Milena Milojevic-Sheppard Ljubljana CDU 802.0-25-563.2 PRONOUNS A~ THE PROBLEM OF REFERENCE IN TRANSFORMA TIONAL GRAMMAR The problem of reference is one of the main problems in transformational gram- mar. It does not concern only the pronominalization and reflexivization rules (both transformational and interpretive) but also many other rules, such as, for instance, deletion rules, movement rules, etc. Inspite of its importance for trans- formational grammar as a whole the problem of reference is stil! largely unsol- ved. Attempts to deal with it have been only partially successful; severa! points still have to be explained before a complete solution can be found. In what follows a critical survey of some approaches to the problem of referen- ce is given. l. The Referential Index Approach l. O This approach, commonly known as the Index Approach, was first put for- ward by Chomsky (1965). He suggests that ". . . certain lexical items are designated as referential .•• nl and then proposes that "... by a general conven- tion each occurrence of a referential item is assigned a marker2, say an inte- ger, as a feature". 3 In this way reference is marked in the deep structure (DS) by means of lexical features. The Index Approach has been most frequently used in the transformational_ rules of pronominalization, reflexivization, Equi-NP dele- tion and relative clause formation. l. l. Let us now examine in greater detail how the Index Approach can work within the transformational theory of pronouns and reflexives. According to the transformational theory anaphoric pronouns and reflexives are not generated in the DS but are derived transformationally from underlying noun phrases (NP' s). Reference is marked in the DS by means of indices on NP' s. The semantic component will interpret two NP' s as being coreferential if, and only if, they have been assigned the same index in the DS. For instance consider (1): (1) Mary said that Mary was tired. NP1 NP2 Let the referential indices be !. and j. If NP1 is assigned the referential index ..!. and NP2 is assigned the same re- ferential index i, then NPi is coreferential with NP2. PRONOMINALIZATION therefore takes -place, generating (2): (2) Mary said that she was tired. The semantic component then interprets Mary and ~ as having the same reference. 81 With respect to REFLEXIVIZATION the Index Approach works in a similar way, the only difference being that there is an additional condition, besides the coreferentiality condition: NPi_ and NP2 must be in the same simple sen- tence. Thus, for instance, (3) is derived from (4): ( 3) John hurt himself., (4) Jo~ hurt Johni. NP1 is coreferential with NP2 NP1 and NP2 are in the same simplex S NP1 NP2 REFLEXIVIZATION ==-- (3) l. 2 On first sight it would seem that the Index Approach can successfully deal with the problem of reference, at least as far PRONOMINALIZATION and RE- FLEXIVIZATION are concerned. However, there exist a number of cases for which this approach proves to be inadequate. l. 2.1 . Lakoff (1968) discussing the problem of reference at considerable length, presents severa! cases which show the inadequacy of the Index Approach on both syntactic and semantic grounds. l. 2. l. 1 Lakoff' s first counterexample to the Index Approach involves what he calls "the participant - observer distinction". According to Lakoff sentences like ( 5) have two different readings. (5) John dreamed that he was robbing the bank. On the first reading (the participant reading) John dreamed that he was ac- tually taking part in the bank robbery, whereas on the second reading (the observer reading) John dreamed that he was observing himself committing the bank robbery. The participant-observer distinction, hidden at first sight in sentence (5), can be perceived much more readily in sentences where it corresponds to a syntactic distinction, as for instance in (6) and (7): (6) John imagined robbing the bank. (participant) (7) John imagined himself robbing the bank. (observer) (8) Lakoff claims that sentences such as (6) and (7) cannot be accounted for by means of the Index Approach. Both (6) and (7) have the same DS (8): /s"' NPl VP 1 / ........... John v NP 1 1 imagine s /"' NP2 VP 1 /~ John V NP 1 ~ rob the bank 82 In order to derive (6) EQUI-NP-DELETION must be applied. However, this can only be done if NP1 is identical to the participant NP2 but not to the observer NP2. In the derivation of (7) REFLEXIVIZATION is applied after SUBJECT RAISING has made the subject of rob the superficial object of imagine. To make the application of REFLEXIVIZATION possible, NP1 must be conside- red identical to the observer NP2 • Within the Index Approach the identity of NP' s can be marked, but there is no way of indicating whether an NP is identical with a participant NP or whether it is identical with an observer NP. The cases discussed above are obviously. beyond the scope of the Index Approach. l. 2.1. 2 Lakoff mentions one further class of counterexamples to the Index Approach. These counterexamples involve a curious occurrence of ungrammati- cality in. pairs of sentences like the following: (9) Mary wants to buy a car and she wants to drive it. (10) *Mary wants to buy a car. and she will drive it. (9) is derived from (11) and (10) from (12): (11) Maryj wants to buy a cari and Maryj wants to drive a caq. ~ ~ (12) Maryj wants to buy a cari and Maryj will drive a cari. NP1 NP2 In the DS' s of both (9) and {10) the coreferentiality condition on NP]_ and NP2 is met and PRONOMINALIZATION can therefore apply, replacing NP2 with a corresponding pronoun it. Yet (10) is tingrammatical. If the Index Approach is correct then why is this so? l. 2. 2 The Index Approach obviously does not provide a satisfactory solution of the problem of reference. Chomsky (1965) himself points out that ". . . inte- resting problems arise when the referential items are plural". 4 Unfortunately Chomsky does not give any examples to show what these "interesting problems" might be. That plural NP' s pose difficulties for the Index Approach has been noted also by Partee (1973), again with no examples given. In addition to plural NP' s she considers that quantified NP' s can cause problems for the Index Appro- ach, too. She does provide some pairs of sentences like (13) and (14) but gives no explanation as to why such sentences give rise to difficulties. (13) Every philosopher · argues with himself. (14) Every philosopher argues with every philosopher. } [Partee (44a, b)J A possible explanation can be obtained by considering the DS' s of the two sentences. It turns out that both (13) and (14) have the same DS, (15): 83 (15) ~s..._______ VP NP1 /~ every philosopher 1·~ v PP / -----argue with NP2 L·~ every philosopher The main difficulty lies in the fact that in order to cierive (13) . NP1 and NP2 would have to be marked coreferential, whereas for the derivation of (14) they would have to be marked non-coreferential. As the DS is the same for both sentences it is not at all clear how the referential index system could mark the same two NP' s once as coreferential and once as non-core- ferential. l. 2. 3 The next class of problematic cases to be dealt with involves sentences like (16): (16) The girl hurt her. (17) In (16} the !@:! and her are without doubt non-coreferential. The DS of (16) is (17): s ----- ----NP1 1 VP /~ the girl V NP2 1 1 hurt the girl The referential indices of the two NP' s have to be identical if PRONOMINA- LIZATION is to apply. But, as has been pointed out above, the two NP' s are non-coreferential. So sentences like (16) cannot be derived at all within a theory of pronouns and reflexives based on the Jndex Approach. And yet (16) is perfectly grammatical and belongs to one of the most common types of EngliSh sentences. 2. The Jnterpretive Approach 2. O According to the interpretive theory pronouns and reflexives are generated in the DS and their reference is inarked by semantic rules of interpretation. The 84 need to use the Index Approach is thus avoided but this does not mean to say that a semantic reference determining mechanism proposed by the interpretive theory of pronouns and reflexives could fully cope with the problem of reference. 2. 1 Let us now take the examples put forward by Lakoff (1968) again in order to see whether they can be solved by means of the Interpretive Approach. First consider (5): (5) John dreamed that he was robbing the bank. The pronoun he is now present in the DS of (5). John and he can optionally be interpretedas coreferential since (5) satisfies the conditions for the ope- ration of the interpretive pronominalization rule. 5 However, there seems to be no way for the semantic reference determining mechanism to be able to mark the participant - observer distinction; the Interpretive Approach ob- viously cannot handle cases like (5) any better than the Index Approach. The Index Approach failed to account for the difference between (6) and (7). (6) John imagined robbing the bank. (participant) (7) John imagined himself robbing the bank. (observer) The Iriterpretive Approach on the other hand, is not able to relate (6) and (7) at all. Although (6) and (7) if taken individually, do not present any prob- lems for the Index Approach, the latter cannot be considered satisfactory since it fails to perceive the relationship which is intuitively felt to exist between these two sentences. According to the interpretive theory of pro- nouns and reflexives (6) and (7) have different DS' s; in the DS of (7) the lexical item under the node NP2 has the feature ( + REFL) , whereas in the DS of (6) the lexical item under the same node has the feature ( + PRO) Since the DS of (6) and (7) differ, the derived sentences, (6) and (7), also differ. Thus the distinction between (6) and ( 7) has been shown by the use of the Interpretive Approach, but not in a satisfactory way. This is because the difference shown between (6) and (7) is not based on the participant-ob- server distinction but instead on a distinction in the DS' s of two sentences, thus completely obscuring the relationship which exists between (6) and (7). As has been shown by Lakoff (1968) the ungrammaticality of sentences such as (10) cannot be explained within the transformational theory of pronouns and reflexives based on the Index Approach. No reason seems to be given for it within the interpretive theory either. (9) Mary wants to buy a car and she wants to drive it. (10) *Mary wants to buy a car and she will drive it. The conditions which must be fulfilled for the interpretive pronominalization rule to be able to apply are satisfied in both (9) and (10). Therefore a car and it can be marked coreferential in both sentences. The reason for the ungrammaticality of (10) remains a mystery. 85 2. 2 Sentences such as (13) and (14) can be accounted for within the interpretive theory if taken individually, but like in cases (6) and (7) the relation existing between (13) and (14) is not perceived. The problem of how to account for pairs of sentences which are clearly related obviously cannot be solved by simply ignoring the fact that the two sentences in a pair are related. 2. 3 Cases like (16) can be handled by the Interpretive Approach but only if the interpretive reflexive rule6 is modified (Sheppard (1974)), as follows: If NP1 and NP2 are in the same simplex S, then NP2 can be coreferential with NP1 if and only if it is reflexive. OBLIGATORY According to this rule NP1 (the girl) and NP2 (her) can be only marked non-co- referential. We have seen that inspite of the fact that both existing theories of pronouns and reflexives, the transformational theory and the interpretive theory, involve the notion of reference neit4er of them can cope with ali problems connected with it. For this reason severa! grammarians have been led to think that perhaps some other notion, instead of reference, is needed if a satisfactory treatment of pro- nominalization and reflcxivizat10n is to be obtained. Two approaches adopting a notion other than reference now follow. 3. The Counterpart Approach 3.0 When discussing cases such as (5) - (7) and (9) - (10) Lakoff (1968) pro- poses that the notion of coreferentiality be replaced by that of counterpart. The notion of counterpart has been taken from a new form of moda! logic, developed by David Lewis 7. In contrast to the traditional forms of logic this new form lets two possible worlds be related in such a way that two entities in one world can correspond to one entity in the other. Lakoff suggests that this approach to lo- gic be employed in order to solve cases like (5) - (7) and (9) - (10) since these sentences involve more than one possible world. The fact that they do is the very reason why such sentences present problems to both the Index and the In- terpretive Approach. In (5) Jolin of the actual world in which he does his dreaming is split into two persons in the world of dream: John the participant and John the observer. Shematically the situation looks like this: John o / '>,, 86 ACTUAL WORLD WORLD OF DREAM The problem of why ungrammaticality occurs in cases like (10) can also be ex- plained in terms of two possible worlds. In (9) a car exists in the world of Mary' s desires which is defined by Mary wants. The same world is involved in the second part of (9) since the latter again contains Mary wants. It follows that the car bought by Mary exists in this second possible world as well as in the first one. In other words, the car in the first half of the sentence has a counterpart in the second half of the sentence, so the pronoun it can be used when referring back to the car in the first half of the sentence-:- In (10), however, two different possible worlds are related: the world of Mary' s desires, in which the car bought by Mary exists, and the actual world, defined by will, in which the car Mary wants to buy does not exist. Therefore the car in the first half of the sentence does not have a counterpart in the second half the sentence, and consequently, the car in the first half of the sentence cannot be referred back to with the pronoun it. If it is, then ungrammaticality results, as in (10). To demonstrate the usability of the Counterpart Approach, Lakoff considers some further examples which are unexplainable both within the transformational theory and within the interpretive theory of pronouns and reflexives: (9a) Mary wants to buy a car and she intends to drive it. (lOa) *Mary intends to buy a car and she wants to drive it. According to Lakoff the car in the world of Mary' s desires has a counter- part in the world of Mary' s intentions, but not vice versa. This would ex- plain the ungrammaticality of (lOa). Thus the entities that exist in the world of one' s desires automatically exist in the world of one' s intentions, but not vice versa. 8 On the basis of such cases, Lakoff concludes that the notion of counterpart, and not the notion of coreferentiality is relevant for pronominalization. 4. The Forma! Identity Approach 4. O Another approach to pronominalization which tries to avoid the notion of coreferentiality has been presented by Partee (1973). According to her pronomi- nalization involves two processes: during the first process a repeated N in an either definite or indefinite NP is reduced to the proform one( s) and during the second the proform one(s) deleted after certain determiners. The referential noun identity is not relevanf for either N-REDUCTION TO ONE(S) or for sub- sequent DELETION of ONE (S). The only requirement to be fulfilled is forma! noun identity. 4. 1 Anaphoric personal pronouns are assumed to be derived by the reduction of a repeated the N to the one(s) followed by the deletion of one(s), leaving only the definite pronoun the under the NP node. Therefore anaphoric personal pro- nouns are considered to be the suppletive forms of the. 9 87 4. 1.1 Since personal pronouns are derived only from definite NP' s there must be some process introducing definite articles which precedes what Partee calls "pronominalization proper". Definite articles are introduced by DEFINITIVIZA- TION. Partee points out that, unlike in the case of PRONOMINALIZATION, in the case of DEFINITIVIZATION the notion of coreferentiality plays an important role. Another process which, according to Partee, precedes PRONOMINALIZATION i.s REFLEXIVIZATION. Here, once again, the problem of reference cannot be avoi- ded; the application of REFLEXIVIZATION requires coreferentiality between the antecedent and the reflexivized NP.10 4. l. 2 According to the Formal Identity Approach pronouns can be generated both transformationally from underlying NP' s by the processes described above and also in the DS, i. e. from underlying the one( s). The first way of derivation corresponds to the transformational theory and the second to the interpretive theory. The only difference between the interpretive theory and the system suggested by Partee is that within the latter only the DS determiner the and the proform one(s) are present in DS and not all pronouns as is stated by the interpretive theory. 4.2 However, even the "combined.'1 system proposed by Partee cannot handle the problematic cases presented in l. 2, p. 2. 4. 2.1 Sentences such as (16) clearly have a non-ambignous non-coreferential reading, but according to Partee' s system (16) can be derived in two di.fferent ways: (18) The girl hurt the one. (19) The girl hurt the girl. In (19) the first the N is formally identical to the second the N, and yet, as mentioned above, they must not be considered coreferential. However, the condition of formal N - identity has been met, and since this is the only requirement for the application of N-REDUCTION TO ONE(S) the latter process takes place. Two non-coreferential NP' s are not necessarily always accompanied by formally different modifiers, as can be seen in (19). Besides ~ all that is left in the second NP after N-REDUCTION TO ONE(S) has taken place is the, and therefore there is no way within the Forma! Identity Approach by means of which the conversion of the one into her could be prevented. During the derivation of (16) the girl will thus be automatically turned into her and consequently coreferentiality between her and the prece- ding NP will be implied, inspite of the fact that in reality cases like (16) cannot have a coreferential reading. 4. 2. 2 Sentences such as (13) can be accounted for by the Formal Identity Approach since according to this approach the identity condition is not on the whole NP. Thus (13) can be derived from (20). (13) Every philosopher argues with himself. 88 (20) Every philosopher argues with the philosopher. Sentence (14) (14) Every philosopher argues with every philosopher, however, presents the same difficulty for the Forma! Identity Approach as sentence (19). The difficulty involves the false implication of coreferentiality. 4. 2. 3 Like the Index Approach and the Interpretive Approach, the Forma! Identity Approach fails to account for pairs of sentences such as (9) and (10). (9) Mary wants to buy a car and she wants to drive it. (10) *Mary wants to buy a car and she will drive it. Whereas the Index and the Interpretive Approach can account for the gram- maticality of (9) but cannot account for the ungrammaticality of (10), the si- tuation here is merely reversed. The ungrammaticality of (10) can be explained since within the Forma! Iden- tity Approach the assumption has been made that an indefinite NP can be the antecedent only if it is ( +specific ) and not if it is ( -specific) . A car in (10) is < -specific ) and this is th~ reason for the ungrammatica- lity of this sentence. However, if this is correct, then how is it possible for (9) to be grammatical inspite of the fact thaf it includes the same ( -specific NP } , ~? Since within the Forma! Identity Approach the grammaticality of (9) cannot be accounted for, the relationship between the two sentences once again remains unexplained. 4. 2. 4 Finally, there seems to be no way in which cases involving participant- -observer distinction such as (5) - (7) can be handled by the Forma! Identity Approach. 5. Conclusion 5. O During the discussion of the reference problem in l. - 4. above the follow- ing two facts have become obvious: - Although both the transformational theory and the interpreti ve theory of. pro- nouns and reflexives draw heavily on .the notion of reference neither of them can provide an entirely adequate system for its representation. - The importance (if any at all) of the role which reference plays with regard to pronominalization and reflexivization is by no means clear. 5.1.1 It has been shown that neither the Index nor the Interpretive Approach developed within the transformational theory and the interpretive theory respec- tively can handle cases like (4) - (7), (9) - (10), (13) - (14), (16). Sentences such as these occur frequently enough in standard English so they cannot be la- belled as exceptional in order to justify either of the two approaches mentioned above. 89 5.1. 2 Furthermore, there exists a fact which bears more general significance than individual cases and therefore casts serious doubt on the validity of both the Index and the Interpretive Approach. This fact has to do with the definition of the notion of reference. The problem is that so far it has never been expli- citly stated what exactly is meant by the terms "reference", "referential", "co- referential". When Chomsky (1965) put forward the Index Approach he was at the same time well aware of the fact that ". • . there are problems in specifying the noti on 'referential' properly". 11 Partee (1973) points out how difficult it is to define the notion of coreferentiality since ". . . in many cases the two coreferential NP' s do not refer to the same phy- sical object1112 as, for instance in (21): (21) My home used to be in Baltimore, but now it' s in Los Angeles. (Partee (13l)J. Like the Index Approach the Interpretive Approach involves a vague notion of reference which is no more explicitly defined than within the former. Thus with regard to the general question of what exactly "reference"· is there is no reason for preferring either one of the two approaches to the other. It is very difficult to imagine how an approach can have any hope of being satisfactory if the notion on which it is based is left unexplained. In parti- cular, how can an adequate system of representing reference be worked out if it is not clear what reference really is? · 5. 2 We have seen that the In is coreferential with NP1 if 1-either NP1 is to the left of NP2 or 2-NP2 is dominated by a clause subordinate to the clause immediately dominating. NP1• OPTIONAL (Jackendoff (1968), p.11.) NP2 can be coreferential with NP:J. if and only if it is reflexive. OBLIGATORY (Jackendoff (1968), p. 6.) See Lakoff (1968), p. 4. 94 8 9 10 11 12 13 However, it is debatable to what extent (lOa) is ungrammatical. One can cer- tainly argue that (lOa) is much less ungrammatical than (10). The reason for this is that in the latter two basically different worlds are concerned (the real world and the world of desires) whereas in the former, the two worlds invol- ved, the world of intentions and the world of desires, are not so widely sepa- rated This was first suggested by Postal (1966). The other two processes which must take place before PRONOMINALIZATION, FEATURE TRANSFER and SURFACE CASE MARKING, are irrelevant to the problem of reference and .therefore are not discussed here. Chomsky (1965), p.146. Partee (1973), p.202. Lakoff (1968), p. 7. 14 Cf. Lees and Klima (1963), Langacker (1966), Sheppard (1974). Povzetek ZAIMKI IN PROBLEM REFERENCE V TRANSFORMACIJSKI SLOVNICI Članek obravnava problem reference s stališča dveh osnovnih teorij o zaimkih v okviru transformacijske slovnice, transformacijske in interpretativne. Avtor prikazuje in poskuša kritično ovrednotiti štiri različne pristope k temu proble- mu: pristop referencialnih indeksov, ki se je uveljavil v transformacijski teoriji o zaimkih, interpretativni pristop, ki tvori osnovo interpretativne teorije, t. i. "kontrapar" pristop in formalni pristop. Pristop referencialnih indeksov temelji na predpostavki, da je referenca označe­ na v globinski strukturi stavkov s pomočjo referencialnih obeležij določenih lek- sikalnih enot. Po interpretativnem pristopu pa je referenca semantičen koncept in se določa s pomočjo semantičnih pravil interpretacije. Analiza teh dveh pri- stopov, ki sta v transformacijski slovnici sicer najbolj razširjena, je razkrila številne slabe strani tako prvega kot drugega pristopa. Pri tem ne gre le za po- samezne konkretne primere pronominalizacije in refleksivizacije, ampak tudi za pomembno pomanjkljivost splošnejšega značaja. Niti pristop referencia!nih indek- sov, niti interpretativni pristop namreč ne vsebuje točne definicije pojma referen- ce. Natančna opredelitev določenega pojma pa je vsekakor nujno potrebna, če naj uspešno rešujemo probleme, ki nastajajo v zvezi s tem pojmom. Pomanjkljivosti dveh zgoraj omenjenih pristopov so dovolj resne, da nastane vprašanje ali je pojem reference sploh relevanten za adekvatno teorijo o zaim- kih. Da bi odgovorili na to vprašanje, smo analizirali dva pristopa, ki povsem 95 zanikata pomen reference: pristop, v katerem je pojem reference nadomestil po- jem "kontrapar" ("counterpart") in formalni pristop, po katerem je pogoj za pro- nominalizacijo in refleksivizacijo le formalna, ne pa tudi referencialna identič­ nost imenskih skupin. Ugotovili smo, da t. i. "kontrapar" pristop zaenkrat pred- stavlja le parcialen, čeprav obetajoč, poskus, da bi problem reference rešili z uvedbo nekega novega pojma. Formalni pristop pa tak kakršen je, ni sprejemljiv. Lahko torej zaključimo, da noben izmed obstoječih pristopov k problemu refe- rence ni povsem zadovoljiv. Možnosti za ustreznejše reševanje tega problema vidimo bodisi v izpopolnjenem "kontrapar" pristopu, bodisi v novem pristopu, ki ga predlagamo v tem članku in ki upošteva tako formalno kot referencialno identičnost imenskih skupin. Adekvatna teorija o zaimkih namreč mora vključe­ vati pojem reference oziroma odgovarjajoč nadomesten pojem, in ne more te- meljiti izključno na formalni jezikovni strukturi. 96