¡Ethical Dative and Possessor Omission Sí, Possessor Ascension No!

[This paper was originally issued in 1980, in the Workpapers of the Summer Institute of Linguistics, University of North Dakota Session XXIV 97-141. This version for the Internet, prepared in 2005, contains, at least in intent, the original text unchanged except for typographical adjustments, a few corrections, and redrawing of the figures.

An updated statement of the central argument of this paper was published in 1985 (and is available on this website). Nevertheless this paper has been cited fairly often and I am occasionally still asked for copies of it. It also has detailed argument and data presentations that are not in the 1985 paper. It is for such reasons that I have decided to try and make it easily available. I do recommend reading the 1985 paper for what is to me a more satisfying and complete vision of the phenomena in question. Basically, in this paper the argumentation is all in an either/or mode; the 1985 paper expresses a both-and synthesis that was (and is) more satisfying to me.]

0. Introduction; PA-type sentences

There is a very common type of Spanish sentence which has the following properties: (i) One of the arguments of the verb is a noun phrase which consists of an article (almost always a definite article) and a noun. (ii) There is also a dative pronoun dependent on the verb, and (iii) the dative pronoun is understood as the possessor of the definite noun. (1)-(3) are examples of this kind of sentence.¹

It has been suggested that sentences like (1)-(3) should be accounted for under the theory of Relational Grammar by a relational configuration called Possessor Ascension (PA).² In this structure the possessor in a possessor-head construction is a non-initial indirect object (“3”) in the same clause in which the possessor-head construction bears an initial grammatical relation (GR). The relational network (RN) which defines PA is given in figure 1, along with the RN involving PA which would be used for sentences (1)-(3).

0.2 Arguments A and B for PA in Spanish

Two main arguments have been given to support a PA analysis for Spanish.³ They are as follows:

• Argument A: PA has already been posited, and argued for against at least some reasonable alternatives, in other languages (e.g. Chamorro, French, Georgian, Southern Tiwa, Tzotzil).⁴ Thus it is independently motivated as a universally possible configuration. Using it to account for the Spanish data is therefore more parsimonious; there is no need to posit a new kind of RN.

• Argument B: Positing structures with PA allows one to reflect the similarity in meaning between sentences in languages with PA (e.g. the Spanish sentences given above) and sentences in languages without PA (e.g. the English translations of those sentences), where the Possessor remains as Possessor.

I would like to argue that PA is not the best way to account for sentences (1)-(3). They are better viewed as resulting from a structure with an “ethical dative” (ED) and what we will call “Possessor Omission” (PO), both of which can be independently motivated in Spanish.

1. PA is not necessary in Spanish

1.1 In answer to argument A

Argument A is strong only if there exist no other universally available structures which will account for (1) to (3) and other such Spanish sentences. Such is not the case. Positing PA allows us to account for two facts: (i) the presence of the nominal understood as the possessor as a dative in the clause, and (ii) its absence as an overtly marked possessor. Both of these facts can be accounted for by independently needed mechanisms in Spanish. We will take them up in reverse order.

1.1.1 Possessor Omission

1.1.1.1 PO and PD

Spanish in many constructions besides the PA-type construction exemplified in (1)-(3) permits a nominal which is understood as possessed to appear with no overt possessor. Sometimes the nominal which is understood to be the possessor will appear in the same clause with the nominal which it is understood to possess. In sentences (4) and (5) such constructions are illustrated with the understood possessor as subject (“1”) and as direct object (“2”) respectively.

English speakers learning Spanish are likely to ask, when first confronted with sentences like (4), whose hand is being referred to. Spanish speakers know that, 99 times out of 100, it is the hand of the subject of the sentence. Sentences like (5) cause English speakers no trouble, because English has a structure similar to the Spanish one. Yet a similar question would be perfectly reasonable: whose face is being referred to? In fact a speaker of a language like Aztec where the face must be obligatorily marked as possessed in such a construction would likely be puzzled on just that point. But English and Spanish speakers both understand that it is the face of the direct object that is being referred to.

Two possibilities suggest themselves for representing sentences like (4) and (5) by means of RN’s. One is, in the spirit of Argument B above, to have an initial Poss arc, with the nominal understood as Possessor multiattached, being also the head of the 1- or 2-arc of the main clause. The Poss arc will then be ignored or treated however Equi-victims are treated under the theory.⁵ We can call this approach Possessor Deletion (PD). The PD proposal is represented by the RN’s in figure 2. The other possibility is to omit the Poss arc entirely, leaving the nominal understood as possessor represented only as head of the 1- or 2-arc. We can call this approach Possessor Omission (PO). The PO proposal is represented by the RN’s in figure 3.

The PD and PO approaches differ over whether the conception of a possessive relation is to be represented syntactically (the PD model) or only at some semantic or conceptual level (the PO model). The difference between them, while important, is not crucial to the argumentation here; however, data and argumentation relevant to deciding between them is presented in Sections 1.1.1.4, 1.2.1, and 2.3.

1.1.1.2 PA will not work

A third approach might be to try to account for sentences (4)-(5) by PA. Assuming that PA advances possessors to become non-initial 3's, as it must to account for (1)-(3), this approach would predict a 3 in the clause, and some device would be needed to syntactically delete those 3’s so they would not show up as dative pronouns. This device (3-Deletion) would presumably delete 3’s under coreferentiality with a 1 or a 2, but it would have to be constrained to delete only 3’s which are produced by PA; other 3’s would not be deleted. Up to this point this approach might seem to be on a par with PO or PD; PA may be independently motivated from (1)-(3), and it is not clear that a rule deleting 3’s which have been produced by PA is any more complex or otherwise less desirable than one deleting or omitting possessors.

However, there are also cases in which a PA-type structure like those in (1)-(3) occurs, but in which the DAT is coreferential with the 1. For instance, (6);

Thus it would not be true that all 3’s produced by PA and coreferential with the subject are or even can be deleted. For cases like (6), 3-Deletion would have to be constrained somehow not to apply. On the other hand, if either PO or PD is what is going on, it does not have to be so constrained, but can be used to explain the absence of the possessor in sentences like (6) as well as in sentences like (4) and (5). (It is true that models with PO and PD will have to give an explanation for why there is a dative in (6) but not in (4); such an explanation is given in Section 1.1.2).

Another fact that militates against the idea of using PA to account for these sentences is that in sentences like (4) the 1 is not necessarily to be understood as the possessor. If someone from an anatomy lab were carrying around someone else’s hand and raised it, (4) would be appropriate to describe the situation; it would have to be translated in that case as He raised the hand. PA and 3-Deletion would not be able to take care of such cases because there would be no coreferential nominal to trigger 3-Deletion. (Notice that PD would be involved in a similar problem; cf. Section 1.1.1.4.) Thus PO would be needed anyway to account for the reading of (4) where the 1 is not the Possessor. If it is needed for that case, it is more parsimonious to let it also account for the other reading of (4) and for (5) and (6), rather than to posit PA and 3-Deletion to account for them, for then 3-Deletion will not be needed. Using PO alone is simpler than the alternative, which uses PA, 3-Deletion, and PO, and it is therefore preferable.

1.1.1.3 PO or PD with a coreferential 3

Sentences (7) and (8) give examples of a similar construction in which the understood Possessor is the (final) 3 of the clause.⁶

(7), in the English translation as well as in the Spanish, is ambiguous or vague: It is not clear whose son is being referred to, though it is almost certain that it is either the son of the 1 or the son of the 3. In the Spanish sentence it is not clear who is the possessor because there is no possessor marking; in the English sentence it is because the 3 p. sg. possessor marking could bear an anaphoric relationship to either the 1 or the 3, since both are 3 p. sg. It is also possible, though unusual, for the possessor to be understood as some other 3 p. sg. nominal (masculine in English). (8) also is ambiguous (or vague) in Spanish as to whether the son of the 1 or the son of the 3 is referred to, though in English the difference in person of the possessor disambiguates the two senses, since it is overtly marked. (Once again, the Spanish sentence or the English sentence with his son could be interpreted with some other 3 p. sg. nominal as the possessor.) Thus under PD (8) would result from either of the two RN’s in figure 4. Under PO both versions of (8) would have the RN given in figure 5.

These sentences also should not be accounted for by PA. In the first place, if they were from PA there would be no explanation of the fact that the DAT is understood as goal; sentence (8) does not mean He sent my son. Also there would be no explanation of why the person referred to by the DAT need not be the possessor, but someone else may be understood as possessor (here, either the 1 or some other 3 p. sg. nominal). In other words, PA would need goal-DAT and PO anyway to account for these sentences, and once you have goal-DAT and PO you do not need PA. Crucially then, as was the case with (4) and (5), sentences (7) and (8), on the reading in which the son is the son of the Indirect Object, will need RN’s involving a device like PD or PO. Given such a device, and given an explanation for the datives in sentences like (1)-(3) (which will be offered in Section 1.1.2), the absence of an overtly marked possessor in sentences (1)-(3) can be accounted for without PA.

1.1.1.4 ¡PO sí, PD no!

While the difference between PD and PO is not crucial here, I would like to present a couple of considerations which make me think that PO is preferable to PD. The first is that one would expect the deletion rule to act like other syntactic deletion rules such as Equi-NP Deletion in having a coreferential trigger that commands the nominal to be deleted.⁷ If we assume that PD must have such a trigger to operate (and that PO need not), we can argue against PD in favor of PO. Under that assumption the readings of (4), (7), and (8) on which the possessor is not understood to be coreferential to any nominal in the main clause could not be accounted for by PD. Neither could sentences such as the following:

In (9) (from Ricardo Palma) the nominal which is understood as the possessor of marido does not appear in the same clause as the possessed nominal at all, but only in a conjoined clause (in the pronominal shape la) and embedded way down in another clause which is sister to the clause in question, this time referred to as su consorte. In neither occurrence does that nominal command the possessed nominal. Therefore, if PD requires that the trigger command the target, it cannot account for this sentence.

There is no trigger in sentence (10) to cause the possessor of cuernos to be deleted. The nearest available trigger is un toro bravo, in the preceding sentence. Yet, although the bull is clearly understood as the possessor, it is not marked as such or even represented overtly in the sentence at all. Again, if PD requires a trigger which commands its target, or even a trigger at all in the same sentence, it cannot account for sentences like (10).

In fact, it is quite possible to find sentences in which there is no overt trigger at all in the linguistic context. For instance, sentences like (11):

and many sentences in which possession is clear from the non-linguistic speech situation (see the discussion of (12) and (13) ahead).

Thus, if PD depends on a trigger commanding the nominal to be deleted, it cannot account for many sentences in which possession is understood even though the possessor is not overtly marked. Something like PO would thus be needed besides. But PO can account for these S’s and also those like (4)-(8) which would motivate PD in the first place. PD needs PO, but PO does not need PD. Thus a theory with only PO is simpler than a theory with both, and therefore preferable.

Notice that all these sentences are even less susceptible to analysis by PA than by PD: PA would predict a 3 in the clause with the nominal understood as possessed. Some totally weird and ad hoc mechanism would be needed to delete those 3’s while leaving many other 3’s produced by PA alone.

Another consideration that makes me prefer the omission concept over the deletion concept is this: There are many sentences in both English and Spanish (and probably every other language in the world) in which no possession is overtly marked but in which possession is clearly understood from the speech situation. For instance in sentence (12)

the fridge may be understood in a given speech situation to be my fridge, your fridge, or Fred’s or Harriet’s or Herman’s. I do not think it really proper to include a Poss arc in the initial syntactic structure of sentences like (12). In fact, I do not think such a relationship is properly represented even in the semantic structure of such a sentence, though it would be in the non-linguistic conceptual structure.⁸ Thus (12) involves PO rather than PD. Yet I do not think that it will be possible to draw a hard and fast line between cases like this and cases of the sort discussed above. The possessive relationship will have different degrees of salience and the identity of the possessor will be identifiable to different degrees throughout a whole continuum of expressions and situations. Some English examples are given in (13).

All of these sentences have a noun phrase of the form ‘the noun’. A relationship of possession is at least very probable in each of them; so probable as to be certain in the last ones, less probable in the earlier ones. Where does one draw the line? Why does one need to draw the line? Why can one not simply say that even when possession is clearly perceived to be present, one need not necessarily specify it linguistically? That the cases in which it must be specified are determined by each language and cannot be universally predicted? That in a language like Spanish, a nominal that is conceived of as possessed may simply be coded as definite (contextually unique), whereas in a language like English the parallel nominal may be required to be coded as well as conceived of as possessed?

The argument, then, is a sort of reductio ad absurdum based on the assumption that it is improper to account for the absence of a possessor in sentences like (12) by syntactic means. If sentences like (1)-(8) (including the English translation of (5)) are to be accounted for by PD (or PA) and not PO (i.e., by syntax and not by the interface between conceptualization and linguistic coding), then, unless someone can come up with a principled way to determine where to draw the line, so should sentences like those in (13) and (12). Which is absurd.

1.1.2 Ethical Datives (ED’s)

In the previous section it was argued that if an accounting could be given for the datives in sentences like (1)-(3), either PO or PD would permit us to account for those sentences without recourse to PA. In this section I will attempt to account for those datives.

There is, in Spanish, a class of nominals marked as datives which have a meaning something like “person affected intimately (and usually adversely) by the action or state predicated.”⁹ Often they can be translated into English by a prepositional phrase with on. (14b) gives an instance of such a nominal.

Let us refer to these nominals (following traditional terminology) as Ethical Datives (ED’s). I will assume that they bear on the final stratum a GR of 3: that will explain the fact that they are marked with the dative case. It is probable that they would not be considered to be initial 3’s under most relational analyses: they can occur with clear indirect objects, which, if both were initial 3’s, would violate the Stratal Uniqueness Law.¹⁰ In the absence of any clear indication as to what their initial GR would be, I will simply mark them as initially bearing GR_x. The RN for (14b) would be as in figure 6.¹¹

The meaning of these ED’s can be illustrated as follows. Sentences like (14b) are usually quite appropriate when talking about the deaths of one’s close relatives. Suppose that a man’s son dies; (14b) would normally be appropriate in describing the situation—men are normally affected intimately and adversely by the deaths of their offspring. However, if the father had disowned the son and was unaware of his whereabouts or of his death, then (14b) would be inappropriate. Similarly, (15a) would be totally inappropriate for me to say, but it would be appropriate for Tito’s doctor, who presumably would want Tito to live and would be adversely affected psychologically and/or professionally by his death. Similarly, (15b) would be inappropriate for Americans to say, but it would be quite appropriate for the Yugoslavs, especially if they loved Tito and were therefore psychologically hurt by his death, or if they felt endangered by his death.

It is possible to specify that the person who died in (14b) was the son of the person affected. The resultant sentence is (16).

The absence of the possessor in (16) should be explained by PO or PD, just as it was in the case of (8). RN’s for (16) under PO and PD are given in figure 6.

The final structure of (16) bears a crucial similarity to that of (1)-(3). (1)-(3) are transitive, and this is intransitive, but the important thing is that the final stratum has a noun phrase with the definite article, which is understood to be possessed by a person who is represented in the sentence by a dative pronoun. I claim that the structures of (1)-(3) and (16) are in fact exactly parallel, and that the datives in all those sentences are ED’s. They would have the structure given in figure 7 under PD, and that in figure 8 under PO.

In Section 1.1.3 it is shown that this contention is consistent with a number of facts in Spanish, and in Section 2 it is argued that the analysis which results is preferable over the PA analysis.

One final fact about (16): At least 90 percent of the time it will be understood as glossed: that the son is the son of the referent of the ED. However, the sentence may also be understood in certain contexts as referring to someone else’s son. For instance, if a doctor is treating a father and son for injuries from an accident, and the son dies, (16) would be appropriate in that situation with the ED referring to the doctor. Or if an anthropologist wants desperately to study certain familial interactions and the son of the only family that will do for his study dies, (16) would again be appropriate. In other words, where someone other than the parent of the son can be conceived of as affected by the death of the son qua son, the ED in (16) can be understood to refer to him. This same pattern holds true for other sentences with ED’s also.

1.1.3 The data do not demand PA; ED’s and PO will work

In the following sections I will present relevant data with which the ED-PO hypothesis is consistent. These same data will be used in Sections 2.2-2.5 to argue for this hypothesis against PA; here my purpose is merely to show it to be consistent with the facts of Spanish.

1.1.3.1 PA occurs only where the possessor can be viewed as affected

As far as I know, every sentence like (1)-(3) where PA would be posited can have the implication that the possessor is affected by the action of the predicate. This is quite apparent in the examples given. One is typically affected adversely by having one’s car dirtied, one’s money stolen, or one’s hand cut (off). Other sentences are quite conceivable in which the possessor will not be affected; in these PA cannot occur. For instance:¹²

Also, when the person understood as possessor is just recently dead (during the period when a dead man’s possessions are still considered his) sentences like (1)-(2) are quite inappropriate—versions with possessive pronouns are called for (sentences (18)-(19) below). This fits with the fact that a dead man is not affected by what happens to his possessions. Yet PA-type constructions can be used in reference to a dead person when the physical state of the corpse is changed; the corpse is thereby affected. Thus (3) is still appropriate; the dead man is affected by having his hand cut as much as a dead man can be affected by anything.

Thus whenever PA is posited, ED’s could occur, since the possessor can be viewed as affected. In any situation in which an ED could not occur, because the possessor is not affected, PA-type structures do not occur. This is entirely consistent with the theory that claims that these structures in fact have ED’s in them.

1.1.3.2 PA must occur where the possessor is clearly affected;
it need not occur where the possessor is not clearly affected

To the extent that the conceptual situations represented in (1)-(3) can be viewed as not connoting that the possessor is affected by what is predicated about his possession, those situations can also be represented by a non-PA-type structure with no dative and with an overt possessor marking. Thus (18)-(20), which parallel (1)-(3), are possible.

(18) tends to imply that the person was not affected by his car’s getting dirty. It would be most appropriate if he were absent when the heinous offense was perpetrated, or especially if someone else were using the car.

(19) is of un- or questionable grammaticality. I think this is because of the semantics of robar; like the English rob, it tends to denote a crime against a person (or a property-owning entity like a bank or the corner gas station) rather than against property per se. (Contrast with steal.) This victim is usually coded (appropriately enough) by an ED. The victim need not be coded at all, however; it is quite possible to say robaron tres mil pesos ‘they stole 3000 pesos’. But if the possessor of the money is brought into the picture at all it will be as the victim of the crime (coded by an ED) rather than as the possessor of the property taken.¹³ It is for this reason that (19) is anomalous.¹⁴

(20) will also be judged in vacuo by many Spanish speakers as ungrammatical. I believe that this is because of the difficulty of conceiving of a situation in which one is not affected by having one’s hand cut. That brings us up to the issue of “inalienable possessions”. This is a somewhat elastic class that at least involves as clear members body parts and clothing which is being worn. It is often assumed that PA-type constructions like those in (1)-(3) are obligatory with “inalienable possessions”. If this were true, (20) would be ungrammatical as a matter of course. It seems clear to me that the “inalienability” of possessions can be translated into the probability that the possessor will be affected by whatever happens to his possessions. Consider for instance (21)

(21a) is appropriate when the possessor is wearing the shoes (in which case he is almost certain to be affected by their being stepped on) and also in any other situation in which he is viewed as affected. For instance, in recounting a list of atrocities which A has committed against B, it would be quite apropos to include (21a) even if B's shoes were in the closet when A stepped on them. (21b) is not very appropriate if the shoes are being worn; it rather implies that the possessor was not affected by their being stepped on. However, it is quite appropriate when the shoes are sitting out in the middle of the floor or in the closet and someone steps on them, because in such a situation the person can easily be viewed as not affected by what happens to his shoes. In other words, whenever the possessor can be viewed as not affected by the action of stepping on the shoes, (21b) will be appropriate.

An even more interesting case is the sort of thing that happens with the most “inalienable” possessions of all—body parts. In at least two situations what happens to a body part can be viewed as not affecting its possessor. One is unconsciousness or inattention. Thus, (22b) is quite appropriate to say in talking about the procedure followed in an operation

(22a) is appropriate, because even an unconscious person can be viewed as affected by having his stomach cut open, but (22b) is also appropriate, because as long as he is unconscious he is not directly or clearly affected by it as he would be if he were conscious. Similar comments apply to (23a) and (23b).

The other case in which the possessor of a body part can be viewed as not affected by what happens to the body part is dismemberment. Thus if in an accident a person’s arm was cut off and was lying in the road getting run over by the cars, he might well say (24b), but hardly (24a).

(24a) and not (24b) would be appropriate if the cars smashed his arm while it was still attached to him; i.e., when he was still clearly affected by what happens to it. A very similar case is that of teeth. While they are still in one’s mouth, he is usually affected by what happens to them, but when they have been taken out, although they are still his teeth, he is not affected by what happens to them. Thus (25a) is appropriate to use when speaking to a dentist, but (25b) is much less appropriate in that situation, since the dental manner of examining teeth usually has (adverse) effects upon the patient. However, once the tooth has been extracted, and the patient is showing it to a fried, (25a) is quite inappropriate and (25b) is called for.

In exactly the same way, (20) will be appropriate in cases of unconsciousness or dismemberment, because the person can be viewed in those circumstances as not affected by what happens to his hand, and (3) will be inappropriate in cases of dismemberment, because the person can no longer be viewed as affected by what happens to his hand.

In sum, then, in situations where the possessor is perforce viewed as affected by what happens to his possession (including the cases usually subsumed under "inalienable possession"), the PA-type structure is required; the structure with the possessor overtly marked and with no dative is inappropriate. If the possessor can be optionally viewed as either affected or not affected, either type of structure can optionally be used. Once again, these facts are at least entirely consistent with a theory that claims that the datives in PA-type structures are in fact ED’s marking the person affected by the action or state predicated.

1.1.3.3 Ambiguous (or vague) possession

Sentences (1)-(3) were glossed with the DAT translated as a possessor. That is appropriate for most instances of those sentences. However, they may also be used in certain situations in which the DAT cannot be translated as a possessor. For example, (1) is quite appropriate when A’s car is the one that is dirtied and yet the DAT refers to B, as long as B is affected. The sentence would then have to be translated in English as They got the car dirty on him. Similarly (2) would be quite appropriate if it was A’s money but B was affected by the stealing; say he was carrying the money at the time. The sentence would be translated as They stole all the money from him, or They robbed him of all the money. In fact, (2) in the case where it is B’s money could be felicitously translated as They stole all his money from him. Even (3) can be construed with the hand belonging to A and the dative referring to B, as along as B is affected. If it is already known that B was carrying A’s hand around, trying to protect it, (3) would be appropriate with some translation such as They cut the hand on him. This construal is quite odd, but that is simply because it is odd to think of B as the person affected by A’s hand being cut.

Thus it is apparent that whenever a person other than the possessor can be viewed as affected by a predication relative to the possession, the dative in sentences like (1)-(3) can be interpreted as referring to the person. This parallels the case with ED’s noted in the last paragraph of Section 1.1.2. Notice too the parallel with the cases noted in Section 1.1 of PO where there are no ED’s. Once again, these facts are at the least consistent with the view that the datives in PA-type structures are in fact ED’s and that the possessors are simply omitted rather than ascended.

1.1.3.4 Clear possession by another

It is quite possible to get sentences like (1)-(3) with the construal by which the person represented by the dative is not the possessor, but in which the real possessor is overtly marked. Thus (26)-(28), which parallel (1)-(3).

(28) is quite odd, but again that is explainable by the fact that it is very atypical for B to be the person affected when A’s hand is cut.

Once again, this is not the sort of thing that one would expect from PA, but it is not at all surprising given the analysis of (1)-(3) as having ED’s and PO.

1.1.3.5 Possessive pronouns on preverbal subjects

For most speakers the possessor must be omitted (or deleted) in sentences like (4), (5), (7), (8) and (16), where PO (or PD) is clearly motivated. For some speakers, however, the added proviso must be made that PO is often or even usually suppressed when the possessed nominal is a preverbal subject. (Actually, it may be that the important fact is that such a subject precedes the other occurrence of the nominal understood as possessor in the clause. I think that discourse considerations are involved here.) For these speakers, grammaticality judgments like the following hold.

For these same speakers, a preverbal subject will retain its possessor in a PA-type situation as well. Thus (32)-(34), which parallel (1)-(3).

These facts fit in beautifully with the claim that the absence of the possessives in both (1)-(3) and (29a), (30a), and (31a) is a result of PO. It makes sense under that theory that the possessor should be specifiable under exactly the same conditions in PA-type sentences as in other sentences. Once again, then, the facts are at the least consistent with the theory that PA-type sentences are to be accounted for by ED’s together with PO.¹⁵

1.1.4 Summary and conclusion

In Section 1.1.1 it was shown that either PO or PD is needed to account for certain facts in Spanish, and it was argued that PO is preferable. It was claimed that PO (or PD) can account for the fact that no overt possessor shows up in the sentences for which PA would be posited. In Sections 1.1.3.1 to 1.1.3.3 it was argued that the behavior of the datives that would be produced by PA is consistent with an analysis which posits that they are in fact ED’s. In Sections 1.1.3.3 to 1.1.3.5 data were given which showed that the behavior of the possessors in PA-type sentences is consistent with an analysis that posits that they are omitted rather than ascended.

I conclude that the independently motivated PO and the independently motivated ED’s, working together, can account for the same facts as PA would. Therefore Argument A, which claimed superiority for PA on the basis that it would require no universally new types of RN’s, is invalid. A theory positing ED’s and PO (henceforth ED-PO) also requires no universally new types of RN’s and accounts for the same facts.

1.2 In answer to Argument B

Argument B for PA was that positing it allows one to reflect the similarity in meaning between sentences like (1)-(3) and parallel sentences (like the English glosses) in languages without PA. Argument B can show that PA is necessary only if (a) the assumptions underlying it are valid, and (b) there is no alternative analysis that appropriately reflects the similarity in meaning between sentences like (1)-(3) and their English glosses. In Section 1.2.1 I will question the assumptions underlying Argument B, and in Section 1.2.2 I will claim that an analysis with PD would do much the same thing as PA even if the assumptions are valid.

1.2.1 Argument B’s assumptions are questionable if not wrong

It seems to me that at least three assumptions underlie Argument B. They are (i) Sentences like (1) and its English gloss are in fact the same in meaning; that is, the meanings are identical with respect to the relevant aspects, in particular the relationship of possession. (ii) This identity of meaning must be reflected by an identity of structure at some linguistic level. (iii) In fact, it should be represented at the initial syntactic level. I think that all three assumptions can be questioned. Assumptions (i) and (ii) will be hard to handle separately, so I will take them together first.

1.2.1.1 What is meaning? Where is it represented?

1.2.1.1.1 Three meanings of “meaning”

It is by no means clear that sentence (1) and its English gloss mean the same thing in the ways and to the extent necessary to sustain Argument B. “Sameness of meaning” can be judged by at least three criteria.

(i) In common parlance we say two expressions “mean” the same thing if they are functionally equivalent in general. In other words, do two expressions’ truth conditions coincide in most cases, or are they good translations for each other in most contexts? If so, we say they “mean” the same thing. I will call this the Functional Criterion. The distinctions this criterion makes are obviously matters of degree and may be held for one situation or purpose but not for another. For some purposes and situations ball and sphere have the same Functional “meaning”, but ball will not do where geometric accuracy is necessary, nor will sphere do in sports, especially American football.

(ii) A stricter test for identity of meaning is identity of truth conditions.¹⁶ By this criterion (the Truth-value Criterion), two expressions “mean” the same thing if and only if they have exactly the same truth values under all conditions. Ball differs in Truth-value “meaning” from sphere because there are situations in which one of them is appropriate and the other is not: it is true that an American football is a ball; it is not true that it is a sphere.

(iii) What we will call the Imagic Criterion makes more fine-grained distinctions. It has been pointed out that even when two expressions have identical truth conditions they may still differ in “meaning” in some sense. It is not the same to say Each of the men is a sailor as to say All of the men are sailors, even though the truth conditions for the two sentences are identical. To say that a bottle is half full and to say that it is half empty is to say slightly different things about the amount of liquid in the bottle, even though the amount of liquid is the same. The two sentences “mean” something different even though the difference in “meaning” is difficult to pin down. The nature of the Imagic Criterion will be discussed in Section 1.2.1.1.3.

“Meaning”, then, as judged by the Functional or Truth-value Criterions, can be (and has been) viewed as essentially a function of truth conditions: the greater the extent to which the truth conditions of two expressions coincide, the more felicitous it is to say that the expressions have the same “meaning”. The Imagic Criterion, however, appeals to some notion of distinctions of “meaning” that go beyond what is revealed by truth value judgments.

I would claim that sentence (1) and its English gloss (and the rest of the PA-type sentences and their English glosses) “mean” the same thing only in the sense implied by the Functional Criterion, not in the senses implied by the Truth-value and Imagic Criterions.

1.2.1.1.2 Different meanings by the Truth-value Criterion

If “meaning the same thing” means being functionally equivalent to a rather high degree, as implied by the Functional Criterion, then it seems clear that sentences like (1) and their English equivalents “mean” the same thing. Most situations (including the most common ones) about which you could felicitously say Le ensuciaron el coche could also felicitously be reported by They got his car dirty, and vice versa. But not quite all, as we saw in Section 1.1.3. For instance, in 1.1.3.3 it was pointed out that sometimes Le ensuciaron el coche refers to a situation in which They got his car dirty is inappropriate (e.g. when it is someone else’s car), and one must instead say They got the car dirty on him. Also, as pointed out in the case of a dead man in Section 1.1.3.1, They got his car dirty may be appropriate where Le ensuciaron el coche is not. In both cases there is a discrepancy in the truth conditions: it may be true that Le ensuciaron el coche where it is not true that They got his car dirty, and vice versa. Thus by the Truth-value Criterion the sentences do not “mean” the same thing.

This in itself is probably enough to undermine Argument B. The two sentences differ in “meaning” on the questions of whether the referent of the dative in the Spanish sentence must correspond to the possessor in the English sentence, and whether the referent of the possessive in the English sentence can always surface as a dative in Spanish. PA would predict that both of these questions would be answered affirmatively, but we see that neither of them can be.

It might be countered that PA need not be the only source for sentences like (1); i.e., that (1) is ambiguous rather than vague about who is the possessor. That is, the predictions of PA hold true, but only for a subset of instances of (1); the other instances are derived from a different source. Since the Truth-value Criterion distinguishes “meaning” only where there is a truth value discrepancy, it is possible (and even logical) to claim that in the cases where there is no discrepancy the “meaning” is the same. (Notice that this contention can not be urged against the Imagic Criterion distinctions I claim in the next section, since Imagic “meaning” distinctions hold even when truth values coincide.) Argument B would then take the following form: We should posit PA because it allows us to reflect the similarity (=identity of the relevant aspects) of meaning between those cases in which it applies and the corresponding sentences in English (and other languages) in which it does not apply. To sustain the argument, one would also have to posit that the English glosses are ambiguous between two kinds of possession. Thus the English gloss for (1) would have one “meaning” which would be the same as that of (1) on the reading where the possessor is coreferential with the DAT, and another which would be different. (One might suggest “affected possessor” vs. “non-affected possessor”.) Rather than the identity of meaning between (1) and its English gloss being at issue, it would be the identity of meaning between some instances of (1) and some instances of their English counterparts. Positing this double ambiguity here is ad hoc and comes perilously close to being argument in a circle: We decided that the sentences, both English and Spanish, are ambiguous, because otherwise PA will not work, and we know that PA works because the sentences are ambiguous. As a further consideration, for what it’s worth, constructions with so did or and … too have been proposed (Lakoff 1970) as a test for vagueness vs. ambiguity. These constructions, it is claimed, are possible with different readings in the case of vagueness but not in the case of ambiguity.¹⁷ Thus My uncle is a butcher and so is John’s (or and John’s is too) is vague rather than ambiguous as to whether maternal or paternal uncles are referred to, and therefore any reading is possible. By contrast, in I like my /ænts/ and John likes his too, /ænts/ is ambiguous between the aunts and the ants sense, and therefore only those readings are possible where both John and I like the same kind of /ænts/. By this test both (1) and its English gloss are vague rather than ambiguous:

Similar results are obtained by applying the same test to (2) and (3) and other PA-type sentences. Thus it seems not to be the case that these sentences are ambiguous.¹⁸

It would appear, then, that sentences like (1) and the corresponding English glosses do not “mean” the same thing in the sense required to support Argument B, as judged by the Truth-value Criterion.

1.2.1.1.3 Different meanings by the Imagic Criterion

The Imagic Criterion is more sensitive than the Truth-value Criterion. By the Truth-value Criterion we can judge that sentences like (1) [and its gloss] differ in “meaning” in those instances in which their truth values do not coincide. But I am convinced that the sentences differ in another sense of “meaning” even in those instances where their truth values do coincide. In other words, given (1) and its English gloss both making felicitous reference to the same situation, I would claim that they are saying slightly different things about it, that in fact they “mean”, in the Imagic sense, slightly different things.

It is not easy to characterize the sense in which the sentences “mean” different things by the Imagic Criterion, in part because there is no simple and generally accepted test like similarity of truth conditions which will distinguish the “meanings”. Different notions of “meaning” have been appealed to; I would not be at all surprised if what I refer to as the Imagic Criterion is in fact a bag of different criteria capable of making finer distinctions than those possible under the Functional and Truth-value Criterions. Often such criteria are lumped together under a heading of “Intuition”. That is to some extent enlightening; intuitive judgments are based on such distinctions in meaning, and introspective examination of one’s intuitions can often give a good start on a characterization of what those distinctions are. It was my intuitions as a speaker of Spanish and English that convinced me that Argument B was false, and indirectly led to the writing of this paper. But of what do such “intuitions” consist? Langacker (1979:88-89) speaks of differences in conceptual viewpoint being conventionally coded by different semantic units (including both unitary predicates and constructions), which represent different “images” or views of the conceived referent. Thus The statue is on the pedestal and The pedestal is under the statue, when applied to the same scene, represent two different viewpoints on or images of that scene. Perhaps this is as good a way as any to characterize the difference between the Truth-value Criterion and the Imagic Criterion. Truth-value distinctions show that different scenes are being referred to, while Imagic distinctions claim that the same scene may be being referred to, but that it is being construed differently, through a different Image, from a different conceptual, social and/or emotional perspective.¹⁹

I believe (and hope to illustrate, if not demonstrate, below) that the Imagic distinctions are primary over the Functional and Truth-value distinctions in that they entail, and thus can be used to explain, the Functional and Truth-value distinctions, but not vice versa.²⁰ When two expressions view the same scene through different images, it is often (though not necessarily always) possible to imagine a scene which one of those images fits but the other does not. To change the metaphor, you can usually find a scene on which one of the two viewpoints is possible but the other is not. This will amount to a distinction by the Truth-value Criterion. In other words, once you know what the expressions “mean” in the sense of the Imagic Criterion, you will know where and why the Truth-value distinctions will hold; but finding a Truth-value distinction does not tell you automatically why it occurs, nor which Imagic distinction is responsible for it. Truth-value distinctions are just that tip of the iceberg which truth-value judgments can make visible; they tell you little about the shape of the iceberg as a whole. If Truth-value distinctions occur often enough, they will amount to a Functional distinction: the two words will not “mean” the same thing at all except perhaps in specific specialized contexts.

I would also claim that Imagic “meaning” is primary in that it can explain similarities in Functional and Truth-value meanings, but not vice versa. Imagic “meaning” not only can make distinctions too fine for the Truth-value Criterion to reveal, but it can also make subtler semantic connections, showing similarities between expressions too different for the Functional Criterion to show as “meaning” the same thing. An example of this sort of thing is involved in the almost instantaneous grasping of many jokes. Most people understand They’re called wisdom teeth because they smart the first time through. Involved in that understanding is (among other things) the ability to perceive an important semantic similarity between wisdom and one sense of smart. Yet it is improbable that everyone who gets the joke has ever heard wise and smart or wisdom and smartness used as functional substitutes. I do not remember hearing them so used myself; I would as soon class them as antonyms as as synonyms. Their truth values do not coincide except accidentally (the same person may be both wise and smart, but then the same person may be both tall and cross-eyed) and they are rarely if ever acceptable translations for each other. Thus the similarity between them is not a clear Functional similarity, and certainly not a Truth-value identity. Yet the similarity is clearly and easily perceived. Such similarities are also crucial to an understanding of metaphor. Whoever first called a narrow part in the road a bottleneck was clearly responding to Imagic similarities between two things clearly distinct by the Functional Criterion. Indeed, whoever first called a bottleneck a bottle neck was making a similar response. The reason such metaphors catch on is that other speakers are already aware of the Imagic meanings of (in this case) bottle and neck, and can see the appropriateness of naming a bottleneck by those words. Or consider the words brother and sister. Clearly they have a lot in common Imagically. This common Imagic material can be coded by sibling in academic dialects of English; other dialects which do not have that word also perceive the similarity between brother and sister. Yet they are clearly not Functional equivalents: it is difficult to find any case in which both are true or appropriate. Again, we have a clear Imagic similarity which does not correspond to a Functional similarity. In many cases, however, such similarities will result in two expressions’ being alternative in some environment—in other words, a Functional similarity of “meaning” will arise. There is at least one Imagic similarity to account for every Functional similarity, but not vice versa; Functional similarities are just another tip of the iceberg of Imagic meaning. If there are enough such similarities (and no egregious dissimilarities) between two expressions, a Truth-value similarity (or identity) will obtain.

Here we are more concerned with Imagic distinctions of meaning than with similarities. While such distinctions are subtle and difficult to characterize, they are very pervasive. They occur, of course, between every two expressions that differ greatly in meaning, but in those cases they are often ignored (at least by linguists) since the difference in “meaning” can be characterized by the Functional Criterion or the Truth-value Criterion. However, the need to posit them shows up more clearly in the consideration of such phenomena as language-internal synonymy and paraphrase and cross-language translation. It is very difficult to find a synonym, paraphrase, or translation that does not involve some difference in Imagic “meaning”. For instance, do the verbs gaze and stare “mean” the same things? Do be quiet and shut up “mean” the same thing? Are hors d’oeuvres and appetizers the same in “meaning”? They are by the Functional Criterion and possibly by the Truth-value Criterion, but not by the Imagic Criterion. Does John bought the dog from me “mean” the same as I sold the dog to John? Or, perhaps more controversially, does John hit me “mean” the same as I was hit by John? They apparently do by the Functional and Truth-value criteria, but it is at least arguable that they differ by the Imagic Criterion.

To return to a previous example, does half empty “mean” the same thing as half full? They are clearly Functionally equivalent. They are apparently the same in meaning even by the Truth-value Criterion—I have not been able to think of any situation in which it is true that something is half full but not true that it is half empty, nor vice versa. Certain infelicities do arise. It would be rather odd to say about someone who was filling a glass with water, He stopped when it was half empty.²¹ Yet that oddness would not amount to a clear violation of the Truth-value Criterion: it is still true that he stopped when it was half empty. However, I would claim that there is a clear Imagic difference between half empty and half full. Half empty locates a point by Imagic reference to a scale of emptiness, where one scans from full towards empty. Half full locates that same point on a scale of fullness, where one scans from empty towards full. Any point between empty and full can be specified by reference to either scale. That is why three-fourths full can refer to the same point as one-quarter empty, or one-quarter full to the same point as three-quarters empty. To get at the same Image in a slightly different way, half full Imagically measures the amount of liquid in the container, whereas half empty measures the empty space above the liquid. This accounts for the infelicity noted above; when a glass is being filled the direction of scanning most naturally goes with the movement of the liquid, from empty towards full. To say half empty in such a situation practically forces you to scan from both directions at once, for no good purpose, with odd or humorous sounding results.

Similarly, does the expression four plus one “mean” the same as the expression five? (Or does seven minus two, or ten divided by two? They are functional equivalents in some sense, and it is difficult if not impossible to find a case in which one is true and the other is not. Yet they differ Imagically in that four plus one arrives by a complicated route at a conceptual situation comparable to the one achieved directly by five. The same number is being referred to, but in two different Imagic ways.

Or, to take a cross-linguistic case, is the Spanish sentarse (seat:REFL) the same in “meaning” as the English sit down? Does acostarse (lay:REFL) “mean” the same as lie down? They are functionally equivalent in most situations, and thus “mean” the same thing by the Function Criterion. The Truth-value Criterion differentiates between them, however, in a few instances. Consider the case of a person lying on a couch, who then assumes a seated position on the couch. Here sentarse is appropriate to describe the action but sit down is not; sit up is called for. Note that there is no parallel case to distinguish between acostarse and lie down; either expression is appropriate to describe a person assuming a prone position, no matter what his previous position was. Or consider the case of an action of forcing a physically resisting child into a seated or lying position. It is appropriate to say I made him sit (or lie) down,²² but it is not appropriate to say Le hice sentarse (DAT I:made seat:REFL) or Le hice acostarse (DAT I:made lay:REFL). These sentences would be appropriate if psychological rather than physical pressure were used. but for cases of physically forcing the child to sit or lie down, Lo senté (ACC I:seated) or Lo acosté (ACC I:laid) are called for. Thus by the Truth-value Criterion acostarse and lie down don’t mean the same things, nor do sentarse and sit down. Sentarse and sit down are differentiated by the Truth-value Criterion in two cases, but acostarse and lie down are differentiated in just one of those cases. However, I would claim that by the Imagic Criterion both of these Spanish expressions differ from their English glosses in exactly the same way. There is inherent in the conception of a prototypical act of sitting or lying down a notion of departure from the canonical vertical orientation of human posture. I would claim that that notion is represented in the semantics of the English forms but not in the semantics of the Spanish forms. In other words, there is an Imagic distinction between the English and the Spanish forms at this point. This distinction is coded by the presence of the word down in English and the absence of any such word in Spanish.²³ This Imagic distinction happens to result in a Truth-value distinction between sentarse and sit down, because some acts of sitting result in an approximation to, rather than a departure from, the canonical vertical posture. Those acts must be coded by sit up. The fact that there is no similar Truth-value distinction between acostarse and lie down is explained by the “meaning” (Imagic sense) of lie: it is difficult if not impossible to conceive of a situation in which assuming a horizontal posture leads one to a more rather than a less close approximation to the canonical vertical posture. Thus there is no English expression lie up in opposition to lie down. (Similarly there is no stand down in opposition to stand up.) Also inherent in the prototypical conception of sitting or lying down is a notion of reflexivity. When one sits or lies down one does something that affects the state of one’s body. I would claim that that notion is not represented in the semantics of the English form but that it is in the Spanish. It is coded by the use of reflexive forms of the transitive verbs sentar (‘seat’) and acostar (‘lay’). This Imagic distinction can lead to a Truth-value distinction in certain cases in which one’s body achieves the specified state (seated or prone) without one’s doing anything to cause it (e.g. the case discussed above of a child being physically forced to sit or lie down). The fact that Le hice sentarse/acostarse can be used when psychological rather than physical pressure is employed is explained: in such circumstances the child is still seating himself (or laying himself down) even if under duress. Thus it appears that the occurrence and nature of the Truth-value distinctions can be explained by the occurrence and nature of the Imagic distinctions, though not vice versa.

I think that the case with (1) and its English gloss (and the other PA-type sentences and their glosses) is similar. The two are often functionally equivalent, thus “meaning” the same in terms of the Functional Criterion. As we have seen, the Truth-value Criterion distinguishes between them in certain cases. Yet even when they refer to identical conceptualizations as far as the Truth-value Criterion shows us, I would claim that they have different “meanings” in the sense of the Imagic Criterion. The semantics of the English sentence contains a reference to possession, which is coded explicitly by a possessive pronoun his; I would claim that the semantics of the Spanish sentence does not contain such a reference. This explains why the identity of the possessor is vague in the Spanish sentences; it simply is not specified. It also accounts for the Truth-value distinctions where Le ensuciaron el coche is true (appropriate) but They got his car dirty is not because it is not his car. The semantics of the Spanish sentence, on the other hand, contains a reference to a person’s being affected by the action of the car being dirtied, whereas the semantics of the English sentence does not. This reference in the Spanish sentence is coded by the presence of an ED. Positing this Imagic distinction accounts for the fact that the English sentence is vague as to the extent to which the possessor (or anyone else) was affected by the dirtying of the car. It also accounts for the Truth-value distinctions where They got his car dirty is true (appropriate) while Le ensuciaron el coche is not, because he was not affected by the dirtying of his car (being e.g. dead). Positing these Imagic distinctions is entirely consistent with the Functional equivalence of the sentences: they are equivalent precisely because in the prototypical or most common cases the possessor and the person affected are the same. Thus these cases can be viewed through either Image.

Thus it would appear that by the Imagic Criterion all instances of (1) and its English gloss differ in “meaning” precisely in that the English sentence specifies possession whereas the Spanish one does not, and the Spanish sentence specifies that someone is affected whereas the English one does not. Similar considerations show the same to be true of (2) and (3) and other PA-type sentences with respect to their English glosses. Thus it makes sense to claim that these meaning distinctions are properties of the constructions, not just of the individual pairs of sentences. English speakers use a construction which specifies possession but leaves affectedness vague; Spanish speakers use a construction which specifies affectedness but leaves possession vague. English speakers, of course, are aware that very often a possessor is affected by what happens to his possessions, and similarly Spanish speakers are aware that very often the person affected by what happens to a possession is its possessor. But in each case the sentences they use do not code those notions explicitly but rather leave them vague.

Thus it would appear to be clear that PA-type sentences and their English glosses do not “mean” the same thing in the sense required to support Argument B.

1.2.1.1.4 Where is identity of “meaning” represented?

Another way to get at the same problem is to inquire whether the identity of “meaning” between sentences like (1) and its English gloss is a linguistic identity at all. I do not think that it is. Linguistically the sentences are similar in various ways. There are parallel semantic entities with sometimes parallel semantic relationships, such as, for sentence (1), the agent THEY and the patient CAR which is semantically definite, both related to the action DIRTY (or CAUSE-INCHOATIVE-DIRTY, if you like) occurring in PAST time, and a 3 PERS SG entity somehow involved. There are parallel syntactic phenomena and even parallel phonological phenomena. Yet though these parallels exist, I would claim that they do not amount to identity, but only to similarity. And I would claim that the differences include precisely those that most fit in with the ED-PO hypothesis; namely, that the semantically involved 3 PERS SG is involved as a possessor in the English sentence and as affected person in the Spanish sentence, and that the nominal representing that 3 PERS SG is syntactically a (surface) possessive in English and an ED in Spanish.²⁴ It seems to me that the only identity that there is between the sentences is a sort of conceptual identity which is not really linguistic in nature, though the ability to perceive it is deeply involved in the use of language. We have the ability to perceive that both sentences “fit” many of the same conceptual scenes just as we have the ability to see that two different paintings may “fit” the same landscape, or that different views of a face or the back of a head may “fit” the same person, or that both G⁷ and D⁷ chords may, in certain contexts, “fit” the idea of a dominant seventh. But I would claim that at every linguistic level, including semantic levels, the two sentences differ because the English one specifies possession whereas the Spanish one does not, and the Spanish one specifies that someone is affected whereas the English one does not.

1.2.1.2 Is possession in PA sentences a syntactic thing?

Even if it is wrong to claim that (1) and its English gloss are different at all linguistic levels, it is not at all clear that they are syntactically to be viewed as having identical structures with respect to possession. I know of no syntactic arguments for assigning initial Poss arcs to the DAT nominals in PA-type sentences. Such syntactic evidence is crucial; cf. the discussion in Section 3.3. Thus even if the sentences were semantically the same, that would not be sufficient to argue that they are syntactically the same with respect to possession, as we would have to posit were Argument B to have any validity.

1.2.1.3 Conclusion

I conclude that it is very doubtful (in fact it seems wrong to me to claim) that sentences like (1)-(3) and their English glosses are equivalent in meaning in the sense required for Argument B, and that even if they were, there is no clear reason why that equivalence should be represented in the syntactic structures of the language. I thus conclude that Argument B is invalid.

1.2.2 Argument B does not exclude PD

Even if all the questions raised in the preceding section with respect to assumptions (i)-(iii) were settled in favor of what Argument B demands, Argument B would still not show that PA is necessary. It would constitute an argument for PA against PO, but it would not distinguish between PA and PD, since both of them represent the similarity of meaning with respect to the possessive relationship between the Spanish and the English sentences in exactly the same way.²⁵ So a model with PD and ED’s would still be on a par with a PA model. Thus, even if the assumptions underlying it were valid, Argument B would fail to prove that PA is necessary.

1.3 Conclusion

On the basis of the material presented in Sections 1.1 and 1.2, I conclude that both Argument A and Argument B are invalid. In the absence of any further arguments for PA, I further conclude that PA is therefore not necessary in Spanish.

2. ED-PO is preferable to PA in Spanish

In this section are presented arguments to the effect that ED-PO is superior to PA in accounting for the relevant facts of Spanish. The first argument to be presented is an argument from simplicity. The second through fifth arguments derive from the data presented in Section 1.1.3. The second is an argument from the behavior of the datives in sentences like (1)-(3). The third and fourth arguments involve ways in which the superficially non-existent possessors in these sentences behave as if they are omitted rather than ascended. The fifth argument involves the fact that possessors can be overtly marked in PA-type and non-PA-type sentences under exactly the same circumstances. The final argument is the converse of Argument B: ED-PO is preferable to PA because it adequately represents the differences in meaning between sentences like (1) and its English gloss, whereas PA obscures those differences.

2.1 A theory of Spanish without PA is simpler

This argument is very simple. ED’s and PO (or PD) are motivated within Spanish quite apart from sentences like (1)-(3). PA, on the other hand, is not independently motivated within Spanish. So why do we need it? A theory without it is preferable to a theory with it, by Occam’s razor.

2.1.1 A digression on the universal availability of grammatical devices

Although the argument is basically simple, it appears that the terrain has been confused by claims that if a device is universally available, then it costs the grammar of a particular language nothing in terms of simplicity to utilize it. I believe that such claims are misleading, if not erroneous. The following is a summary of how I think such claims should be evaluated.

Given two devices A and B which equally well account for a range of data in language X, and given that A is independently attested in X, while B is not, there are four logical possibilities:

(i) If both A and B are universally available, I claim that a theory which uses A to explain the data is preferable to one that has recourse to B. Thus, any theory that would explain English data in terms of some clearly attested English phenomenon such as SVO word order is to be preferred over a theory which would explain the same data in terms of say a Modalis Case marking 2-Chomeurs (which is attested in Eskimo, and therefore universally available). (This is, at least as far as this argument goes, the sort of situation we are dealing with here; both PA and ED-PO are universally available, but only ED-PO is independently attested in Spanish. However, in the next sections, I will argue that actually ED-PO also accounts for the data better than PA.)

(ii) If A is universally available while B is not, then A is obviously superior.

(iii) The opposite case, where A is not universally available while B is, never occurs. Any time a device A is really clearly attested in language X, it is by definition universally available; the test for universal availability is clear attestation in some language.

(iv) Similarly, the case where neither A nor B is universally available, never occurs. If (as is given) A is clearly attested in the language independent of the data in question, it is ipso facto universally available.

Thus, if the preceding judgments are accepted, any time there is a choice between two devices, one of which is independently attested within the language in question and the other of which is not, and both of which account equally well for the same range of data, the independently attested one is preferable, irrespective of which of them may or may not be universally available.

The question of universal availability only comes up as a factor in the following few cases, as far as I can see.

(v) A and B both cover the same language X data equally well; neither of them is independently motivated within X; A is not universally available (i.e. attested in other languages) but B is. In this case B is preferable to A. (This is the only clear case where universal availability should cast a deciding vote.)

(vi) A and B both cover the same language X data equally well; both of them are independently motivated within X; A is not attested in other languages (though it is, of course, universally available) but B is. This might provide a very weak argument for preferring B over A. Actually, in a case like this it might even be better to claim that both B and A should be appealed to to account for the data (cf. Hankamer 1977).

(vii) A covers a range of data in X better (e.g. more elegantly) than B. A is not universally available (i.e., attested in other languages) but B is. Neither A nor B is independently motivated within X. Here it seems that a judgment needs to be made, based on the degree to which and in which sense A handles the data better than B. If A very clearly handles the data better than B, then that amounts to clear attestation for A, at which point A is universally available, and clearly to be preferred over B. Its lack of attestation in other languages should be irrelevant. But if A is only very slightly or not clearly better than B, then perhaps B would be preferable. Making such a decision is tantamount to putting one’s faith in the underlying unity of human language, and implying that if we understood things better, the superiority of A over B would be seen to be illusory.

2.2 The datives in PA sentences are ED’s

It was claimed in Section 1.1.2 that the datives in PA sentences behave like ED’s in the following ways: they may occur wherever the understood possessor (let us call him Y) may be viewed as affected by the predicated action of state; they need not occur where Y need not be viewed as affected, they must occur wherever Y must be viewed as affected and they must not occur wherever Y must not be viewed as affected. If these claims are true, it would seem clear that those datives are in fact ED’s.

This can be reduced to an argument from simplicity of the following form: A theory with PA would need two extra constraints, one guaranteeing that PA will occur where the possessor is viewed as affected, and another guaranteeing that PA will not occur where the possessor is not viewed as affected. Under the theory that claims that the PA sentences have ED’s in them, these facts are an automatic consequence. ED’s code the notion “(person) affected”, and thus an ED will appear to represent Y whenever Y is viewed as affected, and will not appear when he is not viewed as affected. Thus ED-PO does not need the extra constraints, and it therefore is preferable.

To put the same argument in yet another form, a theory saying that a possessor is coded by a dative if and only if it is viewed as affected is simply a notational variant of one that claims that that dative “means” affectedness. Any time a morpheme or construction occurs if and only if a given semantic specifications holds, we say that it “means” that specification. Thus PA would be claiming, in effect, that the datives it produces have the same meaning as ED’s. These datives, however, are not ED’s, because they come from a different source. Thus PA has two sources for datives with exactly similar meanings, whereas ED-PO has only one source for those datives. In this way ED-PO is simpler and to be preferred. Also, PA would be claiming that the semantic relation of “(person) affected” corresponds to a GR_x in some cases, but not in others, whereas ED-PO can claim that that notion always corresponds to a GR_x in Spanish. Thus ED-PO is again simpler and to be preferred.

2.3 The possessors in PA sentences are omitted, not ascended

As the data presented in Section 1.1.3.3 show, the possessors in PA-type sentences act more like they are omitted or never specified than like they are ascended. The possessor is usually the same as the person affected, but not always.²⁶ Imagine such a fact being true of any ascension, e.g., Subject Raising in English. If such were the case, John seems to be tired would be able to bear a reading where Aloysius is the one who is tired, or He expects John to put his foot in his mouth could mean that George or Mehetabel or someone else is expected to put someone's foot in his mouth. I do not think that anyone would ever have proposed raising if such facts had obtained. However, where something is simply omitted and is never present linguistically, such vagueness is to be expected.²⁷ For instance, consider (12), where the possessor is omitted, or a sentence like I hit him, where it is usually assumed in vacuo that the instrument of hitting is the hand but where it could perfectly well be a stick or even a car. Or consider I bought a Ferrari yesterday, where the benefactee is left vague; it will be assumed in vacuo that it was bought for the speaker but it could perfectly well have been bought for someone else. It thus seems clear that what is going on in sentences like (1)-(3) is omission rather than ascension.

This can be reduced to an argument from simplicity very similar to that in Section 2.1. The PA theory is going to need two devices to account for the absence of an overt possessor in (1)-(3); PA itself for those cases where the DAT is the same person as the possessor, and some form of PO or of PD (perhaps fed by PA) for the cases where the DAT is not the same person as the possessor. ED-PO, however, needs only PO to account for all the cases. Thus ED-PO is simpler and to be preferred.

A related argument is the following: as we have just seen, PA will have to posit the PA structure for cases of (1) which have coreferentiality of the DAT and the understood possessor, and some other structure for those cases which do not have that coreferentiality. I will assume (following usual practice within RG) that those structures would differ at the initial stratum. In fact I would expect that in one structure the referent of the DAT would be marked as initial Possessor, while in the other one it would not be so marked. I will further assume that different initial syntactic structures reflect different semantic structures, again following usual practice within RG. Under these assumptions, the fact that PA would have to use different syntactic structures would predict ambiguity rather than vagueness of possessorhood in the semantics of sentences like (1). ED-PO, however, predicts vagueness rather than ambiguity, since all instances of (1) come from one initial structure, and possession is never specified. As shown in Section 1.2.1.1.2 by sentence (35), (1) is vague rather than ambiguous by the and so did … test. Since the predictions of ED-PO rather than those of PA are borne out, ED-PO is preferable.

The PA theory is also going to need two devices to account for the DAT in sentences (1)-(3): PA when the possessors and the DAT’s are the same, and ED’s or some such thing to account for the other cases. ED-PO can account for all the cases with only ED’s. Again ED-PO is simpler and to be preferred.

2.4 Possessors can appear overtly in otherwise PA-type sentences

ED-PO is also supported by the fact that the possessors need not be omitted but can in fact be specified, as the data in Section 1.1.3.4 show. If PA is posited for those sentences, we have no explanation for why the possessors remain as possessors. Again, think what it would mean to posit this for another ascension. It would mean that by Subject Raising you could get English sentences like *John seems for Aloysius to be tired, or *He expects John for Mehetabel to put his foot in his mouth. Again, I do not think that anyone would have posited Raising in English if such facts obtained. However, when an item is simply omitted, it is not surprising to find that it can be specified if desired. For instance, contrast (12) with I put it in your mother’s fridge, or contrast I hit him with I hit him with a noodle, or I bought a Ferrari yesterday with I bought a Ferrari for my grandmother yesterday. Thus it appears that we are dealing with an omission rather than an ascension.

Again, this can be reduced to an argument from simplicity; for these cases a theory with PA is going to need something like ED’s to account for the presence of the DAT, plus a constraint prohibiting PA from applying. ED-PO, however, need say nothing other than that it is permissible to include rather than omit the possessor when it is desirable (and non-redundant). And even this statement is exactly what we should expect; it is probably a universal of language in some sense that one is permitted to specify items left vague when it is desirable and does not contradict the norms of the language. Thus ED-PO is simpler and more preferable.

2.1 PA would be suspended exactly where PO is

As shown in (29)-(31) (Section 1.1.3.5), PO is suspendable for some speakers in preverbal subjects; the possessor, even though coreferential with a term nominal in the main clause, may be overtly specified in this position. Under either the ED-PO or the PA grammars some statement of this fact will be needed. As (32)-(34) show, the same pattern holds for sentences of the PA type: when the possessed nominal is a preverbal subject, its possessor may be overtly specified. Under the PA grammar this is a new kind of pattern. A separate statement would be needed that PA need not occur from preverbal subject position, but that instead a coreferential dative may be put into the clause (for no particular reason). Alternatively a new process of Copy-PA might be proposed, to occur only from preverbal subjects and after which PA could not apply. However it is done, complication is entailed. Under the ED-PO model, however, the independently needed statement accounts for the same facts without any complications. In this way ED-PO is simpler than PA and to be preferred to it.

2.6 PA obscures semantic differences

In Section 1.2.1 it was argued that the constructions represented by PA-type sentences like (1)-(3) and their English glosses differ semantically in that the English construction specifies possession but leaves affectedness vague, whereas the Spanish construction specifies affectedness but leaves possession vague. These systematic differences in the semantics are admirably represented by ED-PO, which has an ED specifying affectedness but no specification of possession in Spanish. (Both models would presumably have a Poss arc but no ED in English.) PA, on the other hand, has an initial Poss arc in the structure of the Spanish sentence with a specific nominal heading it. This corresponds to no semantic relationship at all. Its presence argues against PA. PA also has no GR coding affectedness in the initial stratum in the Spanish sentences. If, as is often tacitly if not explicitly assumed, the only articulation of semantics with syntax in RG is at the initial stratum, this lack also counts against the PA model. Thus the PA model for Spanish both implies specific possession, and does not imply, under certain assumptions, affectedness. Both of these implications obscure the semantics of the Spanish sentences.

To put the same argument in another form, under ED-PO the link-up between the semantic and syntactic structures will be simple and direct, whereas under PA extra and ad hoc machinery will need to be added in order to make the linking. Thus ED-PO is simpler and to be preferred.

2.7 Conclusion

I therefore conclude that ED-PO is clearly preferable to PA for accounting for Spanish sentences like (1)-(3).

3. Implications

If the conclusions of the preceding sections are accepted, they have important implications. First, they imply that other analyses using PA should be re-examined. Secondly, they imply that the relationship of semantics to syntax had better be re-examined and certain practices of syntactic research and argumentation severely questioned if not abandoned. Thirdly, they have implications for translation theory. These topics will be addressed briefly in the following sections.

3.1 What about PA in other languages?

If the foregoing argumentation is valid, there is in Spanish a construction which at first glance looks very much like a PA construction but which on further examination turns out not to be one. A clear practical consequence is this: there is at least one universally available close look-alike to PA. Analyses using PA should therefore be closely examined to determine whether PA is needed or whether an analysis along the lines of ED-PO is equally viable. Perhaps it is the case that all, or certain kinds of, PA analyses that have been proposed are better explained by something analogous to ED-PO. I will discuss these possibilities vaguely and briefly in Sections 3.1.1 and 3.1.2.

3.1.1 PA is undesirable universally

PA is undesirable universally for at least two reasons. One is simply that it is an extra device. We could say that universal grammar would be simpler if it did not have to contain a description of PA as universally available GR configuration. Another way to say the same thing is that if PA were non-existent we could make universal grammar stronger by being able to make the generalization that possessors do not ascend. We would be able to cross off another item from the list of ways languages differ.

Another reason why PA is undesirable universally is that it violates two proposed universals: the Relational Succession Law and the Host Limitation Law (Perlmutter and Postal, to appear (b)). The Relational Succession Law states that when an ascension takes place, the ascendee assumes the GR of hits Host (the structure out of which it ascends). Thus a nominal which ascends from within a 1 will be a 1 upstairs, or a nominal which ascends from a 2 will be a 2 upstairs. This Law would be violated in Spanish by sentences like (1)-(2) if PA were posited for them, and it is violated by the analyses that have been proposed using PA in Tzeltal and Georgian and French (at least) (Aissen 1979, Harris 1976, Frantz 1979). The Host Limitation Law states that only Terms (1's, 2's, or 3's) can serve as Hosts. This Law would be violated by Spanish sentences like (37) if they were accounted for by PA.

The Host Limitation Law is also violated by the analysis posited using PA in Georgian.

Both of these Laws can be modified fairly easily to apply only to ascensions from clauses (i.e., all well-known ascensions other than PA). However, it would be preferable from a universal perspective if PA could be shown to be unnecessary or if it could be shown that the only real cases of PA are those that do not violate the universals, which then could be allowed to stand in their more general form.

3.1.2 PA may be unnecessary universally

I find it hard not to feel that PA is probably not really needed in the other languages for which it has been posited. I do not know the data in any of those languages in any depth, so I speak in ignorance and am ready to be corrected. But it seems to me very probable that some model similar to the ED-PO model might handle those cases as well. I am aware that not all of those languages will have something exactly like ED’s, and also that some of them do have PA-type structures in places where ED’s could not occur. For instance, Perlmutter (personal communication) says that in Rumanian the parallel to (17) is grammatical, as it is in Southern Tiwa.

Yet it seems to me that those cases might be able to be handled by positing some GR_y which instead of coding “(person) affected by the action or state predicated” would code some more tenuous or less specific semantic connections, such as “(person) with reference to whom the predication occurs”. This GR_y could advance to 3 for those languages with PA to 3, or to the GR of the nominal by virtue of which the person is referred to for languages with PA by the Relational Succession Law. PO (or PD) would be necessary to complete the picture.

Frantz (personal communication) says that he knows of no evidence against (or for) such a solution for Southern Tiwa. It is also interesting to note that in Chamorro (Crain 1979) sentences for which PA has been proposed apparently have a semantic relationship of “in spite of” between the clause and the putative ascendee. In this case I would posit a GR_y corresponding to that semantic relationship.

Not knowing the other languages, I do not know what evidence can be found independently in them which would support such proposals or militate against them. I also do not know how to evaluate the difference in universal terms between a model which makes a configuration like PA available and one which instead allows a new GR_y. If this approach could be made to work, however, it would make the relationships between languages like Spanish and these others clearer; they would differ only in the degree of involvement necessary conceptually for a nominal to qualify to be coded by GR_x or GR_y.

At the least I feel that pursuing the possibility of explaining PA-type structures by means of a model similar to the ED-PO model is likely to be a fruitful field for investigation.

3.2 Implications for practical syntactic analysis

It has been quite common practice since the mid 1960’s for syntax to be done following a sort of rule of thumb that when two expressions are paraphrases of each other, they should be given identical deep or initial structures and the difference in their form should be explained syntactically if possible. The assumption is that the paraphrase relationship indicates that the two expressions have essentially the same meaning, and positing identical initial structures will reflect this fact, simplifying the link-up between semantics and syntax. In particular, this strategy has been common in Relational Grammar. Thus Frantz (1979:30) gives the reasoning behind a PA analysis of Stoney data as follows: “We can account for the paraphrase relation of these two sentences, as well as their structural differences, by saying that [the second] involves ascension of the possessor …”

This study is by no means the first to deprecate such practice, nor the first to point to what I believe to be the proper alternative (see e.g. Langacker 1976, 1980). But the facts and arguments here presented do, it seems to me, show at least one case where such an analysis is clearly wrong; where even though there is a paraphrase relation between two constructions they have different semantic structures and should be given different initial syntactic structures. This means that some different way of accounting for the paraphrase relationship is necessary. I think that such a way is provided by a proper view of the complexity of Imagic meaning and of the conceptual ability of humans to view a situation through more than one Image. If, as I have suggested and strongly believe to be true, paraphrases differ Imagically more often than not, this way of accounting for paraphrase will be the ordinary one; cases with identical semantic structures will be the exception rather than the rule, and the burden of proof will be on anyone who would claim that any case of paraphrase is to be accounted for by an identity of semantic structures.

If the paraphrase relationship can and usually does exist with different semantic structures, what reason is there to suppose that it will require identical (initial) syntactic structures? I would judge that there is no reason to suppose it, and that therefore it would be more practical to assume that where there is a difference in surface form it is more likely than not to correspond to some difference in semantic and initial syntactic form.

3.3 Implications for syntactic argumentation

Quite apart from the practical question of which strategy is more likely to lead to insightful analyses is the question of what is necessary in argumentation to support an analysis once it has come to mind. A weakness in many RG analyses has been that initial relations have been claimed with little or no argumentation to support them. This is in part because it is quite difficult to find syntactic arguments for many proposed initial relations, and also in part because analysts rest on the assumption that similar semantic relations will link up with similar GR’s cross-linguistically. Thus Frantz (1979:1) speaks of analysts having “come to expect a fairly straightforward correlation between the semantic role of a nominal and its syntactic function,” and of “the claim that there are fairly straightforward principles for assigning (initial) grammatical relations on the basis of semantic notions such as agency, recipiency, affect, etc.” Or Perlmutter and Postal (1977:402) speak of grammatical “relations as given cross-linguistic substance (in part) by universal connections between the relational signs 1, 2, etc., and some representation of semantic relations.” Indeed, Frantz (1979:67) lists as a Principle of RG the “Universality of initial termhood: initial GR’s are predictable from semantic relations.” That this is necessarily the case is far from clear, however. Consider cases like buy vs. sell, which (according to Perlmutter, personal communication) are probably best viewed as encoding the same basic semantic material, but having the semantic relations linked to different initial GR’s. Or consider the case of many American Indian languages which have no clear language-internal evidence of a GR of 3 (Indirect Object), but instead apparently code the nominals which correspond to 3's in other languages as 2’s (see e.g. Comrie 1979, Tuggy 1979a, 1979b). The whole problem of exactly how to link up the semantics with the (initial) GR’s has not to my knowledge been worked out in any detail within RG; that is one aspect of the theory that is in dire need of development. It is true that most other theories are also deficient in this respect, but for at least some of them it is not as crucial because they are not positing such abstract initial relations and can give clearer syntactic evidence for the more superficial relations they do posit.

To this problem must be added the problem presented in this paper of Imagic distinctions in meaning which are easily missed or glossed over when working in a foreign language (or even one’s own!), and which cast strong doubt on whether the semantic relationships themselves are actually the same from language to language, much less the syntactic relationships which depend more or less “straightforwardly” on them. The resultant picture should make it clear that strong syntactic argumentation to establish initial GR’s is quite crucial in arguing for RD analyses. If you cannot be sure that the semantic relations are the same, nor that they will always correspond to the same initial syntactic relations, you will need pretty strong evidence beyond a correlation with intuitively like semantic roles to establish an initial structure different from the final one. Unless such evidence is available, any such analysis will be dubious.

3.4 Implications for translation theory

Another area in which the argumentation of this paper and the notion of Imagic distinctions in meaning is relevant is the theory of translation. Much has been said about translation under the assumption that it is essentially possible; that one can convey all and only the meaning of a message in a source language by a translation in a receptor language. Thus Beekman and Callow (1974:20) define the task of translation as “to communicate the meaning of the original”, and they quote with approval Hollander’s (1959:207) dictum that, viewed customarily and common-sensically, “to translate a sentence from one language to another is somehow to discover its meaning and then to construct a sentence in the new or target language that possesses the same meaning.”²⁸ The idea is that the two languages will have the same meaning structures; one must exegete the source expression to arrive at the meaning and use the grammar of the receptor or target language to construct the form appropriate to that meaning in that language. Translation is possible because the two languages will have identical meaning structures.

This study would indicate that this is not always the case; rather it suggests (and my personal experience corroborates) that it is rarely if ever the case. Imagic distinctions in meaning are so pervasive and so subtle that it is virtually impossible to translate any stretch of speech longer than a few morphemes from one language to another without making some change in some facet of some image, winding up saying slightly different things in the different languages. I am obviously not the first to notice this: compare Nida’s (1959:13) comment: “No translation in a receptor language can be the exact equivalent of the model in the source language. That is to say, all types of translation involve (1) loss of information, (2) addition of information, and/or (3) skewing of information.” I most heartily agree. It is almost always impossible to capture in the receptor language all that was meant in the source language. Similarly it is almost always impossible to render in the receptor language only what was meant in the source language. The translation of sentence (1) as They got his car dirty loses the Image of affectedness meant by the Spanish, and specifies the Image of possession, which is not meant by the Spanish. Translations like They got the car dirty on him or They got his car dirty on him also wind up changing the Image slightly, either through specifying unpossession or through specifying possession again. There is no good way in English to say exactly what sentence (1) says. Of course, any of the three sentences may be a good translation for (1) in some particular context, but they will be saying something slightly different for all that. Part of what makes translation such an extremely complex task is the fact that a translator is constantly faced with decisions about what he will consider an important component (or omission) of meaning in the source expression. He cannot render everything in exactly the balance it had originally—and that balance itself is a part of the Imagic meaning. Often it will be very awkward to render certain Imagic notions at all. At other times it will be possible, but only at the expense of upsetting the dynamics of the passage or straining the norms of the language, as well as usually introducing extraneous Imagic material. Translation is a continual compromise between the desire to render the source message faithfully and the desire to communicate well in the target language.

This way of looking at things also has implications for the traditional debate concerning idiomatic vs. literal translation (e.g. Beekman and Callow 1974:19-32, Nida and Taber 1969:1-31). Idiomatic translations will often use a target language expression that has the same Functional meaning as the source language expression, at the expense of obscuring some Imagic difference which could have been preserved. Literal translations attempt to keep such Imagic distinctions, usually at the expense of naturalness, since the distinctions will be awkward in the target language. I suggest that it is this, rather than a slavish adherence to the form of the source language, that lies behind much literal translation: the literal translator is eager and willing to pay a high price to render as much of the Imagic meaning as he can. This conception helps make it clear why some people so much dislike literal translations (because they do not sound natural) and others dislike idiomatic translations (because they do change the meaning).

4. Summary

In this paper I have argued that the Spanish construction exemplified in sentences (1)-(3) is not an example of PA and does not have the same meaning as the English construction exemplified by the glosses to those sentences. I have suggested that this implies that other cases where PA has been posited are quite possibly not best analyzed in that way, but as having a construction parallel to the Spanish one. I have also suggested that the notion of Imagic meaning and the fact that a paraphrase relationship can and often does coexist with differences of Imagic meaning imply that the way we do syntax should be re-examined to make sure that it is not based on a covert assumption that paraphrase implies semantic identity. And I have suggested that translation theory should allow for the fact that there is often no translation of a given expression into a given language that will convey all and only the Imagic meaning of the original expression.

Footnotes

¹ The following is a list of abbreviations used in this paper, including some that are introduced in the text.

1	Subject	GR	grammatical relation
2	Direct Object	H	head
3	Indirect Object	OBJ	object marker (“personal ‘a’”)
1p. sg.	first person singular	P	Predicate
2p. pl.	second person plural (and so on)	PA	Possessor Ascension
ACC	accusative (2-marking) or 3p. sg. accusative pronoun	PD	Possessor Deletion
DAT	dative (3-marking) or 3p. sg. dative pronoun	PO	Possessor Omission
ED	ethical dative	Poss	Possessor
ED-PO	ethical datives together with Possessor Omission	RG	Relational Grammar
		RN	relational network

Masculine forms (“he”, “him”, “his”) will be used to gloss 3p. sg. Spanish forms.

⁸ It would, I think, be in one sense in the semantic structure of the word fridge in that one of the specifications of fridge would be that fridges typically are personal or familial possessions. But that specification would be totally internal to the word fridge and would not be part of the semantics at the sentence level. In any case, it would not be specific as to who is the possessor. Possession might also be involved in the semantics of the phrase the fridge if it is one of the things contextually utilized to provide uniqueness or “definiteness”.

13 Robaron sus tres mil pesos: ‘They stole his 3000 pesos’ is grammatical, but only because the possessive is used in a “restrictive” sense. It is implied that there was a sum of 3000 pesos belonging to the man (usually all the money he had) that the hearer was aware of before the sentence was uttered. Le robaron sus tres mil pesos: ‘They stole his 3000 pesos from him’ is thus also appropriate.

Note also, that, as mentioned in Section 1.1.3.4, the possessor may be specified in any such case where a different person is specified as the victim of the crime (e.g. sentences (26)-(28)).

¹⁵ Under the following assumptions some sort of argument could be made from these facts for PD as against PO: (i) Semantics is articulated with syntax only at deep structure (or initial level), and (ii) Rules like Passive and Subject Postposing (or Preposing) are optional and not keyed to the presence or absence of a possessor nor to any relevant semantic features. Under such a model PO would have no way of getting needed information about the application or non-application of these rules, whereas PD could simply be ordered after them (i.e. they could be constrained not to apply to structures involving PD). In this as in other things the problem of accounting for the absence of possessors is very reminiscent of the problems of accounting for Pronominalization.

¹⁷ Although it has been pointed out that tests of this sort (involving conjunction and/or reduction) may not necessarily test exactly for vagueness versus ambiguity and that they sometimes give equivocal results, it is clear that they at least distinguish prototypically “vague” from prototypically “ambiguous” pairs of expressions. Thus, although passing the test does not prove that the two expressions are vague, it at least damages the position that they are ambiguous. In any case, the only alternative to a test like Lakoff’s seems to be fiat declaration, which I am willing enough to make. “They are vague.” So be it.

¹⁹ Comparable conceptual situations can be construed in many different ways (i.e. different perspectives can be taken on a scene and different facets of it singled out for explicit attention) both at the concrete level and with respect to the more abstract relations symbolized by ‘grammatical morphemes’. Conventional imagery of this kind is an important aspect of linguistic structure and leads to the situation where two languages code the same conceptual scene in semantic (hence grammatical) structures that differ greatly in specifics despite being functionally equivalent.” (Langacker 1980:33)

²⁰ In another sense, the Functional and Truth-value distinctions can be viewed as primary in that they are more easily accessible, both to the linguist and to the language learner. My experience as an adult language learner (which is quite compatible with what I remember of language learning as a child) is that one usually learns the “meaning” of an expression in the Imagic sense by observing its “meaning” in the Functional and Truth-value sense: seeing regularities in the usage and non-usage of the expression (or of its parts) and inferring or deducing the “meaning” of the cases where it is not clear from the cases where it is. (Whether, and to what extent, this is influenced by inbred predispositions to certain “meanings” as opposed to others, I do not know. Also, I am ignoring the important part that observation of such things as periphrasis and antonymy or even direct explanation through periphrasis or translation may play in all this.) Similarly it often proves necessary for linguists to argue from Truth-value or Functional “meanings” to establish an Imagic “meaning”, viewing it as a hypothesis justified by the fact that it elegantly accounts for the Functional and Truth-value distinctions. (Often, of course, important aspects of the Imagic “meaning” are assumed rather than argued for, and conveyed through periphrasis or translation. Many examples of this can be found in the preceding sections.) In a sense, then, the Imagic distinctions may be viewed as projections or extrapolations of Truth-value or Functional distinctions to cases not distinguishable by the Truth-value or Functional Criterions. However, once the language learner has mastered the expression, that extrapolation or hypothesis becomes part of the semantic structure of the expression. It is, I would claim, “psychologically real”—objectively there whether it is producing Functional or Truth-value distinctions or not.

I would claim, then, that what people do when they judge whether two expressions “mean” the same thing or not is to compare the two Imagic “meanings” (which are the psychologically real ones) and judge (by an extralinguistic conceptual faculty—see Section 1.2.1.1.4) the extent to which these two “meanings” are alike. We know that two expressions “mean” the same thing in a weakened sense (like the Functional sense in that it is a matter of degree and relative to context and purpose but unlike it in that it does not depend on functional interchangeability) when we see important similarities in their Imagic “meanings”.

²² For some speakers of English it is more felicitous to say I sat (or, laid) him down; for some the sentence in the text may actually be inappropriate. For such speakers, it would thus seem that the reflexive component of the prototypical act of sitting down is part of (at least on salient version of) the “meaning” of sit/lie down. Notice that this component of the semantics would not be given a separate overt coding. (For a good example of analysis showing the need to posit more than one version (subschema) of the semantics of predicates and constructions, see Lindner (1980).)

²³ I am not claiming that Imagic meanings always show up coded explicitly by some word or morpheme; see for instance the preceding footnote. However, I do believe that very often morphemes that have been treated as “meaningless” code some “meaning” in the sense of the Imagic criterion; some change of image or shift of conceptual viewpoint, increasing the salience of some elements in the conceived scene, and decreasing that of others down to the point of not specifying them at all; i.e. leaving them vague.

²⁴ The last clause of this sentence is question-begging to a certain extent: crucial to the idea of PA in Spanish is the proposal that sentences like (1) do have an initial Poss arc, though admittedly not a surface one. But the point is that if there is no semantic possession in the Spanish sentence (or perhaps even if there is—see the next section) there is no reason to posit any syntactic Poss arc. Similarly, if “affectedness” is not included in the English semantics, there is no reason to posit any GR_x arc corresponding to that of the ED of Spanish.

²⁵ The meaning with respect to whether or not the understood possessor is affected is not represented equally by PA and PD. Under PD the Spanish has an ED whereas the English does not; under PA neither does. If it could be shown that the English and Spanish sentences do in fact have the same meaning with respect to affectedness this could perhaps be made into an argument for PA over ED-PO. In fact it can be shown (Sections 1.1.2 and 2.2) that they differ exactly as predicted by the claim that the Spanish sentences do have ED’s.

²⁶ It might seem attractive to claim that these cases where I have talked about a different person from the possessor being the referent of the dative are actually cases of some sort of attenuated possession. For example, for all three sentences (1)-(3), on the reading where B is affected by what happens to A’s possessions, the usage is most felicitous where B has either physical possession of or at least responsibility for A’s possession. Thus (1) is most appropriate if the car is in B’s possession, even if it is not his car, and (2) when B is carrying A’s money, or at least responsible for it. Thus, it might be claimed, it would be appropriate to have a Poss arc attached to the nominal, and those sentences would also be examples of PA. The objection to that is that then there would be no way to distinguish between such cases of “attenuated possession” and cases of true possession. This would have the following consequences: The parallel with languages like English would be destroyed (Argument B); to be consistent we would have to posit Poss arcs in English sentences such as They stole all the money from him, and there would be no way to distinguish those Poss arcs (which cannot surface as possessive pronouns) from those associated with real possessives. And, in sentences like those to be discussed in the next section, there would be two Poss arcs, one somehow to be interpreted as attenuated and the other not, one able to suffer PA and the other not, both attached to the same nominal node. Thus I conclude that “attenuated possession” and real possession must be distinguished anyway, and that trying to extend PA to cases of “attenuated possession” is of no real benefit.

²⁸ Hollander is well aware of the problem I am discussing: he sees the alternatives as either “to assert the ‘form-content’ dichotomy against all usual better judgment”, or to succumb (as I have) to “the specter of the ‘organic’ view”, which would “probably end up by asserting that translation is impossible under any circumstances”, or, apparently preferably, to avoid the whole issue of meaning and discuss “how people react to the literary works themselves” (pp. 207-208). (Cf. Nida and Taber’s (1969) embracing of “dynamic equivalence” as the criterion for good translation.)

References

Casad, Eugene H., and Ronald W. Langacker. 1985. “‘Inside’ and ‘outside’ in Cora grammar”. IJAL 51.247-281.

Langacker, Ronald W. 1982. “Space Grammar, Analysability, and the English Passive.” Language 58.22-80.

————. 1985. “Observations and speculations on subjectivity.” In John Haiman (ed.), Iconicity in syntax, p. 109-150. Amsterdam and Philadelphia: John Benjamins.

————. 1986. “An introduction to Cognitive grammar”. Cognitive Science 10.1-40.

————. 1987a. Foundations of Cognitive grammar, Vol. 1. Grammatical prerequisites. Stanford: Stanford University Press.

Lindner, Susan. 1981. A lexico-semantic analysis of English verb-particle constructions with OUT and UP. UCSD doctoral dissertation, also published by Indiana University Linguistics Club.

[Maldonado, Ricardo. 1999. A media voz: Problemas conceptuales del clítico se. México: Universidad Nacional Autónoma de México.]

Tuggy, David. 1981. The transitivity-related verbal morphology of Tetelcingo Náhuatl: An exploration in Space grammar. UCSD doctoral dissertation.

© 2005, David Tuggy
All rights reserved. May be reproduced for nonprofit use
so long as it is not modified in any way.
Derechos reservados. Puede reproducirse para fines no lucrativos
siempre que no se altere en forma alguna.

¡Ethical Dative and Possessor Omission Sí Possessor Ascension No!

David Tuggy

Summer Institute of Linguistics