Extra argumentality--affectees, landmarks, and voice *.

Hole, Daniel

Abstract

This article investigates sentences with additional core arguments of a special type in three languages, viz. German, English, and Mandarin. These additional arguments, called extra arguments in the article, form a cross-linguistically homogeneous class by virtue of their structural and semantic similarities, with so-called "raised possessors" forming just a sub-group among them. Structurally, extra arguments may not be the most deeply embedded arguments in a sentence. Semantically, their referents are felt to stand in a specific relation to the referent of the / a more deeply embedded argument. There are two major thematic relations that are instantiated by extra arguments, viz. affectees and landmarks. These thematic role notions are justified in the context of and partly in contrast to, Dowty's (1991) proto-role approach. An affeetee combines proto-agent with proto-patient properties in eventualities that are construed as involving causation. A landmark is a ground with respect to some spatial configuration denoted by the predication at hand, but a figure at the highest level of gestalt partitioning that is relevant in a clause. Thereby, both affectees and landmarks are inherently hybrid categories. The account of extra argumentality is couched in a neo-Davidsonian event semantics in the spirit of Kratzer (1996, 2003), and voice heads are assumed to introduce affectee arguments and landmark arguments right above VP.

1. The topic: extra argumentality and interparticipant relations

The sentences in (1) from Mandarin, English, and German share a number of properties. (1)

(1) a. Ta si-le muqin.

(s)he die-PRF mother

'His/her mother died on him.'

b. The ship tore one of its sails.

c. Hans schrubbte Paul den Rucken.

Hans scrubbed Paul.DAT the back

'Hans scrubbed Paul's back.'/lit. 'Hans scrubbed [sup.dat]Paul [sup.acc]the back.'

First, the italicized arguments (subjects in [1a] and [1b], an indirect object in [1c]) are unexpected if one considers the canonical argument structure of each verb, and if one considers the thematic relations typically associated with eventualities encoded by these verbs. For instance, verbs denoting events of dying are not typically construed as (noncausative) transitive verbs, as is the case in (1a).

Second, the referents of the italicized arguments are felt to stand in a special relationship to the referents of the more deeply embedded and ccommanded arguments muqin 'mother', one of its sails, and den Rucken 'the back'. The relations in (1) are kinship, part/whole, and inalienable possession. These relationships will provisionally be called "interparticipant relations" before their implementation in terms of predicate abstraction and variable identification is put forward in the second half of the article. The very same interparticipant relations that are instantiated in (1) between syntactic core functions may also occur between subjects or objects and PPs; an example from English would be She stared him in the eyes. In order to keep the empirical domain of this pilot study manageable, only predications with non-PP complements will be taken into consideration. This move should, however, be regarded as provisional.

Unexpected arguments with syntactic core functions of the above kind will henceforth be called "extra arguments," and they are invariably italicized.

For (1a) and (1b) a further characteristic frequently encountered with extra arguments holds: the extra arguments may fulfill syntactic functions that are canonically assumed by the more deeply embedded arguments. Thus, in (1a) and (1b) the extra arguments are subjects, even though from the point of view of canonical intransitive uses of the respective verbs one would expect the other argument to appear in subject position.

Sentences as in (1) have been dealt with in very different frameworks and from many different perspectives, but to the best of my knowledge, precisely that array of data that we'll be concerned with and which I take to constitute a crosslinguistically homogeneous phenomenon has never been under scrutiny. In Section 2, I will critically review results from several frameworks in which data as in (1) have been analyzed, most notably the possessor raising or ascension analyses in the tradition of relational grammar. Section 3 will provide the empirical survey of relevant sentence patterns in German, Mandarin, and English. These three languages have been chosen for exemplification for no particular reason, except that I have a sufficient command of each of them, which allows me to base my argumentation not just on the literature and work with consultants but also on intuitions of my own. In Section 4 the focus of interest shifts from the presentation of data to the semantic modeling of extra argumentality. I will introduce the implementation for "possessor" datives of German in terms of a Kratzerian voice semantics (Kratzer 1996, 2003) as put forward in Hole (2005), and we will check which portion of the data from Section 3 can be covered by this kind of analysis. The last major section will deal with the amendments that are necessary to extend the voice analysis of Section 4 to the other cases of extra argumentality. In the course of this extension a major division between predicate types will be introduced: (i) predicates instantiating the proto-agent/proto-patient scheme in the sense of Dowty (1991), which is rooted in concepts like volitionality, control, and causality, and (ii) predicates denoting locative eventualities, whose participants are mapped to figures and grounds. The figure/ground distinction is usually filtered out as thematically irrelevant in the literature, and I will reopen the discussion about this issue. As a result, the claim is put forward that the agent/patient dichotomy and the figure/ground partitioning are two ways of mental and linguistic construal which are treated on a par by grammars of natural languages.

This article is designed as a pilot study, and therefore it fails to define its empirical domain in a strictly deductionist manner. There remains something ostensive about the delimitation of the relevant data. But it seems to me to be beyond doubt that the data assembled here should be subjected to a uniform treatment. Even though we will come to call the relevant domain "voice" beginning with Section 4, the collection of necessary conditions for our voice phenomena will not reach the point where necessary-and-sufficient conditions can be stated. Still, the insights that emerge along the way render the project, I think, sufficiently interesting for the theoretician and the descriptivist alike to merit treatment in an article like this.

2. Arguments against possessor raising approaches

The most influential approach to the grammatical modeling of facts as in (1c) dates back to Perlmutter and Postal's (1983) (the paper has been circulating since 1974) seminal idea that such facts should be treated as phenomena akin to passives and raising structures. On some nonsurface level of derivation, some argument of a sentence has a syntactic position lower in the tree than in the surface version of the sentence. (See also Isacenko [1965] for a much earlier generative account in the same vein.) In the case of the passive, the direct object (or "initial 2" in relational grammar) becomes the subject. In the case of possessor raising, a possessor nominal which is base-generated as a constituent within some argument constituent (often a direct object in European languages) is raised or ascended to assume a syntactic function of its own (often the function of an indirect object in European languages). Take (2) from German as an example.

(2) a. Paul zerbrach ihr die Brille.

Paul broke her.DAT the glasses

'Paul broke her glasses (on her).'

b. Paul zerbrach ihre Brille.

Paul broke her glasses

'Paul broke her glasses.'

On the possessor ascension analysis, (2a) is derived from (some underlying version of) the more basic (2b). Specifically, the possessor of the glasses, which is expressed as a determining modifier in (2b), becomes the indirect dative object in (2a). The large number of papers treating phenomena in individual languages in terms of possessor raising or possessor ascension is proof of the strong impact that the idea of possessor raising has had over the decades, and this impact has not at all been limited to Generative (sub)paradigms (cf. to name just a few: Aissen 1987; Fox 1981; Gallmann 1992; Keenan and Ralalaoherivony 2000; Landau 1999; some of the contributions to Perlmutter 1983 or Perlmutter and Joseph 1990). (2) This holds despite Bresnan's (1983) seminal lexical-functional paper which fostered growing reservations against raising analyses in other areas. But since Bresnan did not treat extra argumentality in her paper, the ascension/raising analyses remained--on the whole--unchallenged. Apart from the partially theory-dependent arguments exchanged by generative and lexical-functional grammarians, there are empirical reasons to discard any kind of possessor ascension analysis. We will turn to them in Section 2.

Wunderlich (1996, 2000) defends a view of "possessor" datives in German which might be characterized as involving "lexical possessor raising." Even though there is no level of syntactic analysis in his approach at which the dative arguments are constituents of the possessum DP, the semantics interprets them just like DP-internal possessors. (3) is an example.

(3) a. Sie wusch ihm die Fusse.

she washed him.DAT the feet

'She washed his feet.'

b. [lambda]y[lambda]z[lambda]x[lambda]s {WASH(x,y) & POSS(z,y)}(s) (3) (Wunderlich 1996: 339)

(3b) is Wunderlich's representation of the semantics of wasch- 'wash' which has been derived in such a way that the integration of a dative argument becomes possible. After the arguments have been inserted (or their slots have been bound), the truth conditions for (3a) will then come out as follows: 'There was an event of her washing the feet, and the feet belong to him'. In typological studies of the recent past, the terminology of "external" vs. "internal" possession has gained ground (Konig and Haspelmath 1998; Payne and Barshi 1999; the terminology dates back to Vergnaud and Zubizarreta 1992, but this latter paper was written in the government and binding paradigm; more on it will be said in Section 4.3.2). Whenever a person (as opposed to a body part) or a possessor is encoded as an extra argument, this is called external possession, and whenever the possessor term and the possessum term form a single complex DP constituent, this is a case of internal possession. Inasmuch as generalizations in this terminological paradigm rest on the assignment of the possessor role to external possessors/extra arguments, the same objections as will be stated in Sections 2.3 and 2.4 can be used to criticize this analysis. While typological analyses of extra argumentality in terms of external possession are less committed syntactically than the possessor raising solutions, they have greatly enlarged our knowledge of the typical semantic differences between external possession or extra argument constructions and internal possession constructions across languages. (4,5) Let us now turn to the problems that accounts of extra argumentality in terms of syntactic or lexical possessor raising/ascension face.

2.1. The dead possessor argument (6)

If the extra argument in a sentence like (4b) were a possessor, it would be a mystery why its use renders sentences deviant if the alleged possessor is not alive anymore, since bona fide possessors in genitival constructions as in (4a) are not deviant if their referents are not alive or sentient anymore. A similar point is made in (5) for English, where the contrast arises between a sentient/conscious whole (him) and a nonsentient whole (the leg). (7)

(4) [Paul died first.]

a. Dann starb auch seine Mutter.

then died also his mother

'Then his mother died, too.'

b. #Dann starb ibm auch seine Mutter.

then died him.DAT also his mother

'Then his mother died on him, too.'

(5) She kicked him/#the leg in the shin.

The contrast in (4) is easily reproduced in Mandarin (cf. [6]), the difference being that the extra argument is a subject in Mandarin, and not a dative as in the German example. (See Section 3.1 and, in particular, example [14] for evidence that ta '(s)he' is a subject, and not just a topic, in [6b].)

(6) [Paul died first.]

a. Houlai, tade muqin ye side.

afterwards his mother also die-PRF

'Then his mother died, too.'

b. #Houlai, ta ye side muqin.

afterwards he also die-PRF mother

'Then his mother died on him, too.'

I assume that both the dative argument in the German examples and the subject argument in the Mandarin example instantiate the thematic role affectee, with affectee being a thematic relation which combines at least one proto-agent property (sentience/consciousness), and one proto-patient property (causal affectedness) in the sense of Dowty (1991) (cf. also Primus' [1999] proto-recipient). We will repeatedly return to Dowty's theory of proto-roles below; at the present point, a preliminary understanding of Dowty's notions is sufficient.

Lexical possessor raising a la Wunderlich (1996, 2000) is faced with the same kind of problem. While it may be possible to write a special condition into the semantics of the poss predicate in (3b) which takes care of the semantic restriction under discussion here, this restriction wouldn't follow from anything that we know about DP-internal possessors, thereby weakening the claim that extra-argumental participants may be reduced to possessorhood without any special amendments.

2.2. The dative passive argument

The German dative passive is formed with the passive auxiliary bekommen (or kriegen) 'get' and the past participle of a verb. Dative arguments of active sentences are encoded as subjects while the direct objects remain unaffected (Reis 1985; Leirbukt 1997). The subject of the active sentence is demoted to an optional PP. The passive sentence in (7b) exemplifies this construction by opposing it to the corresponding active sentence in (7a).

(7) a. Die Wachen offneten [[dem Angreifer].sub.DAT] das Tor.

the guards opened the.DAT attacker the gate

'The guards opened the gate for the attacker.'

b. [[er Angreifer].sub.NOM] bekam das Tor (yon den Wachen)

the attacker got the gate by the guards

geoffnet.

opened

'The guards opened the gate for the attacker.'/lit. 'The attacker got the gate opened (by the guards).'

Not just any dative going along with an active verb can become the subject of a dative passive. There is a constraint requiring the subjects of such passives to be affectees as characterized in the preceding subsection. Therefore, as witnessed by (8b), (8a) doesn't have a good dative passive counterpart.

(8) a. Die Zensur offnete [[er Beeinflussung].sub.DAT] das

the censorship opened the.DAT manipulation the

Tor.

gate

'Censorship opened the floodgates to manipulation.'

b. #[Die Beeinflussung] bekam das Tor geoffnet.

the manipulation got he gate opened

lit. 'Manipulation got the gate opened.'

The referent of die Beeinflussung 'the manipulation' is not consciously or sentiently involved in the event at hand, therefore the dative passive in (8b) is out. Now consider (9) and (10). Extra dative arguments regularly become the subjects of dative passives, and in my own dialect this is true of all extra dative arguments.

(9) a. Die Arztin hat [[Paula].sub.DAT] Blut abgenommen.

the doctor has Paula blood taken

'The doctor took some of Paula's blood.'

b. [[Paula].sub.NOM] hat (von der Arztin) Blut abgenommen

Paula has by the doctor blood taken

bekommen.

got

'The doctor took some of Paula's blood.'

(10) a. Sic streichelte [[dem Jungen].sub.DAT] den Arm.

she stroked the.DAT boy the arm

'She stroked the boy's arm.'

b. [[Der Junge].sub.NOM] bekam (von ihr) den Arm

the boy got by her the arm

gestreichelt.

stroked

'She stroked the boy's arm.'

If dative passives are only grammatical with subjects whose referents are affectees, extra arguments must be affectees, too. We will therefore have to say that affectee, but not possessor, is the correct thematic relation of the pertinent extra dative arguments in German. A parallel argument can be made for the Chinese passive promoting indirect objects or immediately postverbal extra arguments to subject function. Turn to Section 3.2 (examples [16] and [16']) for the relevant data.

Again, lexical possessor raising as defended by Wunderlich (1996, 2000) (cf. [3b]) will have essentially the same problem; possession alone does not single out a natural class for the phenomenon of German dative passives and analogous constructions in other languages and does not, therefore, seem to be the relevant notion to explicate the observed link between many extra arguments and dative passives.

2.3. Presupposition vs. entailment (8)

The relation of possession felt to hold between the referent of the extra argument and the referent of the more deeply embedded argument is presupposed. If the extra argument encoded a possessor, it would be the only kind of verbal argument that I know of which conveys presupposed information. This in itself speaks against extra dative arguments as encoding the possessor role. The affectee information conveyed by extra dative arguments in German (cf. Section 2.1) is, however, part of the assertion and, thus, entailed. (11) with a dative extra argument in a conditional clause illustrates the difference.

(11) a. Falls die Schwester dem Patienten auf den Mantel

if the nurse the.DAT patient on the coat

tritt, ...

steps

'If the nurse steps on the patient's coat ...'/

lit.: 'If the nurse steps the patient on the coat ...'

b. presupposition: 'The patient possesses a coat, and he is wearing it, is keeping it close to his body, or is locally related to it in some other way.'

c. lost entailment: 'The patient is consciously involved in an event in which he is causally affected.'

As expected in if-clauses, the presupposition persists, while the entailment is lost. On the assumption that the dative argument expresses the affectee involvement of the relevant referent, no problem arises, because the affectee involvement is just as inactive in (11) as, say, the agentive involvement of die Schwester 'the nurse'. If we can make plausible where the presupposed relation of possession has its source in (11) outside the dative DP, the situation will be a lot easier to handle. No two thematic involvements of differing information statuses would have to be expressed by a single DP. In Section 4.2 we will see how the intuition of possession can be taken care of within the more deeply embedded DP.

Wunderlich's representation of lexically derived predicates with a built-in semantics of possession as in (3b) (... WASH(x,y) & POSS(z,y) ...) does not make mention of the differing kinds of information status for the two predicates and may, for this reason, too, not be considered superior to the syntactic possessor raising analyses.

3. The range of syntactic functions of extra arguments

Extra arguments occur in all syntactic functions associated with verbs, except for the function of the most deeply embedded complement in a given syntactic structure. We thus find extra arguments that are subjects or objects. If the extra argument is a direct object, the most deeply embedded complement is sometimes a (directional) PP, and sometimes a second object. For the sake of restricting the empirical domain to be covered in this article, I disregard PP complements (and also PP extra arguments); extra arguments as objects will always be followed by a second object in this article (but see Note 9). I will look at instances of the different syntactic functions of extra arguments in the following, providing examples from each of the languages mentioned above (English, German, Mandarin), supplemented by one Korean example.

3.1. Extra arguments as subjects

Some English sentences with extra arguments in subject function are listed in (12) (cf. Rohdenburg 1974).

(12) a. The ship tore one of its sails.

b. The car burst a tire.

c. The boy grew breasts.

d. The athlete tore a muscle.

All the examples in (12) feature unaccusative verbs, and the direct objects are the arguments that would, in the absence of the extra arguments, figure as the subjects of the sentences. None of the sentences in (12) involves a causativized use of the intransitive verbs. Take (12c) for illustration. The sentence does not mean that the boy grew breasts the way farmers grow tomatoes, it rather states that it happened to the boy that breasts grew on his chest. In all of the examples, the extra argument denotes the whole of which the referent of the more deeply embedded argument constitutes a part. Analogous examples from Mandarin are provided in (13). (9)

(13) a. Ta diao-le hen duo toufa.

(s)he fall-PRF very much hair

'A lot of his/her hair fell out.'/lit. '(S)he fell very much hair.'

b. Ta duan-le tui le.

(s)hebreak-PRF leg PRT

'(S)he broke his/her leg.'

It can be shown that the extra arguments are really subjects, and not Chinese-style topics. Chinese-style topics, because of their backgrounded discourse status, cannot be focal information in a question-answer sequence. (10) This is, however, easily possible for ta '(s)he' as in (13). (14) is a felicitous question-answer pair.

(14) Q: Shei duan-le tui le?

who break-PRF leg PRT

'Who broke his leg?'

A: Ta duan-le tui le.

(s)hebreak-PRF leg PRT

'(S)he broke his/her leg.'

A second argument for the subjecthood of extra arguments in such sentences is the fact that the internal argument of the unaccusative verb surfaces as a postverbal complement. This is so because the preverbal subject function has been "snatched" by the extra argument. Chinese-style topics do not trigger the in-situ surface realization of arguments of unaccusative verbs.

3.2. Extra arguments as objects

In German, extra arguments with the function of indirect objects are the paradigm cases of so-called external possession. (15) is an example, and (16) presents a structurally similar sentence from Mandarin.

(15) Die Mutter flocht der Tochter die Haare.

the mother braided the.DAT daughter the hairs

'The mother braided her daughter's hair.'

(16) Xiao Waing chi-le wo yi-ge dangao.

little Wang eat-PRF I 1-CL cake

'Little Wang ate a cake of mine.'

(15) does not require a lot of explanation since we have seen similar examples above. The hair is part of, or possessed by, the daughter, and the verb flechten 'braid' is transitive, but not usually ditransitive. (16) from Mandarin is more of a challenge. Deviating from the pattern found with verbs of transfer or verbs of creation in other languages, the referent of the indirect object loses the cake in (16) instead of getting it, and the verb chi 'eat' denotes an event in the course of which the cake ceases to exist. It is easily shown that wo T in (16) is not a modifier of the more deeply embedded complement (see Zhang 1998a, 1998b for details concerning the relevant construction in Mandarin). Moreover, it should be noted that, quite generally, the ditransitive construction with minimal coding devices (no preposition) in Mandarin is typically found with verbs that have the referents of the indirect objects lose something, or that exempt them from something, but only rarely with verbs that have the referent of the indirect object come into the possession of something. Among the few verbs of the second kind are the most general verbs of transfer gei 'give' and song 'give as a present', which take an indirect and a direct object with the same zero marking as chi 'eat' in (16). (11) At this point, readers may be suspicious about the objecthood of the extra arguments in (15) and (16). Supporting evidence for the claim that we are really dealing with objects, and not with adjuncts or modifiers, comes from passivization. As shown in (15') and (16'), the Mandarin and the German extra argument may each be the subject of a corresponding passive sentence (cf. also Section 2.1.1 above).

(15') Die Tochter bekam die Haare geflochten.

the daughter got the hairs braided

'The daughter's hair got braided.'

(16') Wo bei chi-le yi-ge dangao.

I PASS eat-PRF 1-CL cake

'I suffered from someone eating a cake of mine.'

I have chosen to present the Mandarin case in (16) as one involving a sequence of indirect and direct object because Zhang (1998a, 1998b) uses the same terminology. It is, however, possible that the Mandarin case really patterns with the double accusative sentences from Korean as exemplified in (17) (cf. Shibatani 1994: 475).

(17) John-i Mary-lul son-ul ttayli-ess-ta.

John-NOM Mary-ACC hand-Acc hit-PST-IND

'John hit Mary on the hand.'

Mandarin has a peculiar system of preverbal objects marked by the notoriously controversial functional element ba (see Li [2001] for an elegant overview of the relevant discussion). Extra arguments are found in this position, too. To understand these data, we first have to familiarize ourselves with the way the more typical ba-sentences work, that is, those ba-sentences not involving extra arguments. Direct objects are (often obligatorily) shifted into the preverbal ba-position if other complementational material is to follow the verb, if a resultative construction is involved, or if, more generally, a highly transitive event (in the sense of Hopper and Thompson 1980) is encoded. (18) illustrates the case where either a canonical postverbal object as in (18a) may be used, or a ba-object as in (18b). As just said, in other cases speakers do not have a choice. The resulting grammaticality contrast for such a case is given in (19) (cf. Li and Thompson 1981: 407).

(18) a. Wo chi-wan-le zhei-dun fan.

I eat-up-PRY this-CL food

'I've eaten up that dish.'

b. Wo ba zhei-dun fan chi-wan-le.

I BA this-CL food eat-up-PRY

'I've eaten up that dish.'

(19) a. Wo ba bao-shi cang zai xiangzi-li.

I BA precious-stone hide at chest-in

'I hide the precious stone in a chest.'

b. *Wo cang bao-shi zai xiangzi-li.

I hide precious-stone at chest-in

intended: 'I hide the precious stone in a chest.'

Ba-sentences instantiate extra argumentality if a preverbal ba-object and a postverbal object co-occur. Examples from, or in the spirit of, Tsao (1987), who in some cases quotes Cheung (1973), are given in (20) and (21).

(20) a. Ta ba juzi bo-1e pi.

(s)he BA orange peel-PRF peel

'(S)he removed the peel of the orange.'

b. Tamen ba zhu fang-le xue.

they BA pig release-PRY blood

'They drained the pig of its blood.'

(21) a. Zhangsan ba men shang-le suo.

Zhangsan BA door put.on-vRF lock

'Zhangsan put a lock on the door.'

b. Ta ba bilu sheng-le huo.

(s)he BA fireplace ignite-PRF fire

'(S)he put on a fire in the fireplace.'

(22) Ta ba qiang ti-le yi-ge dong.

(s)he BA wall kick-PRF 1-CL hole

'(S)he kicked a hole into the wall.'

(20) is a collection of examples in which the referent of the extra argument gets diminished in the course of the encoded event. The examples in (21) have it that something is added as a part to some functional whole or location. In the case of (22) it is not immediately clear whether something is added to the wall or whether its functional integrity is destroyed by the described event. I will return to this problem in Section 5.5, but the issue will ultimately be left unsettled.

3.3. Extra arguments as objects with subject-like properties

A special kind of object extra argumentality is found with verbs of bodily sensation and some psych verbs. To get a better understanding of the domain, let us start out from a well-known quirky case fact from Icelandic. Among the Germanic languages, Icelandic is an extreme case in that it allows for non-nominative subjects with certain verbs. These non-nominative arguments control coordination reduction. An example is given in (23) (cf. Faarlund 1999).

(23) Haraldii gedjast vel ad Mariu og [[??].sub.i] bydur henni oft i Harald.DAT pleases good to Maria and invites her often in bio.

cinema

'Harald likes Maria and often invites her to the movies.'

The facts of coordination reduction clearly show the subject properties of the dative argument Haraldi. If subjects, and only subjects, control coordination reduction in Icelandic, then Haraldi in (23) must be the subject. German dative experiencers are not subjects; it is the nominative (stimulus) arguments in sentences with psych verbs that have the morphosyntactic properties of subjects, such as controlling coordination reduction or triggering agreement on the verb. In terms of word order things are different, though. The neutral relative order of nominative arguments on the one hand and dative or accusative arguments on the other in German psych verb constructions is as in (24). This means that in psych verb constructions and with many verbs of bodily sensation, the nominative argument does not have its canonical position to the left of the accusative or dative argument, but rather to its right. Thereby, accusative or dative experiencers (with psych verbs) or affectees (with verbs of bodily sensation) conform to the (tendential) subject property of being the first argument in a clause (we will turn to the spurious difference between affectees and experiencers in Section 4.4). (25) is an example, and (25') is the same example with a different word order, viz. NOM > ACC; the marked order of this sentence is the unmarked order of the canonical verb classes (cf. Lenerz 1977; Hohle 1982).

(24) DAT/ACC > NOM

(25) Einem Zeugenist ein Hund aufgefallen.

a.DAT witness is a.NOM dog (be.)noticed

'A witness noticed a dog.'

(25') Ein Hund ist einem Zeugen aufgefallen.

a.NOM dog is a.DAT witness (be.)noticed

'A witness noticed a dog.'

Without going into the details of the arguments from the relevant literature, let us just note one piece of evidence supporting the claim of a special neutral word order with such verbs. Typically a focus accent on the most deeply embedded argument of a sentence with canonical word order and an unaccusative verb will allow for an interpretation as an all-new utterance in German even if none of the words without focus accents are contextually given. This effect does hold for (25) if we make the assumption that the nominative is the most deeply embedded argument. If, on the other hand, we put a focus accent on the dative in (25'), the only possible reading is a narrow focus on einem Zeugen 'a.DAT witness'. (12) With German verbs denoting sensations or transformations on or in body parts we get both things: (i) extra-argumentality and (ii) noncanonical neutral word order with the expected all-new utterance readings with stress on the nominative argument. (26) lists some examples. (13)

(26) a. Ihm juckt die Kopfhaut.

him.DAT itches the.NOM scalp

'His scalp itches.'

b. Ihr druckt der Magen.

her.DAT presses the.NOM stomach

'Her belly hurts.'

c. Ihm bricht das Herz.

him.DAT breaks the.NOM heart

'His heart is breaking.'

The German equivalent of The boy grew breasts (cf. [12c]) belongs here, too. I separate it from the examples in (26) because those sentences all involve changes or perceptions relating to body parts that already exist. The breasts in (27), however, come into being in the course of the event described by the verb.

(27) Dem Jungen wuchsen Bruste.

the.DAT boy grew breasts.NOM

'The boy grew breasts.'

With inanimate referents, the acceptability of dative extra arguments with subject-like properties varies among speakers of German. Referee B classifies the sentences in (28) as "perfect," whereas I prefer the versions in (29).

(28) a. Dem Baum wuchsen neue Aste.

the.PaT tree grew new branches

'The tree grew new branches.'

b. Dem Wagen platzte ein Reifen.

the.DAT car burst a tire

'The car burst a tire.'

(29) a. An dem Baum wuchsen neue Aste.

on the.DAT tree grew new branches

'The tree grew new branches.'

b. An dem Wagen platzte ein Reifen.

on the.PaT car burst a tire

'The car burst a tire.'

I admit that sentences as in (28) do occur and that they do not necessarily signal metaphorical animacy. Still, a contrast persists in that PPs as in (29) are always possible with inanimate referents in this construction; if, however, PPs are used instead of the dative arguments of (26) and (27), the deviant sentences in (26') and (27') are the result.

(26') a. *An ihm juckt die Kopfhaut.

on him.DAT itches the.NOM scalp

intended: 'His scalp itches.'

b. *In ihr druckt der Magen.

in her.DAT presses the.NOM stomach

intended: 'Her belly hurts.'

c. *In ihm bricht das Herz.

in him.DAT breaks the.NOM heart

intended: 'His heart is breaking.'

(27') *An dem Jungen wuchsen Bruste.

on the.DAT boy grew breasts.NOM

intended: 'The boy grew breasts.'

The referents of the prepositional complements in (26') and (27') are clearly construed as inanimate.

This concludes the survey of extra arguments with a syntactic core function in German, English, and Mandarin.

4. Interparticipant relations

4.1. Modeling interparticipant relations in terms of voice and some notion of binding

It has become clear in Section 2 that extra arguments will have to be assigned thematic relations that are dependent on the asserted events of the sentences in which they occur. This means that, in terms of thematic relations, Paul in Sic zerbrauch Paul die Brille 'She broke [sup.dat]Paul's glasses' is not the possessor of the glasses, but the affectee in an event of breaking. This, however, does not cover the whole range of native speakers' intuitions. The typical intuitions are that Paul is (indirectly) affected, and that it is Paul's glasses that are broken. Even if raising cannot be the answer, we somehow have to model the interparticipant relation felt to hold between the referent of the extra argument (Paul) and of that of the more deeply embedded argument (the glasses).

I propose that the semantics of possession or, more generally, of relationality of the more deeply embedded argument in extra argument constructions should be modeled in terms of voice and some notion of "binding." The extra argument "binds" the unsaturated variable of the c-commanded relational noun within the DP that denotes the possessum or (body) part of the referent of the extra argument. (14) Implementing this idea in an explicit syntax-semantics framework is not trivial, though. If we do not want to contend ourselves with a purely syntactic notion of binding, we will have to say where quantification enters the picture. Extra arguments may be, and typically are, nonquantificational definite DPs, but semantic binding requires a quantifier. In the next subsection, I will sketch the implementation proposed for German "possessor datives" as laid out in Hole (2005). The other subsections of Section 4 are devoted to a comparison of the proposal advocated here with other proposals. We will then move on to state the adjustments that are needed to cover the wider empirical domain of this article (Section 5).

4.2. The implementation for German dative extra arguments (a.k.a. "possessor datives")

The account proposed in Hole (2005) is couched in a neo-Davidsonian event semantics in the spirit of Kratzer (1996, 2003). In Kratzer's framework, only the internal argument(s) of a verb is (/are) required by the valency of the verb. Agent arguments that surface as subjects of transitive predications, for instance, are merged outside VP, they do not correspond to an argument position pre-specified by the verb, and their thematic role is likewise introduced independently of the verb; an agentive voice head (Kratzer 1996) performs this task. A special rule of composition, event identification, makes sure that the AGENT argument introduced in the specifier of the voice phrase denotes a participant in the event that is characterized by the VP, and not in some other event (see below for more detailed discussion).

The voice phrase implementation for "possessor datives" proceeds along similar lines. An affectee voice head right above VP introduces an affectee argument into the structure (for the characterization of the affectee role, see Section 2.1). By way of a combination of an abstraction rule and variable identification--a more general version of Kratzer's event identification--the identity of the possessor referent of the more deeply embedded argument and of the affectee referent is ensured alongside the identity of the two event variables involved. The combined effect of predicate abstraction and variable identification may be seen as a specific kind of binding.

We will use the bracketed consituent in (30) for illustration (the same sample computation is presented in much greater detail in Hole 2005).

(30) Sie will [dem Jungen [vp den Kopf streicheln]].

she wants the.DAT boy the head stroke

'She wants to stroke the boy's head.'

lit. 'She wants to stroke [sup.dat]the boy [sup.acc]the head.'

One possible interpretation of the VP is given in (31). The 'g(5)'-part of the denotation will be explained in a moment.

(31) [[den Kopf streicheln]] = [lambda]e.stroke g(5)'s head(e)

In accordance with Kratzer's theory, the denotation of the VP in (31) has no unsaturated argument position except for the one of the event argument. With existential binding the denotation in (31) may be paraphrased as 'There is an event of stroking the boy's head', provided the assignment function g maps the index 5 to the boy. Note that, quite standardly, Kopf 'head' in (31) is analyzed as a relational noun which brings along an argument slot for a possessor even in the absence of an overt possessor DP. An implicit argument with the index 5 fills this argument slot in our example, and the argument with this index is mapped to the boy, but other index numbers would also be good possibilities as long as they are mapped to individuals which are in the universe of discourse and have a head. The VP then merges with the affectee head. The function of this head is spelled out in (32).

(32) Function of [Aff.sup.0]:

a. Denotation: [[[Aff.sup.0]]] = [lambda]x[lambda]e.x is an affectee in e

b. Abstraction:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

c. Variable identification: f([Aff.sup.0] g(VP) [??] h(Aff') <e,<s,t>> (15) <e,<s,t>> <e,<s,t>>

The denotation of [Aff.sup.0] in (32a) states that an affectee as characterized in Section 2.1 is involved in the event as a sentient/conscious and causally affected participant. This argument slot will be filled by the dative argument, which enters the computation in SpecAffP. But before this happens, the abstraction rule (32b) applies. What it does is turn the VP denotation, which only had one unsaturated argument position for the event argument, into a denotation with another unsaturated argument position in the position where the argument with the index n is located. Let's say n is 5 (recall that the boy bears the index 5 in our example). (33) states the denotation of the VP before predicate abstraction (the assignment function maps the index 5 to the boy), and (34) provides the denotation after predicate abstraction (the slot of the possessor of the head is "reopened" by modifying the assignment function: whatever bears the index 5 is mapped to x). The outcome of combining the VP denotation of (34) with [Aff.sup.0] (cf. [32a]) is (35).

(33) [[den Kopf streicheln]] = [lambda]e.stroke g(5)'s head(e)

(34) [[den Kopf streicheln]][5[right arrow]x] = [lambda]x[lambda]e.stroke x's head(e)

(35) [[[Aff.sup.0] den Kopf streicheln]][5[right arrow]x] = [lambda]x[lambda]e.stroke x's head(e) & affectee(x)(e)

(35) with existential closure of the open argument slots will then be something like 'There is an event e and an individual x such that x's head is stroked in e and x is an affectee in e'. If then the argument dem Jungen 'the.DAT boy' is merged in the next step, we get the truth conditions: 'There is an event e in which the boy's head is stroked and the boy is an affectee in e', again with existential closure of the event variable. For these truth conditions to come about, (32c), which I haven't discussed yet, is indispensable. What it does in short is to ensure that the affectee referent is identical to the possessor referent of the head, and that the event in which x is an affectee is the same event in which x's head is stroked. It does so by mere stipulation. Worse still, it does so by means of a special kind of composition rule, namely, variable identification. This rule states that if two constituents whose denotations each have an unsaturated position for (i) an individual and (ii) an eventuality are to combine, then the resulting denotation will identify the two variables of the same ontological type. In other words, the resulting function will not have four unsaturated argument slots--two for individuals and two for events--but only two: one for an individual, and one for an event. The move of introducing such a rule of composition into the system may seem absurd at first, because it is a theoretically costly implementation of something highly basic. I still think that it is fully warranted within the general framework adopted here. Let us recall the general idea of neoDavidsonian event decomposition which underlies my whole proposal. If we find it attractive to model (i) verb denotations as predicates of events and (ii) argument structure (and adverbial adjunctions) as the conjunction of separate predicates which typically take just one argument apart from the event argument (cf. Kratzer 1996, 2003) and which all contribute to the overall specification of the eventuality at hand, then we need a mechanism which explicitly states that the conjoined predicates of events all specify one and the same event. Even if we may not need such a special mechanism of unification at other levels of description (in psychology, for instance, the counterpart of event identification might be derivative of something else), we do need it in formal semantics because there is no well-formedness condition in the formalism which only allows for conjoined predicates if they specify the same event. Kratzer (1996: 122), who has introduced event identification into the framework, says nothing about the balance of necessity and costs of this rule of composition, she just states it as in (36), and she needs it to tie the AGENT argument to the event denoted by the VP function, whose denotation is represented as g in (36).</p> <pre> (36) Event identification (Kratzer 1996: 122) f

g [??] h <e,<s,t>>

<s,t> <e,<s,t>> (denotation and

(denotation (event (denotation semantic type of

and identification) and the agentive voice semantic

semantic head [Voice.sup.0]) type of VP)

type

of voice') </pre> <p>This rule is different from variable identification in (32c), which is repeated in (37) for convenience.</p>

<pre> (37) Variable identification (Hole 2005) f

g [??] h <e,<s,t>> <e,<s,t>>

<e,<s,t>> (denotation and (denotation and (variable (denotation semantic type semantic type

identification) and of the affectee of VP after

semantic head [Aff.sup.0]) predicate

type of Aff') abstraction) </pre> <p>Event identification takes a two-place function and a one-place function as input and delivers a two-place function. Variable identification takes two two-place functions as input and delivers a two-place function as output. In a sense, variable identification is simpler than event identification, because both variables of f undergo a parallel identification process in the case of variable identification, but not in the case of event identification. From a less formal viewpoint, we may perhaps say that the further integration of arguments into the specification of an event is simpler to the extent that the additional argument has as much as possible "to do" with the VP denotation. In our realm, "having something to do" is nothing but the fact that the affectee referent plays a role already in the denotation of VP. In the example sentence (32) (Sic will dem Jungen den Kopf streicheln 'She wants to stroke [sup.dat][the boy]'s head') the affectee is also the possessor of the head. Other double involvements are conceivable, and in Hole (2005) I argue that so-called beneficiary datives should really be analyzed as affectees which "bind" a variable in a purposive constituent inside VP (with "binding" again taken in the sense of the combined effect of an abstraction rule as in [32b] and variable identification as in [32c]/[37]).

In Section 5 we will check how much mileage we can get out of this proposal and what modifications are necessary to cover all the examples that we have surveyed in the empirical first half of the article. Before doing this, I will devote Sections 4.3.1 and 4.3.2 to justifying why I have chosen the modeling in (32) and why I think that other solutions won't do.

4.3.1. Possessor raising. Possessor raising approaches to extra argumentality as defined in this article have already been evaluated in Section 2. The conclusion for both syntactic and lexical possessor raising was that raising accounts must be discarded because of the facts in (38) (the bits of text in parentheses state how the voice analysis defended here evades each problem).

(38) i. Possessor raising cannot (easily) explain why the thematic requirements of raised possessors differ from those of DP-internal possessors.

(Here: The thematic requirements are different because the two positions are not directly linked by movement or a lexical process.)

ii. Possessor raising cannot explain why raised alleged Possessors behave like affectees with respect to the dative passive and similar constructions.

(Here: They behave like affectees because they are affectees, and nothing else.)

iii. Raised possessors would combine presuppositional information about possession and entailed or asserted information about affectedness in a single DP. (Here: The presuppositional information about possession comes from the DP-internal (possibly implicit) possessor, and this information is expected to be presuppositional. The entailed/asserted information about affectedness is tied to the affectee argument.)

To these three arguments I would like to add a fourth one which hasn't been given in Section 2 because I lack relevant evidence for languages other than German. Still, since in my eyes it is a strong argument, I do not want to leave it unmentioned. It rests on the interpretive behavior of allegedly raised possessors in certain constructions involving ellipsis. (39) is a pertinent example; the elided constituent is crossed out.

(39) Du sollst [die Haare schneiden] wie ihm you should the hair cut like him.DAT

[[begin strikethrough]die Haare scheiden[end strikethrough]]

the hair cut

'You should cut the hair the way you cut his.'

(lit. 'You should cut the hair like dathim.')

This is a natural sentence to say if a hair dresser instructs her apprentice in the presence of two customers that are referred to without being addressed (say, because the interlocutors stand behind the customers). The hair referred to by the first object DP die Haare 'the hair' is definite because it is contextually given. The utterance of ihm 'him.DAT' in the second half of the sentence is probably accompanied by a pointing gesture. Now, if ihm were a raised possessor, there should be a trace in the elided constituent. On the assumption that elided constituents and the parallel pronounced ones are identical in their structures, the first bracketed VP in (39) would have to contain a trace as well. This cannot be the case, because there is no argument in the first half of (39) which could bind the trace. If the first VP doesn't contain a trace, then the elided one doesn't, either. But this means that ihm cannot have moved out of the elided VP. This argument is fatal for all theories defending possessor raising that put an identity constraint on ellipsis and require traces to be bound/governed by the moved item. (39) is also fatal for lexical "possessor raising" as proposed by Wunderlich (1996, 2000). Even if ihm 'him.DAT' in (39) is not moved in the syntax, but is inserted into the syntactic structure because the specially derived verb in the elided VP projects one more argument, the identity constraint on ellipsis would still predict that the overt VP in (39) should have the same argument structure. As witnessed by the first half of the sentence, though, no dative argument is projected by the pronounced VP. Therefore, the elided VP cannot project a dative argument either.

Example (39) constitutes a really hard nut to crack for all the competing theories that I am aware of in our domain. How does the proposal defended here escape this problem? The important thing to note is that the two VPs as such are fully identical. Only when [Aff.sup.0] right above VP is merged in the second half of the sentence does predicate abstraction apply, and the possessor slot of Haare 'hairs' in the elided VP is reopened by modifying the assignment function. Up to this point (as perceived in a bottom-up perspective), the denotation of the elided VP and the pronounced VP are fully identical. Specifically, the index of the (doubly) implicit possessor of the hair in the second half may, up to the VP level, be identical to the index of the possessor of the hair in the first half. The abstraction rule (cf. [32b]) annuls this index.

4.3.2. Vergnaud and Zubizarreta (1992). Except for two small hints in Section 2 and Note 14, I have so far remained silent about the relationship between the present proposal and its most similar predecessor, Vergnaud and Zubizarreta's (1992) (henceforth, V&Z) analysis of their inalienable constructions in French and English as in Le medecin leur a radiographie l'estomae (lit.) 'The doctor x-rayed them the stomach'. V&Z's treatment stands in the tradition of Kayne (1975) and Gurron (1985) and is conceptually related to Borer and Grodzinsky's (1986) in that (some kind of) binding is identified as the relevant dependency relation holding between indirect objects and (implicit) possessors of the pertinent structures. Although at first blush this general line of argumentation is similar to mine and the approach developed here might seem to be just a more specific implementation of V&Z's, the exact relationship between the two proposals is surprisingly difficult to assess. This is so for the following reasons. First, and quite generally, V&Z present a government and binding implementation, and it is not evident what their analysis would look like in Minimalist terms. Second, V&Z do not give an explicit account of the semantics that they assume, and no denotations of syntactic nodes are given at all, except for the (rather unusual) denotations of inalienably possessed nominal predicates like 'throat' as gorge(x), as opposed to the alienable counterpart gorge (V&Z 1992: 597). (16) Third, V&Z make recourse to a type/token ambiguity between definite DPs that is crucial for their analysis and which arises because of what they identify as two homophonous variants of the definite article in French. I do not assume such an ambiguity to have a crucial status in the explanation of the facts that I am discussing and, what is more, I cannot make much sense of it because it is unclear to me what the result is when a verbal denotation combines with an object denotation that denotes a type (i.e. a French doctor doesn't examine a type when he examines somebody's throat, but V&Z must be read so as to propose precisely this). (17)

4.4. Summary on extra arguments denoting affectees (and experiencers)

Let us start the summary of this larger section on interparticipant relations by recapitulating the gist of the voice analysis for so-called "possessor datives." We may say that it is just an illusion if we classify extra arguments as affectees and possessors at the same time. Affectedness comes into play in the functional voice phrase where the affectee argument is first merged. Predicate abstraction and variable identification (cf. [32]) make sure that the referent denoted by the affectee argument is identical to the referent of a possessor inside a more deeply embedded argument. The relationship between the affectee argument and the more deeply embedded possessor must, by virtue of the involved tree geometry and the involved compositional procedures, be a c-command relation. This allows us to speak of a kind of binding relationship between the affectee argument and the possessor argument. Affecteehood was defined as sentience/consciousness (proto-agent property) plus causal affectedness (proto-patient property) above, and the same ingredients seem also sufficient for experiencers. Experiencers are simply affectees in eventualities without any other participants that are more prototypically agentive. This reduction of experiencerhood to 'affecteehood plus highest degree of agentivity in an eventuality' significantly broadens the coverage of the voice proposal made in this section. (18) For the cases which cannot be handled yet, turn to the following section.

5. Extra argument structures that necessitate amendments

5.1. The problem with inanimate and nonagentive extra arguments

Inanimate extra arguments as in (40) (= [lb]/[12a]) and animate arguments which are not consciously or sentiently involved in an eventuality as in reading B of (41) pose a challenge for the account defended in Section 4.

(40) The ship tore one of its sails.

(41) Dem Sanger ist die Jacke zu eng.

the.DAT singer is the jacket too tight

reading A: 'The singer finds that his jacket is too tight.'

reading B: '(It is objectively the case, or someone other than the

singer finds that) the singer's jacket is too tight.'

The problem centers around the thematic role of extra arguments like the ship in (40), or dem Sanger 'the.DAT singer' in (41). In (40) the ship can't be an affectee as defined in this article, because ships are neither sentient nor conscious, that is, they don't have the proto-agent properties criterial for affectees, and there is no requirement whatsoever for (40) to be understood in a metaphorical sense such that the ship would have to stand for an animate being. In (41), similarly, the singer is animate, but reading B completely abstracts away from this fact. The only repercussion of the ontological animacy of the singer (as opposed to his nonagentive involvement in the reported state of affairs) is its encoding as a dative (as opposed to a PP, which would be required for an inanimate referent; such PP extra arguments have been neglected in this article for the practical reason of restricting the factors to be controlled; cf. the beginning of Section 3). I will return to an example like (41) below and concentrate on (40) for the moment. From among Dowty's (1991: 572) proto-agent properties, the ship in (40) only qualifies as "exist[ing] independently of the event named by the verb," and Dowty includes this property in the list of role entailments only tentatively. (19) This hesitation is the starting point for the argumentation to follow. Primus (2004: 2) adds "possession of another entity" to the list of proto-agent entailments, and one may argue that, in some sense of the word, the ship in (40) possesses its sail. It is, however, unclear whether the addition of possessorhood to the list of proto-agent entailments is really necessary in the light of the idea which was first introduced into general linguistics from Indo-European studies by Benveniste (1966) and which is becoming increasingly popular (cf. among many others Den Dikken 1997 or Meinunger forthcoming), namely, that possession boils down to (the control of) co-occurrence (in space) of the possessed item with another entity. Put differently, Dowty's tentative property of independent existence would, together with his corresponding proto-patient entailment of "dependen[cy] on another participant [...]," allow for the formulation of possession in these related terms of independence and dependence. But what is the gain if Dowty's tentative properties of independence of, or dependence on, other participants in a situation are included in the definition of proto-agents and protopatients? The answer is that recalcitrant stative eventualities as described by A house has walls and a roof may be viewed as instantiating the protoagent/proto-patient scheme. If only this move wouldn't put such a strain on our intuitions! In my eyes, a house as a whole which has walls and a roof simply isn't agentive at all, no matter how much we stretch the notion of agentivity. No other of Dowty's proto-agentive entailments is detached from common sense intuitions in a comparable way. The reason why he includes independent existence into his list is that this move allows him to do away with the figure/ground distinction in the theory of thematic roles. The following digression on locative predications and gestalt properties critically reviews the filtering out of gestalt roles as possible thematic roles. Readers who have no problem with the assumption of a ground (or landmark) role at least for some predications may skip the following subsection. The mainstream in linguistics seems to favor the idea that gestalt roles have no place in a theory of thematic roles, and therefore the digression appears to be justified and necessary.

5.2. Digression on gestalt roles

The idea that gestalt features of eventualities might qualify for inclusion into the set of thematic roles pops up again and again in the literature, most notably perhaps in Gruber's (1965) and Talmy's (1985, 2000: 339-341) work. The typical reaction to such considerations is that, while considered attractive at first, the assumption of thematic roles like figure or ground is judged superfluous, and that gestalt is relevant at a level of language description different from the theory of thematic roles (cf., e.g., Dowty 1991:563-564 or Dfirscheid 1999: 254). Dowty's arguments in this domain are endorsed by other researchers, and one gets the impression that Dowry has once and for all times discredited the idea of gestalt notions as thematic roles (but cf. Talmy 2000:339-341 or Maienborn 2001: 192). I would like to take a fresh look at the issue, and the result will be that Dowty's argument is not valid and that we indeed need something like the ground role (termed "landmark" below), at least for stative locative predications, and possibly for all locative predications.

By "locative predication" I refer to predications describing configurations in space that conform to the criteria in (42).

(42) Criteria for locative predications:

a. The predication asserts or entails a specific spatial configuration between at least two entities in space.

b. The position in space of one entity is identified relative to the other entity.

c. The predication may only be assigned a truth value if the position of the entity with respect to which the other entity is situated in space is identifiable (i.e. the position of this entity must be presupposed).

Apart from the differences between conflation patterns for the different notional components of locative predications that are identified in Talmy's (1985) seminal paper, Talmy also posits a major cross-cutting dividing line between stative locative predications and others that involve movement of the referent-to-be-situated. '[BE.sub.L]' is Talmy's term for stationary existence in space, which stands on a par with the pure 'MOVE' component of locative predications involving movement. Locative predications involving movement may further be subdivided into those where the moving entity has agentive control over its movement, and others where it does not. The same split holds mutatis mutandis for BEL predications such that the enduring existence of the stationary spatial configuration may require agentive control or not (cf., e.g., the control ambiguity of verbs like stand if used with animates as opposed to inanimates: Paul couldn't stand any longer and fell to the ground vs. ??? The statue couldn't stand any longer and fell to the ground). In order to keep the present argument manageable I will disregard those locative predications that involve movement. We will, for the time being, only look at stative locative predications which may or may not involve the sentient/conscious involvement of the ground referent.

Dowty's argument for rejecting a ground role for the locative which denotes the place with respect to which something is located by a locative predication rests on the identification of the figure/ground distinction with the given/new distinction relevant for information structure. He observes that, depending on the discourse in which it is used, the lamp in The lamp is over the table may preferably be given, but that it may also be new or focal information in an appropriate discourse environment. A thematic role, however, should be constant across different discourse embeddings of a single sentence. The problem with this argument is the identification of locative gestalt features and information-structural properties and gestalt. While it is possible and fruitful to talk about new information in a discourse in terms of the figure notion, this is an application of the gestalt concept which is independent of the gestalt properties of locative eventualities as such. The information status of a referent within a spatial configuration may change with the embedding discourse, but in a sentence like The apple is on the table the locative gestalt properties are fixed; the apple is the (locative) figure and the (surface of the) table is the (locative) ground. (To appreciate this fact, one must keep in mind that for a DP to bear a focus accent and be [part of] focal information does not mean 'denoting a referent whose position in space has so far not been in the discourse background'.) A sentence like The lamp is over the table instantiates the same fixed figure/ground scheme, even though at first it may seem that the lamp and the table might compete for the figure status because they are more easily comparable in general saliency and size. The important thing is that the locational figure/ground distinction that we are dealing with here is built into the prepositions that are used in such sentences in English, and therefore they do not change across contexts (the complement of a locative preposition always denotes a ground). If this is true, then the figure/ground mappings that we are interested in are stable across different contexts.

With this in mind, let us look at the extra argument sentence in (43).

(43) Dem Stinger istdie Jacke am Bauch zu eng. the.DAT singer is the jacket at.the belly too tight

Reading A: 'The singer finds that his jacket is too tight around his waist.'

Reading B: '(It is objectively the case, or someone other than the singer finds that) the singer's jacket is too tight around his waist.'

The ambiguity of (43), which is reflected by the two translations, is well-entrenched and is easily elicited with speakers of German. (20) If the arguments given in this article to assign the affectee role to the dative argument in the A-reading of such sentences are correct, then we must ask ourselves what role the dative argument bears in reading B where the dative referent is not entailed to perceive anything. Note that we have seen arguments above that the dative argument is not a possessor. The possessor of the belly is implicit in the PP am Bauch 'on the/his belly'. The spatial relation denoted by am holds between the jacket and the singer's belly. The dative DP just serves to tie this configuration to a referent that may serve as a ground for this configuration. It is important to see that the figure role of the jacket in the spatial am-configuration does not correspond to a pronounced constituent. This gets clearer if we replace the PP am Bauch with a relative clause as in (43'), and this time only the B reading is provided.

(43') Dem Stinger istdie Jacke da, wo sie am Bauch

the.DAX singer is the jacket there where it at.the belly

ist, zu eng.

is too tight

Reading B: '(It is objectively the case, or someone other than the singer finds that) the singer's jacket is too tight where it goes around his waist.'

If we want to say that reading B of (43') has the same truth conditions as reading B of (43)--and this seems desirable--then the simplest assumption is that the subject argument of the relative clause of (43') and, more importantly, its role is implicit in (43). We thus have a recursive figure/ ground mapping. The jacket (figure) is in the region of the belly (ground), and this spatial configuration as a whole constitutes the figure whose ground is the singer. Thus, the dative can plausibly be argued to denote a ground in a figure/ground configuration, and no obvious alternatives for a thematic role assignment present themselves.

Another straightforward argument against the discrediting of gestalt roles relies on Mandarin ba-sentences like (44) (= [21c]; recall that ba allows for the integration of additional preverbal extra arguments under special conditions).

(44) Ta ba bilu sheng-le huo.

(s)he BA fireplace ignite-PRF fire

'(S)he put on a fire in the fireplace.'

Judging from the English translation alone, one might want to say that the ba-marked argument is just an internal locative in the sense of Maienborn (2001). This idea receives initial support from the substitutability of ba as witnessed by (45).

(45) Ta zai bilu sheng-le huo.

(s)heat fireplace ignite-PRV fire

'(S)he put on a fire in the fireplace.'

In (45) a preposition is used instead of ba, and the translation does not change. That this cannot be the whole story becomes evident in (46).

(46) Ta zai/*ba shulin sheng-le huo.

(s)heat/BA woods ignite-PRY fire

'(S)he put on a fire in the woods.'

(46) is ungrammatical with ba, and the reason for this is that a fire is not an essential part of any forest, while a fire forms part of the functional whole of a fireplace. We may say that ba may only be used to introduce an extra argument into a clause if (i) there is a figure/ground relation between the referent of the extra argument and the referent of the more deeply embedded argument, and if (ii) this figure/ground relation is conventional, that is, mediated by inalienability or purposivity. Put differently, the relevant figure/ground relation must in some sense be essential, and not just accidental. Pure contiguity in space alone cannot be the sufficient condition for the use of ba, because otherwise (46) should be fine with ba.

Summing up, I claim that one may meaningfully speak of thematic roles that correspond to the ground and the figure of the gestalt notion. In a stative locative state of affairs in which none of the participants is entailed to be sentient or conscious, the ground property of anchoring another referent in space remains as the only function of extra arguments as in (43) or (43'). As said above, it would amount to overstretching the notion of agentivity if one subsumed the ground notion under the agent notion, and therefore, I side with Dowty insofar as I assign "independent existence" a dubious status in the list of proto-agent entailments. Since Dowty wants to reduce all roles to the agent/patient contrast, he has no choice but to include the ground property par excellence, independent existence, among the proto-agent entailments. I would argue that the perspective should be changed. Two fundamentally different kinds of predications should be distinguished. On the one side we have predications that may be subsumed under the heading "eventualities with a cause that forms part of the linguistic conceptualization." This is the realm of the proto-agent/proto-patient contrast. On the other side we have purely stative locative predications which denote "figure/ground configurations in space whose coming into being or change is not part of the linguistic conceptualization." These two kinds of predication types form extremes in that a given complex predication need not belong to one kind to the complete exclusion of the other. For instance, predications denoting agentively caused directed motion form a mixed type. There are other eventualities in the course of which spatial configurations come into being, are changed, or cease to exist, but which have no known or linguistically conceptualized cause. Such eventualities are encoded by unaccusative predicates (cf. Abraham 2004). Whether they instantiate a mixed type between purely locative predications and purely causal predications depends on whether motion is necessarily seen as an instance of the proto-agent/proto-patient scheme, or whether motion may also be seen as a pure instantiation of the locative predication scheme. I will leave the matter unsettled here, but assume that if a sentence like The ship tore one of its sails is, as a whole, classified as denoting an eventuality with a cause, it will at least also denote or entail a subeventuality which is fully locational, in this case relating to the resulting state of there being one (functional) sail less on the mast of the ship (cf. Section 5.3).

For a last refinement, and to make the parallel between affectees and grounds as discussed here more visible, we should note that our grounds are all figures if we move up one level. At the level of the complete situations that are denoted by our extra argumental sentences, the ground referents that we are talking about never define the whole scene. They are always locata which, secondarily, become localizers. Take the example sentence The tree grew leaves. At the highest level of figure/ground mappings, the tree is a figure before a (possibly highly unspecific) ground. But with respect to the leaves and the growing event, the tree constitutes the ground. Analogous reasonings can be developed for all other ground arguments that we have encountered. And this is not a surprise if we take the mixed nature of affectees and the parallelism between affectees and intermediate grounds seriously. Affectees combine proto-agent properties and proto-patient properties, and our extra argumental grounds combine proto-ground properties and figure properties.

For this reason, I think that "ground" is not the best term to characterize the involvement of extra argumental (intermediate) ground arguments. Moreover, this terminology invites the misunderstanding that Dowty falls victim of when he confounds figure/ground configurations in space with those in information structure. I will therefore make use of Langacker's (1987: 217) well-established term "landmark" (for "intermediate ground") to refer to the locative gestalt role of some extra arguments (cf. Maienborn 2001, who also uses this term in a generative context). The corresponding term "trajectory" (for "figure") is not needed because we can use the more established term "theme" instead, if needed.

5.3. Landmark voice

Section 4.2 was devoted to the implementation of affectee extra arguments with the help of the voice head [Aff.sup.0]. The denotation of this voice head is given again in (47), and the corresponding denotation for the voice head for landmark arguments of purely stative eventualities is added in (48a). (48b) rephrases (48a), and it makes use of the specification of the landmark notion given in parentheses in (48a), but it abstracts away from the presuppositionality of the landmark position.

(47) [[[Aff.sup.0]]] = [lambda]x[lambda]e.x is an affectee in e

(48) a. [[[Ldm.sup.0]]] = [lambda]x[lambda]s.x is the landmark for s (with a landmark referent being the ground with its presupposed location in space vis-a-vis which the location of the theme or figure referent of s may be identified)

b. [[[Ldm.sup.0]]] = [lambda]x[lambda]s.s holds in a region of x

The variable s in (48) ranges over states (as opposed to events; cf. Note 15) since the prototypical field of application of landmark arguments is in stative locative predications. For the reasons given in the last subsection, I will leave the matter undecided whether dynamic eventualities involving motion may also integrate landmark arguments, or whether the protoagent/proto-patient scheme may be put to use here. Note that those sentences from the empirical survey of Section 3 that require the assumption of landmark arguments are all dynamic eventualities, but that they all denote or entail stative sub-events whose truth conditions may be spelled out with the help of the landmark role. The sentence in (49a) (= [1b]) may, for instance, be said to entail (49b) as a resulting state.

(49) a. The ship tore one of its sails.

b. 'One of the ship's sails does not exist as a functional entity on the ship (anymore).'

Now, what is the difference between (49a) and (50a)?

(50) a. One of the ship's sails tore.

b. 'One of the ship's sails does not exist as a functional entity (anymore).'

The difference can be read off the paraphrases in (49b) and (50b). They each spell out the resultant state which holds after the sail tore, but the one spelling out the resultant state of the sentence with the landmark voice head, (49b), puts a contiguity constraint on (49a) which is irrelevant for the interpretation of (50a). (50a) may be true in a context in which the sail tore while it was not in the same place as the ship or, more accurately, while it was not in its functional position on the mast of the ship. (49a) is false in such a situation. With locative predications, the fact of a state holding in a region of a landmark referent is thus the counterpart of the conscious or sentient involvement of an affectee referent into an eventuality which instantiates the proto-agent/proto-patient scheme. A German example which illustrates this analogous contrast again is provided in (51) (cf. [4b]).

(51) Dann starb (ihm) auch seine Mutter.

then died him.DAT also his mother

'Then his mother died (on him), too.'

(51) with the extra argument is only good if the dative referent is still alive, while the use of the possessive pronoun inside the more deeply embedded DP requires no such thing. (21)

It would require some more work to develop the fully compositional semantics for locative eventualities with Landmark arguments, especially with respect to the parallel implementation of predicate abstraction and event identification as proposed for the affectee head. At the present level of explicitness, the idea that landmark arguments "bind" a variable in their sister constituents the way affectee arguments "bind" variables in their VPs is only evinced by the bound use of possessive pronouns in sentences like (49a) and by the double occurrence of the ship in the partial spell-out of truth conditions as in (49b). Some more suggestive examples follow in Sections 5.4 and 5.5, but we will basically remain in the sphere of impressionistic arguments. From among the two points which conclude this subsection, I consider (i) sufficiently well-grounded to count as a result of Sections 5.1 through 5.3; (ii) must be substantiated in more detail:

(i) We need the landmark as a thematic role;

(ii) Landmarks and affectees are treated on a par with respect to extra argumentality, the dividing line between their fields of application being whether a locational eventuality or a caused eventuality is described.

5.4. Landmarks as extra arguments must be functional or organic wholes

The contrast exemplified by (52) has not been taken care of yet.

(52) a. The ship tore one of its sails.

b. * The mast tore one of its sails.

Masts are, just like sails, parts of larger functional wholes, namely ships. Judging from (52), extra argumental landmarks must be functional wholes. In other words, extra argumental landmarks must contain, or be anaphoric to a DP with, an absolute noun, as opposed to a relational noun. This statement must immediately be refined in the light of the data in (53) and (54).

(53) a. The tree grew leaves (on its branches).

b. * The branch grew leaves (on it).

(54) The branch in the vase grew leaves (on it).

The deviance of (53b) vanishes as soon as the branch is construed as a functional or organic whole. A branch in a vase is like a whole plant, and therefore it may be used as an extra argumental landmark in (54). The ungrammaticality of (53b) is thus restricted to transitive or relational uses of the noun branch, but it is not tied to the lexeme branch as such. The same holds for the contrast in (55) which goes away if the fender in (55b) is interpreted as a radio set that is shaped like a fender (or actually is a fender which is used as a radio body).

(55) a. The car bent its antenna.

b. (*) The fender bent its antenna.

Still, there remains a certain amount of data that I cannot make full sense of. While speakers of English have no problems with (55b) if the fight context is provided, (52b) continues to be judged deviant even if the mast is contextualized as a decorational mast which comes without a ship and therefore represents a complete functional whole.

The matter will be left unsettled, but this does not undermine the generalization which heads this subsection: landmarks as extra arguments must be functional or organic wholes. This generalization constitutes a necessary condition for the use of landmark extra arguments, but it is evidently not a sufficient condition.

5.5. Explicit and implicit landmark regions

The examples in (53) and (54) were good with or without the explicit mention of the spatial region of the (part of the) landmark referent where the resulting state holds (The tree grew leaves [on its branches]). I have argued above that those regions must be semantically active even if they are not pronounced. The lexical means to link the spatial region of the leaves to the branches (and to the tree as a whole) was the general English preposition of contiguity for such cases, viz. on. Given other conflation patterns of locative predications, other lexical choices become possible. Three pertinent examples from Mandarin are given in (56) (= [20a], [20b], [21a]).

(56) a. Ta ba juzi bo-le pi.

(s)he BA orange peel-PRF peel

'(S)he removed the peel of the orange.'

b. Tamen ba zhu fang-le xue.

they BA pig release-PRF blood

'They drained the pig of its blood.'

c. Zhangsan ba men shang-le suo.

Zhangsan BA door put.on-PRF lock

'Zhangsan put a lock on the door.'

In all of these examples, the information about the regions of the landmarks where the theme or patient referents (the peel, the blood, the lock) are located when the resulting state holds is encoded as part of the verbs' meanings. The verbs bo 'peel' and fang 'release' entail that the themes end up AWAY FROM or DETACHED FROM the landmarks (the orange, the pig). The verb shang in its basic meaning carries a superessive meaning 'on top of'. In (56c) it is used to express the general contiguity-plus-contact relation that was expressed by on in (53) and (54). The conflation pattern underlying (57) (= [21c]) is more opaque.

(57) Ta ba qiang ti-le yi-ge dong.

(s)he BA wall kick-Pro: 1-CL hole

'(S)he kicked a hole into the wall.'

On the one hand, one may argue that the hole is "added" to the wall and that a state is the result in which the wall has a hole in it. On the other hand, the resulting hole in the wall may be regarded as destroying the functionality of the wall, just like the tearing of a sail destroys the functionality of the sail and of the corresponding ship. Perhaps both views can be integrated into a single set of truth conditions for this sentence, but given the ungrammaticality of sentences like (46) with ba further specifications will be necessary. We said above in the context of (44)-(46) that ba may only add extra arguments to a predication if the more deeply embedded theme argument refers to a functional or essential part of the extra argumental landmark referent. Yet, a hole that's been kicked into a wall is certainly not an essential or functional part of the wall. I lack evidence to make full sense of the conflation pattern instantiated in (57) and will leave its discussion for another occasion.

5.6. Summary on landmarks and their relation to affectees

Summing up the discussion on the voice head analysis that I propose for extra argumental landmarks, one may say that landmarks are, in a sense, like affectees. A landmark is there to denote the referent which allows one to identify the place where the VP eventuality, or a substate thereof, holds. An affectee is there to denote the referent which may be sentiently/consciously affected by the causal potential of the VP event. Both affectees and landmarks are, therefore, points of reference--mental ones in the case of affectees, locative ones in the case of landmarks.

6. Conclusions and outlook

In this article, I have investigated the empirical range of extra arguments, that is, of surprising syntactic core arguments, in German, English, and Mandarin. One important result concerns the fact that extra arguments occur in all syntactic core functions. An extra argument may, however, not be the most deeply embedded argument in a predication. This restriction was modelled with the help of the voice analysis that I proposed for extra argumentality. The voice heads that integrate extra arguments into the structure come with an identity requirement. The referent denoted by the extra argument must, in a different role, already be part of the denotation of the sister node of the voice head.

Extra argumentality comes in two major kinds. The extra argument may either be an affectee or a landmark.

The affectee referents studied here fulfilled the identity requirement by simultaneously being possessor referents in the VP eventualities. The modeling of this double role strives to account for the longstanding intuition of "affected possessor" arguments that other theories model as a possessor argument that has been moved to the extra argumental position, or by stipulating a possessor in the argument structure of (derived) verbs. I have tried to show that my voice analysis is superior to such attempts. Hole (2005) extends the analysis of affectee referents that are possessors at the same time to so-called beneficiaries. In that study, being a beneficiary is decomposed into affecteehood plus purposivity. If someone bakes me a cake, I will be an affectee in this event, and the cake will have a purpose for me (to feed me, to make me happy, or whatever may have been the intention on the part of the agent referent).

The integration of landmark arguments is the other big field of application for extra arguments. If the sister node of the extra argumental voice head denotes a state involving a spatial configuration (as a sub-event), the extra argument serves to denote the spatial ground where the figure of the spatial configuration holds.

I have argued against attempts to characterize the landmark notion in terms of the agent/patent contrast, even though this is probably the favorite choice among researchers. I claim instead that there is a major conceptual and linguistic split between eventualities that are structured according to the (locational) figure/ground scheme, and according to the (causal) agent/patient scheme.

An obvious blind spot of the proposal as defended here concerns the lexical or categorial status of the variable that gets identified in the more deeply embedded argument (as, for instance, in She stared me in the x--eyes, i.e., in the eyes of the speaker). Here and in Hole (2005), I have nothing to say about this problem. Still, arguments supporting the existence and linguistic activity of this variable are delivered in Hole (2005), and locality constraints for extra argumental variable identification are stated. We may, therefore, be confident that the variable in the more deeply embedded argument is not just a chimera.

Received 20 January 2004 Revised version received 31 March 2005

Ludwig-Maximilians-Universitat, Munich

References

Abraham, Werner (2004). VP-internal subjects as "unaccusatives": Burzio's "object account" vs. the "perfectivity account". In The Composition of Meaning--From Lexeme to Discourse, Alice ter Meulen and Werner Abraham (eds.), 83-111. Amsterdam: John Benjamins.

Aissen, Judith (1987). Tzotzil Clause Structure. Dordrecht: Reidel.

Benveniste, Emile (1966). Problemes de linguistique generale I. Paris: Gallimard.

Borer, Hagit; and Grodzinsky, Yosef (1986). Syntactic cliticization and lexical cliticization: the case of Hebrew dative critics. In Syntax and Semantics. Vol. 19, Hagit Borer (ed.), 175-217. New York: Academic Press.

Bossong, Georg (1998). Le marquage de l'experient dans les langues d'Europe. In Actance et Valence dans les Langues de l'Europe, Jack Feuillet (ed.), 769-788. Berlin and New York: Mouton de Gruyter.

Bresnan, Joan (1983). The passive in lexical theory. In The Mental Representation of Grammatical Relations, Joan Bresnan (ed.), 3-86. Cambridge, MA: MIT Press.

Buring, Daniel (forthcoming). Focus projection and default prominence. In The Architecture of Focus. Studies in Generative Grammar, Valeria Molnar and Susanne Winkler (eds.). Berlin and New York: Mouton de Gruyter.

Cheung, Hung-Nin Samuel (1973). A comparative study in Chinese grammars: the ba-construction. Journal of Chinese Linguistics 1(3), 343-382.

Den Dikken, Marcel (1997). The syntax of possession and the verb 'have'. Lingua 101, 151-183.

Dowty, David (1991). Thematic proto-roles and argument selection. Language 67(3), 547-619.

Durscheid, Christa (1999). Die verbalen Kasus des Deutschen. Untersuchungen zur Syntax, Semantik und Perspektive. Berlin: Walter de Gruyter.

Faarlund, Jan-Terje (1999). The notion of oblique subject and its status in the history of Icelandic. Working Papers in Scandinavian Syntax 63, 1-44.

Fox, Barbara (1981). Body part syntax. Studies in Language 5(3), 323-342.

Gallmann, Peter (1992). Dativanhebung? Groninger Arbeiten zur Germanistischen Linguistik 5, 92-122.

Gruber, Jeffrey S. (1965). Studies in Lexical Relations. Bloomington: Indiana University Linguistics Club.

Gueron, Jacqueline (1985). Inalienable possession, PRO-inclusion and lexical chains. In Grammatical Representation, Jacqueline Gueron, Hans-Georg Obenauer, and Jean-Yves Pollock (eds.), 43-86. Dordrecht: Foris.

Hohle, Tilman (1982). Explikation fur "normale Betonung" und "normale Wortstellung". In Satzglieder im Deutschen. Vorschliige zu ihrer syntaktischen, semantischen und pragmatischen Fundierung, Werner Abraham (ed.), 75-153. Tubingen: Narr.

Hole, Daniel (2002). Spell-bound? Accounting for unpredictable self-forms in J. K. Rowling's Harry Potter Stories. Zeitschrift fur Anglistik und Amerikanistik 50(3), 285-300.

--(2004). Extra argumentarity--a binding account of "possessor raising" in German, English and Mandarin. Possessives and Beyond: Semantics and Syntax, Ji-Yung Kim, Barbara H. Partee, and Yury A. Lander (eds.), 365-383. Amherst, MA: GLSA Publications.

--(2005). Reconciling "possessor" datives and "beneficiary" datives--towards a unified voice account of dative binding in German. In Event Arguments in Syntax, Semantics and Discourse, Claudia Maienborn and Angelika Wollstein (eds.), 213-242. Tubingen: Niemeyer.

Hopper, Paul; and Thompson, Sandra A. (1980). Transitivity in grammar and discourse. Language 53(1), 251-299.

Isatenko, Alexander V. (1965). Das syntaktische Verhaltnis der Bezeichnungen von Korperteilen im Deutschen. Studia Grammatica 5, 7-27.

Kayne, Richard S. (1975). French Syntax. The Transformational Cycle. Cambridge, MA: MIT Press.

Keenan, Edward L.; and Ralalaoherivony, Baholisoa (2000). Raising from NP in Malagasy. Lingvisticae Investigationes 23(1), 1-44.

Kittila, Seppo (forthcoming). The anomaly of the verb 'give' explained by its high (formal and semantic) transitivity. Linguistics 44-3.

Konig, Ekkehard; and Haspelmath, Martin (1998). Les constructions a possesseur externe dans les langues d'Europe. In Actance et Valence dans les Langues de l'Europe, Jack Feuillet (ed.), 525-606. Berlin and New York: Mouton de Gruyter.

Kratzer, Angelika (1989). An investigation of the lumps of thought. Linguistics and Philosophy 12, 607-653.

--(1996). Severing the external argument from its verb. In Phrase Structure and the Lexicon, Johan Rooryck and Lauri Zaring (eds.), 109-137. Dordrecht: Kluwer.

--(2003). The event argument and the semantics of verbs. Ms., University of Massachusetts at Amherst, [http:ffsemanticsarchive.net/Archive/GU1NWM4Z/].

Landau, Idan (1999). Possessor raising and the structure of VP. Lingua 107, 1-37.

Langacker, Ronald W. (1987). Foundations of Cognitive Grammar. Volume 1: Theoretical Prerequisites. Stanford, CA: Stanford University Press.

Leirbukt, Oddleif (1997). Untersuchungen zum "bekommen"-Passiv im heutigen Deutsch. Tubingen: Niemeyer.

Lenerz, Jurgen (1977). Zur Abfolge nominaler Satzglieder im Deutschen. Tubingen: Narr.

Li, Charles N.; and Thompson, Sandra A. (1981). Mandarin Chinese. A Functional Reference Grammar. Los Angeles: University of California Press.

Li, Yen-hui Audrey (2001). The ba construction. Ms., University of Southern California, Los Angeles, [http://www.usc.edu/dept/LAS/ealc/chinling/synformal.htm].

Maienborn, Claudia (2001). On the position and interpretation of locative modifiers. Natural Language Semantics 9(2), 191-240.

Meinunger, Andre (forthcoming). Remarks on the projection of dative arguments in German. In Datives and Other Cases, Daniel Hole, Andre Meinunger, and Werner Abraham (eds.). Amsterdam: John Benjamins.

O'Connor, Mary C. (1996). The situated interpretation of possessor-raising. In Grammatical Constructions. Their Form and Meaning, Masayoshi Shibatani and Sandra A. Thompson (eds.), 125-156. Oxford: Clarendon.

Payne, Doris; and Barshi, Immanuel (eds.) (1999). External Possession. Amsterdam: John Benjamins.

Perlmutter, David (ed.) (1983). Studies in Relational Grammar. Chicago: University of Chicago Press.

--; and Joseph, Brian (eds.) (1990). Studies in Relational Grammar 3. Chicago: University of Chicago Press.

--; and Postal, Paul (1983). Towards a universal characterization of passivization. In Studies in Relational Grammar, David Perlmutter (ed.), 3-29. Chicago: University of Chicago Press.

Primus, Beatrice (1999). Cases and Thematic Roles--Ergative, Accusative and Active. Tubingen: Niemeyer.

--(2004). Proto-roles and case selection in optimality theory. Ms., Arbeiten des SFB 282 "Theorie des Lexikons" Nr. 122, Universitat zu Koln.

Reis, Marga (1985). Mona Lisa kriegt zuviel. Vom sogenannten "Rezipientenpassiv" im Deutschen. Linguistische Berichte 96, 140-155.

Rohdenburg, Gunter (1974). Sekundare Subjektivierungen im Englischen und Deutschen. Vergleichende Untersuchungen zur Verb- und Adjektivsyntax. PAKS-Arbeitsbericht Nr. 8. Bielefeld: Cornelsen-Velhagen & Klasing.

Schwarzschild, Roger (1999). Givenness, AvoidF and other constraints on the placement of accent. Natural Language Semantics 7, 141-177.

Sells, Peter (1987). Aspects of logophoricity. Linguistic Inquiry 18(3), 445-479.

Shibatani, Masayoshi (1994). An integrational approach to possessor raising, ethical datives, and adversative passives. BLS 20, 461-486.

Talmy, Leonard (1985). Lexicalization patterns: semantic structure in lexical forms. In Language Typology and Syntactic Description, Vol. 3, Timothy Shopen (ed.), 57-149. Cambridge: Cambridge University Press.

--(2000). Toward a Cognitive Semantics. Vol. 2. Cambridge, MA: MIT Press.

Tsao, Feng-fu (1987). A topic-comment approach to the ba construction. Journal of Chinese Linguistics 15(1), 1-53.

Tuggy, David (1980). [??]Ethical dative and Possessor omission si, Possessor ascension no!. Workpapers of the SIL 24, 97-141.

Vergnaud, Jean-Roger; and Zubizarreta, Maria Luisa (1992). The definite determiner and the inalienable constructions in French and English. Linguistic Inquiry 23, 595-652.

Wunderlich, Dieter (1996). dem Freund die Hand auf die Schulter legen. In Wenn die Semantik arbeitet. Klaus Baumgartner zum 65. Geburtstag, Gisela Harras and Manfred Bierwisch (eds.), 331-360. Tubingen: Niemeyer.

Wunderlich, Dieter (2000). Predicate composition and argument extension as general options--a study in the interface of semantic and conceptual structure. In Lexicon in Focus. Studia Grammatica 45, Barbara Stiebels and Dieter Wunderlich (eds.), 247-270. Berlin: Akademie.

Zhang, Ning (1998a). The interactions between construction meaning and lexical meaning. Linguistics 36(5), 957-980.

--(1998b). Argument interpretations in the ditransitive construction. Nordic Journal of Linguistics 21, 179-209.

Notes

* This article is a revised and heavily extended version of Hole (2004). While Sections 2 and 3 were transferred from the other paper largely unchanged, Section 4 has been extended considerably. Section 5 is new. I would like to thank two Linguistics reviewers who made me sharpen my points and corrected inadequacies of a preliminary version. The following people were among those who helped me in the course of developing the analysis presented here: Daniel BOring, Jacquelyn Deal, Andreas Dufter, Volker Gast, Joachim Jacobs, Gerson Klumpp, Ekkehard Konig, Elisabeth Leiss, Peter Siemund, Tham Shiao Wei, Zhang Jie, and Zhang Ning. Remaining errors are mine. Correspondence address: Institut fur Deutsche Philologie, Ludwig-Maximilians-Universitat Munthen, Schellingstr. 3/RG, 80799 Munich, Germany. E-mail: hole@lmu.de.

(1.) The following abbreviations are used in glosses: ACC--accusative; BA--cf. Section 3.2.; CL--classifier; DAT--dative; IND--indicative; NOM--nominative; PASS--passive marker; PRF--perfective aspect; PRT--particle; PST--past tense. I only provide glosses for morphosyntactic features if they are immediately relevant to the discussion at hand.

(2.) Among researchers with an intimate knowledge of continental European languages, the idea of syntactic possessor raising has often met with criticism. Cf. Tuggy (1980) for such a strongly opposing view.

(3.) The s-variable is Wunderlich's event argument.

(4.) Cf., for example, O'Connor (1996) for a diligent overview of relevant contrasts in Northern Pomo which, even though the possessor raising/ascension terminology is agnostically adopted, points to the manifold semantic-pragmatic differences between "raised" and "nonraised" structures. Unfortunately, O'Connor does not challenge the claim of truth-conditional irrelevance of the choice between the two structures.

(5.) Independently of the external possession paradigm, Shibatani (1994) makes an attempt to explain extra argumentality or, as he calls it, extra-thematic licensing of arguments, in terms of relevance. It is not clear to me what the exact empirical predictions of a relevance account as opposed to a thematic-role account are and what they mean in terms of a syntactic and semantic implementation. However, Shibatani's work must be acknowledged for widening the perspective such that a broad range of extra argamentality facts from many different languages are integrated into a single picture.

(6.) The general ideas in the arguments of Sections 2.1 through 2.3 have been around in the literature for decades, even though their exact shapes may differ. The argument in Section 2.4 is from Hole (2005) and, to the best of my knowledge, it constitutes a new finding together with the additional argument delivered in Section 4.3.1.

(7.) Note that extra arguments in English do not occur in postverbal position as frequently as, for instance, in German. They are restricted to certain configurations with directional complements that encode bodily action. Extra subjects are more widespread in English (see Section 3.1).

(8.) As already mentioned in Note 6, this argument is from Hole (2005).

(9.) In German, extra arguments in subject function as defined here are extremely rare, if they exist at all. The reason for this seems to be that German allows PPs in its prefield so easily. A possible example of a sentence with a construction as in (12) is provided in (i).

(i) Der Baum treibt B1atter.

the tree drives leaves

'The tree starts growing leaves.'

Whether this example qualifies as a case in point depends on whether der Baum 'the tree' is interpreted as a marginal agent, or as a kind of landmark (see Section 5.2 for the justification of the landmark role). According to the specifications in that section, we would be dealing with a clear landmark argument only if the tree is not at all agentively involved in the eventuality of growing leaves. Two other sentence patterns have landmark subjects in a rather straightforward fashion, but the complements of the underived verb stems are encoded in a special complement type with the derived verbs; cf. (ii) and (iii).

(ii) Der Platz steht voller Menschen.

the square stands full. of people

'The square is full of people (standing there).'

(iii) Der Saal halltewider vor Larm.

the haal resonated because.of noise

'The hall resonated with noise.'

Even though these patterns should definitely be covered by an overarching theory of extra-argumental voice as propagated here, I will, for the sake of perspicuity, stick to the more narrowly defined type of construction in which the syntactic core functions of extra arguments do not downgrade the initial internal arguments to complement PPs. For one more German sentence pattern with extra-argumental subjects, but demoted initial core arguments, cf. (iv).

(iv) Paul wackelte mit den Ohren.

Paul wiggled with the ears

'Paul was wigging his ears.'

(10.) I would like to thank Waltraud Paul for reminding me of this sound criterion of nontopichood.

(11.) Kittila (forthcoming) demonstrates that 'give' verbs are irregular in many languages in that they display idiosyncratic properties not found with other verbs of the same larger class, typically ditransitives. This fits in well with Hole's (2005) reservations against 'give' verbs as points of departure for the elucidation of clause patterns with multiple arguments.

(12.) This generalization only holds cum grano salis. Schwarzschild (1999) and Buring (forthcoming) show convincingly that focal accents on adjuncts are sufficient to focus-mark constituents that are larger than those adjuncts. What matters is that the focal material outside the adjunct must be given (in the sense of Schwarzschild 1999: 151-152). A pertinent example from Buring (forthcoming) is provided in (i) (Buring uses German examples for illustration because even critical readers will readily agree that the German PP in [i] cannot be a complement, while its right-peripheral counterpart in the English translation might give rise to the idea that with his daughter is a complement).

(i) Q: Zur Tatzeit hat Lolek mit seiner Frau einen Einkaufsbummel gemacht. Aber was ist Boleks Alibi?

'Q: At the time of the crime, Lolek was out shopping with his wife. But what was Bolek's alibi?'

A: Bolek hat [[mit seiner TOCHter einen Einkaufsbummel gemacht].sub.FOC]. Bolek has with his daughter a shopping.stroll made

'A: Bolek was [out shopping with his DAUGHTER].'

In this example, einen Einkaufsbummel gemacht 'been out shopping' may not bear a focus accent, since this bit of information has been given about Lolek in the preceding context; on the other hand it forms part of the new information about Bolek, and therefore it is part of the larger constituent which comprises the whole focus of the sentence. Under these conditions, a focus accent on an adjunct suffices to focus-mark the whole sentence minus the subject (and the auxiliary).

If this is so, and if a parallel argument can be stated for sentences as in (25) and (25'), what remains of the claim that the possibilities of focus projection may serve as a diagnostic for unmarked word order in the question at hand? Isn't the whole diagnostic rendered useless by Schwarzschild's and Buring's findings? It fortunately isn't. The point is that Hohle's original generalizations remain fully valid if (near-)empty, out-of-the-blue contexts are chosen. Hohle's patterns may not be reproduced in all discourse environments, but they are stable if the context is (near-)empty. This is the reason why one should be more careful to choose only indefinite arguments, and avoid all presuppositional elements, in examples used to argue for a specific canonical word order.

(13.) Such configurations are typical of most languages of Europe, except for some languages on its (insular) fringes, among them Breton, English, and Turkish (cf. Bossong 1998).

(14.) Vergnaud and Zubizarreta (1992) defend a similar claim for French indirect objects in what they call the "inalienable construction" in French. Even though the general ideas of Vergnaud and Zubizarreta's proposal are similar to the proposal as laid out in Section 4.2, the present analysis is not just a variant of the analysis of the inalienable construction in French. See Section 4.3.2 for more discussion.

(15.)(i) below is a list of semantic types and variables/values as used by Kratzer (1989 and elsewhere) and here. The confusing homonymies are regrettable, but I don't see that it would be a step ahead if I, as opposed to Angelika Kratzer, proposed a more perspicuous system.

(i) semantic type variables/values

truth-value: <t> 1 (true), 0 (false)

individuals: <e> x, y, z

eventualities/situations: <s> e (dynamic eventualities, events proper), s (states)

(16.) It is confusing indeed that V&Z compare the alienable and inalienable uses of nouns to verbs that may be used transitively (as causatives), or intransitively, but then represent the allegedly "transitive" inalienable variants of nominal predicates with a single argument slot throughout their paper (consequently the predicates corresponding to alienable uses of nouns have no argument slot at all in V&Z's paper).

(17.) The diagrams in V&Z (1992: 614) are obviously intended to contribute to the elucidation of this problem, but the representations are somewhat idiosyncratic, and even with some effort I am not sure that I can fully understand them.

(18.) The semantics of extra argumentality of the following examples from the survey in Section 3 have thus been clarified: (13a), (13b), (14), (15), (16), (17), (26a)-(26c).

(19.) (i) and (ii) are Dowty's (1991: 572) lists of proto-agent and proto-patient properties, respectively:

(i) volitional involvement in the event or state sentience (and/or perception) causing an event or change of state in another participant movement (relative to the position of another participant) (exists independently of the event named by the verb)

(ii) undergoes change of state incremental theme causally affected by another participant stationary relative to movement of another participant (does not exist independently of the event, or not at all)

(20.) The contrast which is relevant in (43) recurs in different parts of grammar, and accordingly it has several names. One area where it surfaces is in the area of reflexivization (cf. Sells' [1987] PIVOT vs. SOURCE/SELF, Hole's [2002: 292] somatophoricity vs. logophoricity, or the triggering of de-re/de-dicto contrasts in intensional semantics).

(21.) As alluded to in the preceding note already, another domain where consciousness/ sentience is dealt with on a par with locatedness vis-a-vis something else is reflexivization, as treated in Hole (2002: 290-293). It is argued there that so-called untriggered reflexives in English may, among other things, either indicate that the predication in which the reflexive occurs reports the mental state of the reflexive participant, or that an eventuality is presented with the reflexive participant as the landmark of the eventuality. I would like to thank Robert Mailhammer and Theo Vennemann (pers. comm.) in this context, who reminded me of the well-known Germanic case syncretism which had led to a merger of the locative and the dative in Old High German already. I must, however, leave for another occasion the more exact explication of the historical implications of my proposal.