis, Lisa. 1991. Inner Aspect and the structure of VP. Paper read at NELS talk, at University of Delaware.
———. 2000. Event structure in syntax. In Events as grammatical objects, edited by C. Tenny and J. Pustejovsky: CSLI Publications.
———. 2010. Inner Aspect: the articulation of VP. Vol. 80, Studies in Natural Language and Linguistic Theory. Dordrecht: Springer.
Vichit-Vadakan, Rasami. 1976. The Concept of Inadvertence in Thai Periphrastic Causative Constructions. In Syntax and Semantics, edited by M. Shibatani. New York: Academic Press.


Notes
1. Such approaches to the representation of (intentional) CAUSE resurrect certain core aspects of the Generative Semantics tradition, as represented, for example, by (Lakoff 1965/1970; McCawley 1968).  Following the demise of Generative Semantics, these ideas were taken up by semanticists, especially (Dowty 1979) and (Parsons 1990), then partially ‘re-imported’ into syntax by Pustejovsky (1991) amonst others: see Travis (2010) for a clear overview.
 2. A reviewer draws attention to more recent work on the topic, of which I was unaware, including that of Kallulli (2006), Schäfer (2009) and Solstad (2009).
 3. There are in fact two similar, but structurally distinct, analytic causative constructions in Vietnamese: ‘simple’ làm causatives on the one hand, and ‘complex’ causatives introduced by (làm) cho, on the other. The focus here is on the former type, which exhibit the thematic constraints predicted under the Inner Aspect analysis: in Duffield (in press, also in prep.), I provide evidence that the complex causatives implicate a fundamentally different, bi-clausal analysis, which explains their insensitivity to these thematic constraints (cf. (Kwon 2004)).
 4. This also holds for Thai causatives, as discussed in (Vichit-Vadakan 1976).
 5. As ever, matters are somewhat more complex than presented here. In particular, the present analysis does not directly explain the ‘indirect’ reading of Vietnamese causatives mentioned earlier. It also seems to exclude—falsely, as it turns out—làm causatives involving inadvertent DP1 subjects with ‘participant DP2 complements’ (the equivalent of ‘The wind blew the boy over’). Both of these issues are discussed at greater length in (Duffield in press).
 6. BR = (temporally) Bound Reading, DR= Dispositional (adjectival) Reading. Discussion of the contrast is complicated by the fact that specific, especially definite, determiners introduce an additional (prior) Topic Time to which the event denoted by the verbal participle may be anchored: call this the INDEPENDENTLY BOUND reading (IB). For ease of exposition, therefore, I will ignore specific interpretations/contexts: indefinite determiners should be interpreted as non-specific.
 7. For many speakers, collocations such as melting cheese are acceptable with a dispositional reading. However, the crucial point to observe is that this reading is only available with a ‘coerced causee’ reading: a melting cheese in this sense is one that can be melted, not one that is predisposed to melt (intransitively), whereas a squeaking chair is one that squeaks; see below.
 8. Similar remarks apply to She loves the sound of breaking glass vs. She loves breaking glass: whereas the former sentence may have a generic reading since breaking glass simply modifies the noun sound, the latter sentence can only be interpreted either with a temporally-bound reading ‘She loves it (at the time) when glass is breaking,’ or (much preferred) with the coerced causative/Theme reading, where breaking is reanalyzed as a transitive.
 9. The sampled set comprised the following predicates (non-italicized items from (Sorace 2000), italicized items added): come, arrive, leave, fall (non-agentive); rise, descend, ascend, become; wilt, bloom, decay, die; appear, emerge, disappear, happen, occur; stay, remain, last, survive, persist; exist, be, belong, sit, lie, seem, suffice, subsist, correspond, consist; tremble, waver, shiver, skid, weep; cough, sweat, sneeze, vomit; ring, resound, rumble, toll, tick, shin; run, roll, dance, swim; chat, work, blow, spit, snap; sleep: yield, surrender, triumph, prevail, join; break, melt, freeze, boil, burn, thaw.
10. Again, to the extent that a dispositional reading is possible for ‘sinking ship’, the available interpretation is the coerced causer reading: a destroyer, say, rather than a submarine. Note also that some speakers allow migrating bird with a dispositional reading; for others though, this participle is temporally bound; for these speakers only the alternative migratory is possible.
 11. From the TV animation series Rocky and his Friends and The Bullwinkle Show: I am grateful to David Birdsong for this example.
 12. I assume that the participial alternant, which is the only possible realisation for unaccusative predicates, projects ‘too much structure’ to permit compounding.  In other words, the restriction is another reflex of a more general constraint on compounding: the NO PHRASE CONSTRAINT of Botha (1983); see also Lieber (1988, 1992), Spencer (1991): see below.
 13. As a reviewer points out, this approach finds close similarities with work by Borer (Borer 1995), and especially with Embick (2004), a paper that I was previously unaware of.
 14. To avert any misunderstanding, note that the assumption is that the perfective morpheme is associated with the Inner Aspect projection. As for the progressive morpheme ing, however, I assume that this is associated with a VP-external functional projection, that which Reuland (1983) labels ‘Infl’, and which Travis would term ‘Outer Aspect’: see (33) above.

No comments: