Thematic Hierarchies and Binding Theory: Evidence from Greek*

1. Introduction

A well-known problem in the syntax of anaphora is a tension between configurational and non­configurational effects on binding. There is evidence that structural relations govern anaphora but there is also counterevidence supporting a non-configurational approach. This asymmetry raises a question concerning the relations anaphora is based on and, ultimately, the module regulating binding theory. We have two phenomena in mind where non-configurational notions play a significant role: the so-called logophoric interpretation in the case of non-clause bound binding, and thematic hierarchy effects in the case of clause-bound binding. We will not have much to say about the former in this paper. We will, instead, focus on the thematic hierarchy effects on clause-bound binding. We will argue that the configurational effects on binding cannot be reduced to argument­structure relations reflecting Th-prominence and, conversely, that Th-hierarchy effects cannot be reduced to syntactic structure. Once it is recognized that there are two distinct modules regulating anaphoric relations (Reinhart & Reuland; R&R93), namely Chain Formation, which is part of the Computational System, and Reflexivization, which applies at the stage of translating from LF to semantic representation, the controversy can be resolved in a principled manner. The empirical evidence in favour of our proposal is drawn from a number of systematic asymmetries between English and Greek.

2. Configurational or Thematic Constraints?

Binding Theory as we know it in the P&P framework (cf. Chomsky 1986, Chomsky & Lasnik 1991) is stated in strictly configurational terms. The binding conditions could be called configurational in the sense that, first, the antecedent of an anaphor a must c-command a and, secondly, that there is a domain in which a must be bound. This domain, the governing category, is determined by the governor of a and a predicational head, both contained in that domain. Now observe the examples in (1-3)

(1) a. I talked to Mary_i about herself_i
b. *I talked about Mary_i to herself_i

(2) a. The boys_i fear each other_i
b. ??The boys_i concern each other_i

(3) a. John_i washes himself_i
b. *Himself_i/Heself_i washes John_i

In (1-3) the anaphors are all coindexed with an NP within their governing category but c-command restrictions are not uniformly satisfied. In (3b) the antecedent does not c-command the anaphor, thus correctly predicting its ungrammaticality. However, it is unexpected that grammaticality judgements in (1) and (2) differ. In both (2a) and (2b) the reciprocal is c-commanded, but still (2b) marginal, if not ungrammatical.[1] In (1) one might argue that the anaphor is not c-commanded by its antecedent, but still (1a) is generally considered grammatical, contrary to (1b) for which there is no straightforward explanation (Chomsky 1981:225-226). On the other hand, the examples in (1­3) are all uniformly and straightforwardly accounted for if Th­prominence governs anaphora, assuming a version of the thematic hierarchy as in (4) (Jackendoff 1972):[2]

(4) Agent > Experiencer >Goal/Source/Location/Benefactor >
Theme (Grimshaw 1990)

The ungrammaticality of the b-examples could all be explained since in these cases the antecedent is less prominent on the Thematic Hierarchy than the anaphor itself. In (1b) a Theme would bind a Goal, in (2b) a Theme an Experiencer and in (3b) a Theme an Agent.

However, there are numerous cases in English which directly falsify a Th­approach. We will briefly mention some examples from the literature. In reaction to Jackendoff's (72) proposal Hust&Brame (1976) and Freidin (1975) brought up cases as in (5):

(5) a. John_i was beside himself_i with fear (th > loc)
b. I_i am getting ahead of myself_i (th > go)
c. They_i were left to themselves_i(th > go/lo)
d. The branch_i was bent so far that it_i was actually touching
itself_i near the trunk(th > lo/go)
e. John_i is always tripping over himself_i (th > lo)

A more systematic case is exemplified by ditransitive constructions as the sentences in (6) (Barss&Lasnik 1986, Pollard&Sag 1994, Williams 1994):

(6) a. I showed Mary_i to herself_i
b. Mary explained Doris_i to herself_i

In all such cases a Theme binds a Goal. Such examples are, therefore, a problem for a thematic approach unless we start refining the definition of thematic roles and their ordering (Wilkins (1988), Jackendoff (1990,1992)). In a configurational approach (6) is explained under a Larsonian/ Pesetskyan/ Kaynian-type structure or by appealing to the structure introduced by the preposition in a ternary-branching structure (or by o-command (Pollard&Sag (1994)).

Passives present another well-known case where thematic prominence seems to be violated. Note, however, that in these cases judgements seem to vary:

(7) Passives: Goal/Theme > Agent

a. John_i was given a book by himself_i (Williams 1994)
b. The only barber_i who was shaved by himself_i was Figaro
(Pollard &Sag 1994)
c. *John_i was killed by himself_i(Kuno 1987; Baker,Johnson
&Roberts 1989)
d. ?*Sue_i was seen by herself_i in the mirror (Pesetsky 1994)
e. ?*John_i was given a book by himself_i (Pesetsky 1994)


(8) Passives: Agent > Goal/Theme

a. A book was given by John_i to himself_i (Williams 1994)
b. ??Mary was spoken to by John_i about himself_i (Kuno 1987)
c. *Himself_i was killed by John_i (Kuno 1987)

The grammaticality judgements in (7a-c) and (8b-d) are problematic in a thematic approach, unproblematic in a configurational approach (unless we accept Baker,Johnson&Roberts (1989) analysis of passives). The reverse is true for (8a). (7d-f) can only be accounted for under a Baker, Johnson&Roberts (1989) analysis of passives (cf. Pesetsky 1995:107 for an alternative to their original proposal).

In view of the facts discussed so far, we conclude that neither the thematic approach nor the configurational approach can accommodate the full range of data. This is not surprising since both approaches treat binding effects as a unitary phenomenon. It appears that we need a framework which allows us to distinguish configurational effects from non-configurational effects. A potential solution might be a rule system in which both configurational and thematic conditions are operative simultaneously. Basically this is proposed in Jackendoff (1972, 1992). There are also proposals which claim that a ranked order of constraints exist. Under the latter approach, there are, in principle, two options. The first option is to first check configurational constraints, and then take thematic constraints as a (secondary) filter. Such an approach is advocated in Grimshaw (1987) and Hellan (1988). In (9) a potential interpretation of its predictions is spelled-out:

(9) First check configurational conditions (Conf), then thematic
(Them):

- if Conf ÷ and Them ÷ -> ÷

- if Conf ÷ and Them * -> ?, wavering judgements

- if Conf * and Them * -> *

- if Conf * and Them ÷ -> *

Alternatively, one might first check thematic constraints, and then take configurational constraints as an additional (secondary) factor into consideration:[3]

(10) First check thematic conditions (Them), then
configurational (Conf):

- if Them ÷ and Conf ÷ -> ÷

- if Them ÷ and Conf * -> ?, wavering judgements

- if Them * and Conf * -> *

- if Them * and Conf ÷ -> *

If we check the alternatives against the judgements given above, (9) seems to work better than (10), for English. Whatever option we take, however, we should address at least two questions. (i) How exactly should we define the interaction between the syntactic and thematic constraints, and (ii) is there reason to expect properties of theta-structure to intrude upon what appears to be essentially configurational generalizations. We will return to the first question in section 4 and the second question in section 5. But first we will introduce some anaphora facts from Greek, a language in which thematic restrictions seem to play a more prominent role than configurational restrictions (cf. also Sells 1988 on Albanian and Clark 1992 on Toba Batak)

3. Greek Thematic Hierarchy Effects

Greek anaphors strikingly differ from the English anaphors discussed above in that their distribution seems to be strongly determined by the thematic hierarchy. This is supported by the facts discussed in the next subsections.

3.1 Nominative Anaphors[4]

Consider the examples in (11). In (11a-b) but not in (11c-e) a nominative anaphor is allowed.

(11) a. [O eaftos tu]_i tu aresi tu Petru_i
The self his(N) Cl(D) like-3sg the Peter(D)

"Himself pleases Peter"

b. [O eaftos tu]_i ton afora ton Petro_i
The self his(N) cl(A) concern-3sg the Peter(A)

"Himself concerns Peter"

c. *[O eaftos tu]_i ton antipathi ton Petro_i
The self his(N) Cl(A) dislike-3sg the Peter(A)

"Himself dislikes Peter"

d. *[O Petros]_i ton afora [ton eafto tu]_i
The Peter(N) Cl(A) concerns the self his(A)

"Peter concerns himself"

e. *[O eaftos tu]_i ton xtipise ton Petro_i
The self his(N) Cl(A) hit the Peter(A)

"Himself hit Peter"

In a thematic approach the facts in (11) are predicted because in (11a,b) the antecedent is thematically more prominent that the anaphor (Experiencer > Theme). In (11c,d), the thematic prominence is reversed and, consequently, there is ungrammaticality. (11e) is an example of a straightforward transitive verb where the subject is an Agent and the verb a Theme, thus violating the prominence hierarchy.

3.2 Double Object Constructions (Dimitriadis 1994)

Double object constructions are well-known case where a thematic hierarchy approach seems to fail (see (6) above). However, Greek seems to behave differently with respect to English, although double object constructions could be analyzed like English from a configurational perspective (cf. Anagnostopoulou&Everaert 1996):

(12) a. Ediksa s-[tin Maria]_i [ton eafto tis]_i
Showed-I to-the Mary the self(A) her

"I showed to Mary herself"

b. *Ediksa [tin Maria]_i s-[ton eafto tis]_i
Showed-I the Mary(A) to-the self her

"I showed Mary to herself"

(13) a. Periegrapsa/Apokalypsa s-[tin Maria]_i [ton eafto tis]_i
Described-I/Revealed-I to-the Mary the self(A) her

"I described/revealed/analysed to Mary herself"

b. ?*Periegrapsa/Apokalypsa [tin Maria]_i s-[ton eafto tis]_i
?*Described-I/Revealed-I the Mary to-the self her

"I described/revealed Mary to herself"

(14) a. Poulisa s-[ton sklavo]_i [ton eafto tu]_i
Sold-I to-the slave the self his

"I sold/traded to the slave himself"

b. *?Poulisa [ton sklavo]_i s-[ton eafto tu]_i
*?Sold-I the slave(A) to-the self his

"I sold the slave to himself"

In the a-examples of (12-14) a Goal antecedent binds a Theme reflexive, and consequently the sentences are grammatical. Reversing the thematic properties in the b-examples leads to ungrammaticality.

3.3 Passives

In the case of passives, Greek and English again show a strikingly different behaviour. In Greek the thematic hierarchy must be observed:

(15) a. ?*To tragoudi afierothike s-[ton Janni]_i apo [ton eafto tu]_i
The song(N) was dedicated to-the Janni by the self his

"The song was dedicated to John by himself"

b. To tragoudi afierothike apo [ton Janni] s-[ton eafto tu]_i
The song(N) was dedicated by the John to-the self his

"The song was dedicated by John to himself"

c. *?[O monos koureas]_i pu ksiristike pote apo [ton eafto tu]_i
itan o Figaro
The only barber that was shaved ever by the self his was the Figaro

"The only barber that was ever shaved by himself was Figaro"

Example (15a) is ungrammatical because a Goal antecedent binds an Agent reflexive and (15b) is grammatical because thematic prominence is observed. Example (15c), a literal translation of (7b) from Pollard&Sag (1994), shows that the thematic hierarchy must be observed in Greek even in cases where this does not happen in English.

3.4 Conclusion

From the literature we know that there are languages in which the distribution of anaphors can be more appropriately characterized in terms of thematic-constraints. If we assume that Greek is such a language, then the facts in (11-15) can be described. However, there is one conceptual problem: Why are grammaticality judgements different for English and Greek. In other words, why would the grammar allow for a (parametric) option between `syntactic' vs. `thematic' prominence?

Anagnostopoulou&Everaert (1996a,b) argue that English and Greek anaphors differ with respect to their internal structure and this difference is responsible for their different distributional properties. Adopting the Reflexivity framework outlined in Reinhart&Reuland (1993, 1995) we will show that a correlation can be established between the structure of anaphors and their behaviour, i.e. their sensitivity to syntactic vs. thematic constraints.

4. The Structure of the Greek Anaphor and its Consequences for Reflexivity

Reinhart & Reuland (1993) propose that there are two modules regulating the distribution of anaphors/pronominals. Configurational effects are due to Ch-Formation (16), while the domain of Reflexivization is defined over predicates without making reference to syntactic structure (17).

(16) a. A maximal A-chain (a₁,....,a_n) contains exactly one link - a₁ - that is both +R and Case-marked.

b. Generalized chain condition

C=(a₁,...,a_n) is a chain iff C is the maximal sequence such that (i) there is an index i such that for all j, 1£j£n, a_j carries that index, and (ii) for all j, 1£j<n a_j governs a_j+1

(17) a. A reflexive-marked syntactic predicate is reflexive

b. A reflexive semantic predicate is reflexive-marked

(18) a. Jan_i haat zichzelf_i
John hates himself

Well-formed in terms of (17) Well-formed in terms of (16)

b. *Zichzelf_i haat Jan_i
Himself hates John

Well-formed in terms of (17) Ill-formed in terms of (16)

Now assume that an anaphor does not undergo Chain-Formation. It will then not be subject to the configurational restrictions linked to the Chain. In such a case examples like in (18) would be equally acceptable. Iatridou (1988) has shown that o eaftos tu has basically the structure of an inalienable possession NP, contrary to himself, which is a pronoun. In Anagnostopoulou & Everaert (1996) we follow Iatridou and treat the difference between the Greek and the English anaphor as in (19):

(19) a. [_DP [_D' [_D him] [_NP self] ]' ]

b. [_DP [_D' [_D o] [_FP [_F' [_F eaftos_i] [_NP [_Spec-NP tu] [_N' [_N e_i] ] ] ] ] ] ]

Given a structure as in (19), there is no chain formation in Greek, cf. (20).

(20) [O Jannis]_i pleni [ton eafto tu_i]_j
the John washes the self his

"John washes himself"

In (20) the subject O Jannis and the object ton eafto tu do not form a chain because they are not co-indexed. The two arguments in (20) that are co-indexed cannot form a chain because either D_ or N_ will act as Minimality Barriers (cf. Chomsky 1986:74-80) blocking antecedent government by the subject which is required by the Generalized Chain Condition in (16a). With respect to Binding, the predicate in (20) is reflexive-marked since one of its arguments is a SELF-anaphor. This means that two arguments of the predicate should be co-indexed to satisfy binding condition A. However, the two elements that are coindexed in (20) are, structurally, arguments of different predicates. This technical problem can be circumvented once we follow Fox 1993 who argues on independent grounds that Condition A, which is defined over syntactic predicates, should be reduced to Ch-Formation. This leaves only condition B which operates on semantic predicates and is defined as in (17'):

(17') Revised Condition B (Fox 1993)

A semantic predicate is reflexive iff it is reflexive marked

We will return to a more exact formulation of the notions `reflexive' and `reflexive marked' under which the Greek anaphor can be accommodated in section 5 below.

A consequence of the assumption that Chain Formation does not apply in Greek is that the configurational effects linked to the chain are predicted to work differently in this language than in languages like English, Dutch, Italian, etc where Chain Formation applies. Above we pointed out that the non-existence of nominative anaphors is attributed to a violation of the condition on Chain Formation (16a) in the Reflexivity framework. Thus, if there is no Chain Formation, then we expect nominative anaphors to be grammatical in Greek, contrary to English:

(21) a. [O eaftos tu]_i tu aresi [tu Petru]_i
The self his(N) Cl(D) like-3sg the Peter(D)

"Himself pleases Peter"

b. *Himself_i pleases Peter_i

Note, however, that this prediction is only partially borne out for Greek: inverse-linking psych verbs (cf. 11a,b/21) with nominative anaphors are grammatical but the prediction fails in the case of simple transitive verbs or experiencer-subject verbs as in (11c), repeated here:

(11c) *[O eaftos tu]_i ton antipathi ton Petro_i
The self his(N) Cl(A) dislike-3sg the Peter(A)

"Himself dislikes John"

In Anagnostopoulou&Everaert (1996a,b) we explore a structural account for this distinction, making use of (covert) noun-incorporation, but, clearly, a thematic approach could also account for the difference. In (21a) an Experiencer antecedent binds a Theme reflexive, while in (11c) a Theme antecedent binds an Experiencer reflexive. This, and the other facts from double object constructions and passives described in section 3 seem to indicate that the following Generalization holds:

(22) In languages (like English) where Ch-Formation applies, Th-prominence is not observed. In languages (like Greek) where Ch-Formation does not apply, Th-prominence is observed.

5. The Analysis

The generalization in (22) implies that contrary to what R&R93 assume, Th-prominence is part of the Binding Module. A first reaction to the generalization in (22) might be to say that the effects of chain formation are in competition with thematic prominence effects. A slightly different and, perhaps, more promising way of looking at (22), is to assume that theta prominence always plays a role but its effects are not always visible due to interference of Ch-Formation. This is the line we will pursue here.

If Th­prominence is part of Reflexivity, then it cannot be part of syntax proper. Given the specific formulations of R&R93 the only other option available is to assume that it is operative at a semantic level, i.e. it constraints the relations among the semantic arguments of a predicate. This is most explicitly formulated in an early definition of the notions 'reflexive' and 'reflexive marked' in R&R 1989:[5]

(23) a. A predicate is reflexive if and only if two arguments on its theta-grid are identified.

b. A predicate (of P) is reflexive-marked iff either (i) P is lexically specified as <x,x> or (ii) one of P's arguments is a SELF-anaphor

c. SELF-anaphors are operators on the thematic grid of P such that a SELF-marked position on the thematic grid of a predicate is identified with a position on that grid:

[V(<x,y>) [ .. SELF(<x,x>) ]] => [SELF+V(<x,x>) [..]]

Given the formulation in (23) the fact that thematic prominence is a well-formedness condition on reflexivization becomes straightforward once we accept that thematic grids are ordered and either directly or indirectly reflect thematic prominence:

(24) If an anaphor a, a an argument of P, is coindexed with b, b an argument of P, b must be more prominent than a on the argument structure of P.

Under (24) Greek behaves as expected. English, however, does not, since thematic restrictions can be overridden (cf. the discussion in section 1). Examples like (1) vs (6) indicate that in English, the thematic restrictions can only be overridden whenever there is Chain Formation and the condition on Chain Formation is not violated. Note, however, that in many such cases the judgements reported in the literature are contradictory. We already mentioned the examples (6b) and (9a), here repeated, but other cases have been mentioned in the literature (25):

(6b) ÷/?? The boys_i concern each other_i

(9a) ÷/?* John_i was given a book by himself_i

(25) ÷/* I mentioned Ringo_i to himself_i (Williams 1994, Jackendoff
1992)

It thus appears that one should describe the cases where Chain Formation and the condition in (24) interact as follows:

(26) a. If the Chain Condition is violated - irrespective of the thematic constraint (24) - then the sentence is ungrammatical. (cf. 18c, 18b)

b. If the Chain Condition is (vacuously) not violated and the thematic constraint (24) is observed the sentence is grammatical. (cf. 1a-3a, 8a-b, 11a-b, 12a-14a, 15b, 18a)

c. If the Chain Condition is not violated and the thematic constraint (24) is violated judgments are wavering. Only under appropriate contextual conditions the effect of the thematic constraint (24) may be lifted. (cf. 2b-3b, 5-7)

d. If the Chain Condition is vacuously not violated and the thematic constraint (24) is violated the sentence is ungrammatical. (cf. 1b, 11c-e, 12b-14b, 15a,c)

What we have suggested so far can be summarized more precisely as in (27), re-interpreting R&R's notion of anaphor and viewing Ch-Formation as an interpretive/licensing device (cf. Reuland 1996):

(27) a. a is an Incomplete Expression (anaphor) iff either (i) a is underspecified for phi-features, or (ii) a contains a morpheme with an `incomplete semantics'.

b. An Incomplete Expression must be interpreted/licensed.

c. Interpretation/licensing is established by Ch-Formation or Condition B.

d. The Th-prominence constraint on anaphoric interpretation is a well-formedness condition at the semantic interface.

e. An Incomplete Expression not interpreted in the computational system (through Ch-Formation) can be licensed at the semantic interface (i) by Condition B when in argument position, or (ii) as a logophor when not in an argument position (Reuland 1996).

From (27d,e) it follows that thematic effects will only be visible in local binding contexts if Ch-Formation is not operative. Greek is such a case as was illustrated in section 3. In English the effect of Ch-Formation will override any thematic effects, unless Ch-Formation does not apply, as is the case in the following examples:

(1b) *I talked about Mary to herself_i

(28) *An offer was made to Mary by herself_i Postal (1971)