Volume 28 Issue Jan , pp. Volume 27 Issue Jan , pp. Volume 26 Issue Jan , pp. Volume 25 Issue Jan , pp. Volume 24 Issue Jan , pp. Volume 23 Issue Apr , pp. Volume 22 Issue Historica vol. Volume 21 Issue Historica vol. Volume 20 Issue Jan , pp. Volume 19 Issue Jan , pp. Volume 18 Issue Historica vol. Volume 17 Issue Historica vol. Volume 16 Issue Historica vol. Volume 15 Issue Historica vol.
Volume 14 Issue Jan , pp. Volume 13 Issue Jan , pp. Volume 12 Issue Jan , pp. Volume 11 Issue Jan , pp. Volume 10 Issue Jan , pp. Volume 9 Issue Jan , pp. Volume 8 Issue Apr , pp. Volume 7 Issue Jan , pp. Volume 6 Issue Jan , pp. Volume 5 Issue Jan , pp. Volume 4 Issue Jan , pp. Volume 3 Issue Jan , pp.
Volume 2 Issue Jan , pp. Volume 1 Issue Jan , pp. About the article Published Online: By using the comment function on degruyter. The languages of the world are seen as dividing into two classes: Hence most typologists must make an underlying specification of head-modifier order. The common factor which unites the typological approaches is the need to define the notions of head and modifier. Thus, Anderson departs from underlying structures which, as we have already indicated, are a version of CG using dependency trees.
As we have said, the only requirement which unifies these approaches is that they should avail themselves of a formal notation which permits the definition of the crucial concept of modification. It is therefore not surprising that ST has not given rise to typological work, since, among other things, its formalism for PS rules does not provide naturally 9 for a definition of heads and modifiers cf. Different, on the other hand, is the situation with R EST, since one of the major merits of the Convention used therein is that it does offer a means to define the relevant notions cf.
To my knowledge, comparative work in the system - a necessary preliminary to typological inquiry - has not yet been carried out, but there are signs that such investigations could have interesting results, since the theory appears to make different empirical predictions from its competitors. Thus, on the principle that the X of an is the head of that construction see Bresnan , Jackendoff , S will be the head of an , i. Vennemann , on the other hand, is forced to argue that the COMP is head of such a construction since it follows the embedded S in a consistently modifier-precedes-head language like Japanese.
Another case where predictions among the various theories differ concerns the relative position of nouns and adjectives. English is then treated as a, by no means isolated, exception to this general rule. A large part of Anderson , however, is devoted to showing that the adjective in English is in fact the head of an Adj-Noun construction, and that therefore English is entirely consistent from a typological point of view. See also Maling , Rosenbaum and the discussion in Werth this volume. It is perhaps in the diachronic domain that the typological approach to word order has had its greatest impact, particularly in the work of Lehmann , Vennemann , and more generally in the contributions to Li These scholars share an effort to establish a universally valid cycle of change, into which attested word order, and associated typological, shifts can be integrated.
See also Vincent for more discussion. It is perhaps not obvious that there is no logical necessity for requiring that both these "quite separate functions" be performed at once by the phrase structure rules. On the contrary, it is obvious that since in probably all natural languages at least some morphemes occur in two or more linear orders relative to one another in surface structure, some transformational rules will be required which have as part or all of their function to redetermine the linear ordering of morphemes.
The editors of the present volume would like to thank the Department of Linguistics at UCLA for their permission to reprint it here. By this proposal the structures defined by the phrase structure rules are conceived of as non-linear. Staal proposed that such a grammar was necessary for describing the free word-order phenomenon in Sanskrit, and also correct on intuitive grounds for universal grammar. The main purpose of the present paper is to present further evidence that linear order is absent in deep structure.
Three arguments will be presented. Such paradoxes or necessary indeterminacies can never arise if linear order is only established at the level of post-transformational structure. In other words, at their best grammars with linear deep structure are notational variants of grammars with non-linear deep structure. This will be shown through examination of analyses of the main constituent order of Amharic. McCawley, indeed, claims that all languages have either VSO or SOV main constituent order in deep structure, and languages that show surface SVO order are always synchronically derived from the former of these by a rule of V-NP inversion.
He gave basically two arguments in favor of his analysis of English as a VSO language. The second argument purported to show that several English transformational rules may be simplified if deep linear order is VSO. What we do find is that, if we posit and justify a posttransformational rule which assures SVO ordered surface structures in English, i. First it will be necessary to deal with the claim that English semantic structure is in VSO order. Only Lyndon pities himself b. The formulas of 3 show, not at all surprisingly, that they can.
Onlyx Pity x, x , Lyndon b. The custom leads to a sort of convenience in reading. But we could learn to read 3 as conveniently as 2. The five transformations are Passive, There-insertion, Subject-raising, Negative-transportation, and Predicate-raising. The same rule with VSO input has only one operation: Provided V-NP inversion can be independently justified, this argument is very neat: McCawley, attributing the argument to Paul Postal p. This rule relates in 4 a to the question b , and 4c to the question d.
The same rule will also relate the string e derived by transformation from the VSO ordered string e' to the passive sentence f. The larger part of the necessary ingenuity is called on to deal with Negative-transportation and Predicate-raising, and a smaller part to handle Subject-raising. We can focus attention on the purported difficulty with Subject-raising in English as an SVO language. McCawley's version of Subject-raising relates in 5 both a to b , and c to d. John believes his brother is a narc. John believes his brother to be a narc.
Art seems to admire Spiro. Compare Subject-raising to subject in 6b and 7b. Subject-raising to object with VSO order 1 3 24 b. Subject-raising to subject with VSO order 1 3 24 7 a. Subject-raising to object with SVO order 1 3 24 b. With the SVO analysis it does not seem to be at all the case that one rule is involved. The two cases of of subject-raising in 7 could only be awkwardly collapsed by the use of complex bracketing. As seen in 8 , the rule of Subject-raising which combines raising to subject and to object is just as simple, and just as unitary with an SOV ordered input.
Subject-raising to object with SOV order 1 3 24 b. In fact the same is true for the Passive and Theveinsertion arguments. Even the purported independent justification of the VSO grammar's rule of V-NP inversion would be an argument neutralized in this case. Who is Ed investigating now?
There is a fly in it. Shall we therefore argue that English is an SOV language? Actually the situation is even worse than this for the VSO analysis, and for the general problem of deciding just what the deep order of English main constituents is. This alternative will also employ the rule which puts the verb in second position. The effect of this rule is to restore SVO order after one-operation rules of Theveinsertion, Passive, and Subject-raising have applied. This rule moves the verb into position to the right of the clause-initial NP. There-insertion a fly is in it there a fly is in it by Passive power has corrupted men men power has corrupted by Rule 10 does not apply.
Notice that in this approach as in McCawley's both cases of Subject rais3 ing are raising to the immediate left. Perhaps the most interesting observation which McCawley made was that ten of the fifteen English transformations which he examined accept with equal facility either VSO or SVO ordered inputs. It is reasonable that this should be so, since transformations typically make no reference to linearity in locating the morphemes which they will affect. This is certainly 3 Postal He agrees with McCawley that under the VSO analysis they are combined, for which he gives the single structural analysis suitable for both cases: This is true of the three transformations just discussed, which McCawley believed to be simplified by VSO input structure.
The fact is that if we allow the linear order of the verb to be established by a late rule, any English transformation that would otherwise require this as one operation can be simplified. This may be considered an indication of the actual absence of linear order in deep structure. This will be seen in the next section through examination of the problem of the deep order of Amharic constituents.
For example, if X and Y are sisters occurring in different environments in both the orders XY and YX, and X is the nucleus of the construction e. N of NP , it is Y that moves, while X is stationary. This section is an abbreviation of an earlier study: Where Bach's examples are used here I have conformed them to the phonemicization commonly used in the literature on Amharic: As with English, there appears to be no deep structure evidence regarding the linear order of Amharic main constituents.
Bach presented several arguments. I wish to limit consideration here to the descriptive arguments, since unless these are effective there is no need to seek explanatory criteria in order to justify the VSO analysis of Amharic. In fact the first descriptive argument will receive no attention here, since it relies completely on the theory of 'gapping' proposed by John Ross , a theory which has been shown inadequate in work by Gerald Sanders and Andreas Koutsoudas Still, Bach's remaining arguments are persuasive insofar as they show the sufficiency, if not the necessity, of a VSO analysis of Amharic.
The following examples will make this clear, The examples of 13 show expansion of a genitive phrase, and those of 14 expansion of a relative clause. Attachment in either case says: The second argument that Amharic is a VSO language is said to "converge on the conclusion that there should be a rule of verb-shift," hence VSO deep structure. The material here is of potentially unbounded length, including all simple modifiers of bet. But this approach would be unsatisfactory due to the absence of precedents for left-to-riqht iterative rules in syntax.
With the VSO analysis we are able to achieve a simpler solution. On each cycle attachment takes place: The simplicity and availability of the VSO analysis, and its clear superiority to the SOV analysis seems to lead to the conclusion that Amharic is a VSO language, Bach can bring in his explanatory arguments to confirm the case for VSO order in deep structure. When the preposition stands before a clause, if the clause is in YSO order the preposition will be attached to the verb of the clause, and the verb always is shifted in the last cycle to the opposite end of its clause.
In other words, 21a should become 22a , and 21b should become 22b. By tree-pruning an S node would be lost in 22a. This is the direct statement of the observable generalization about the position of Amharic prepositions. The rule of verb-movement in the VSO analysis is an only indirect statement of the same generalization, made possible by the unnecessary positing of VSO order.
Note that transformations such as these are unambiguously interpretable despite the unlinearized variables they contain. The phrase structure rules establish only one sister constituent of a single type. Thus, barring cases arising from conjunction reduction which applies after other transformations: The A over A principle cf.
After these rules apply we will have The generalization which these rules express is therefore no argument for any particular linear ordering of main constituents. They apply when and wherever their environments are met. It is enough for the present purpose to have seen that, whether the non-linear solution of the problem of the deep order of Amharic main constituents is considered to be a more general solution or not, the opposing solutions with differing linear deep structure are reducible to the non-linear solution once the actual generalization about preposition-lowering in Amharic is completely recognized.
The task of Section 4 will be to show how a grammar which employs post-transformational linearization rules attains a higher level of generality than a grammar which establishes linearity in deep structure. This will be shown through examination of a long-time problem in English transformational grammar: There is apparently no case where a linearity paradox of 46 GROVER HUDSON this sort arises in which an even moderately convincing argument has been presented for choosing one or the other surface order as the basic, deep structure order.
The problem of the deep linear order of English dative and object noun phrases is a typical one. It has been recognized since work by Fillmore , and has recently frustrated Jackendoff and Culicover The alternation between the order of dative and object NPs is seen in sentences such as these: Jack sent it to the Red Cross.
It is apparent that two factors interact to determine the linear order of the dative and object NPs: Sentences a - c illustrate the effect of realizing the dative with to. Sentences c - e illustrate the effect of the length and composition of the object NP on linear order.
This factor is somewhat more complex than c - e indicate. The order of c would be more acceptable, even completely so, if, for example, the object NP were of greater length, say, as in 29f. The order to-dative followed by object is even preferred if the reverse order leads to ambiguity as is seen in the pair of sentences 29g and h.
In a transformational grammar which makes the dative without to basic, and also establishes linearity as a function of the base rules, sentences which in the course of derivation get a to-dative require a transformation to subsequently put the to-phrase after the object. If the to-dative position is established as basic and if in the course of derivation we delete to, a transformation is necessary to move the dative in the opposite direction.
This is summarized in 30 , where the two-directional arrows indicate that no particular phrase-marker is here considered basic. Nothing hinges on whether there is a VP node. This is just a notational convenience here which permits us to set apart subjects; these, like objects, will have to be marked some way in any theory.
Nixon sent an envoy to China. Nixon sent to China an envoy. Nixon sent him to China. China in these sentences is, of course, not a dative, but an adverbial NP, as shown by the ungrammatically of 31a. It expresses roughly 'direction'. Like the to-dative the to-adverbial NP is definitely excluded from this position if the object is a pronoun.
But, again, the pre-object position becomes preferred as a consequence of the length and composition of the object NP. This last similarity with the to-dative is illustrated in sentences 31f - h. Nixon sent to China his most trusted advisor, Henry Kissinger. Nixon sent to China a secret envoy who would arrange a summit meeting. The fundamental parallel is that if the dative is realized as a to-phrase, it typically follows the object, as does the adverbial to-phrase.
Parallel treatment of the to-dative and the to-adverbial could be expressed in a number of ways. Most would agree that the unmarked position of the to-adverbial, as with other adverbial phrases, is after the object, since the cases where it is just after the verb and before the object examples 31f, h are clearly the special cases. Hence the movement transformation that applies to the dative on the basis of whether or not it gets to seems in effect to 'refer back' to this phrase structure rule which in a grammar with linear deep structure first establishes the position of the to-adverbial.
We could employ a convention of some sort or a notation which will relate the linear ordering function of these two rules in different parts of the grammar. But such devices are only necessary if deep structure is linear. It makes no difference that these distinctions are mostly characteristic of written, rather than spoken, language. In an alternative grammar where deep structure is non-linear, the phrase structure rules establish hierarchical but not linear structure, and linearity is only established by post-transformational rules.
In this alternative not only does the necessity not exist for a convention or notation to link up the linearization function of separate rules in different parts of the grammar, but the linearization rule that establishes the linear order of both the dative and the to-adverbial is itself a statement of the generalization involved. We should desire as a general principle that the grammar perform identical functions identically. Following are rules which express the basic generalizations which hold for the linearity of elements of the sort in question.
Again the commas indicate that no relationship of linearity is established for the items that they separate. The mark is used here to show linear order. These two linearization rules may be collapsed as 33 , without use of the dative label. Rule 33 is of course generalizable to certain other prepositional phrases, not just those with to. It is easy to extend this line of reasoning to numerous other instances in order to show the greater generality of grammars which employ linearization rules.
They build structures the linearity characteristics of which are already specified by ordinary phrase structure rules.
In the VSO analysis of English the rule of V-NP inversion gives the right result if the linear ordering step is omitted from the transformation that introduces theve , treated as an NP, at the right of V. In other grammars with other deep orders proposed in Section 2 different late rules perform the linearizing operation depending on where we choose to insert theve.
Therefore it is important to point out that these rules are no more complex than the phenomena they describe. But linearization rule 33 performs at once both the 'movement' function of the transformation, and the linearizing function of the standard phrase structure rules. We found that while this VSO analysis of English cannot be rejected on grounds of descriptive adequacy, the state of syntactic theory is such at present that it nevertheless cannot be selected over analyses with other theoretically possible orderings of the same constituents.
The transformation which fills the role of verb movement, as well as the role of preposition movement in the SOV analysis of Amharic, and which most directly captures the correct generalization about Amharic syntax is simply oblivious to the deep linear order of the constituents to which it refers.
The structural index of this more general transformational rule will as easily accept SOV, VSO, or other theoretically possible main constituent orders. This was shown through examination of the English dative transformation. The linearizing function of this transformation duplicates the linearizing function of a phrase structure rule. This line of reasoning can be extended to other cases in which transformations in a grammar with linear deep structure perform linearization, since this function will often duplicate the work of the phrase structure rules in such a grammar.
Hence the deep structure of generative grammars is non-linear. For these and other arguments see Sanders and a. This article, though written independently of Sanders' work, nevertheless breaks little ground not broken by Sanders. But this paper has argued that application of linearization rules on pre-terminal lines is irrelevant to the subsequent application of transformations, and even that the use by the transformations of the information which linearity provides constitutes loss of generality.
Such facts, if they existed, would indeed be sufficient evidence against the argument of this paper and others in this volume, since the most precise sense of the notion "unordered deep structure" is that expressed in Sanders' Invariant Ordering Constraint, according to which, once any order is determined between two elements in a derivation, that order is fixed Sanders There are no reorderings.
Metathesis is "the simplest example of reordering", and "surely no one would argue that the underlying phonological representations of formati ves are unordered. But if they are ordered and if metathesis is a possible phonological rule, the invariant ordering hypothesis must make the peculiar claim that reorderings are allowed in phonology, Where this argument goes wrong is in the words "surely no one would argue The linear temporal-spatial modes in which we work in expressing such things makes the latter formulation seem simpler, since we refer to the morpheme as having the lexical form XabY, without noting that this representation has two parts: The non-linear analysis makes both statements explicitly.
True, most morphemes, those with invariable segment order, can be conveniently and properly expressed in lexical form with order, e. XabY or XbaY, whichever that invariable linear order is. In these cases the linearization rule is implicit, since order is explicit in the representation. The following example from Hudson illustrates such description of metathesis.
The brackets are needed since a representation Xa, bY would be wrongly interpreted as allowing the two possibilities XabY or bYXa, The bracketed structure X[ a, b ]Y seems awkward, and is only necessary when we stay on the horizontal dimension. The second argument against non-linear deep structure is found by Bach in the sentences 1 , 2 , 3 , and 4 , below.
We are concerned with the interpretation of these sentences in which John and him are coreferential, as indicated by the subscript x's. Bach's assumptions are that 3 and 4 are derived from 2 , which is derived from 1. Let me clarify this argument by noting that 2 , 3 , and 4 are all good sentences on the interpretation of non-coreferentiality between him and John.
This is clear when another proper noun is substituted for him: Hence 3 and 4 cannot be considered derived from 2 , but from the non-linear and non-pronominal ized structure equivalent to 2 , 2'. Since the non-linearity of sentence 2' is not crucial here, I have for convenience neglected to express it as a non-linear structure, i. Another pair of examples presented by Bach is part of the same argument, sentences 5 and 6: I will still help himx.
Sentence 5 is ungrammatical since the pronoun he coreferential with Harry both precedes and commands its antecedent. This underlying ungrammaticality Bach offers as explanation of the asterisk on the derived sentence. Again, a deep linear order is said to explain surface facts. However, I think that it really does not need much explanation at all, since it is not ungrammatical. It is therefore unlikely that the proper noun Harry would not be pronominal ized when fronted in this topical ized phrase. As evidence for this note that 6 is not all bad in the way that 3 and 4 are; in fact, with repetition and a little imagination sentence 6 begins to sound okay.
This difference in status between 6 and 3 and 4 is a fact unexplained by the theory in which the ungrammaticality of these is explained identically. Why do you think he's guilty? Furthermore, sentence 6 , if in fact it is ill-formed, would not thereby be counter-evidence to non-linearity in deep structure, since its ill-formedness would be consistent with the clearest expression of the non-linear hypothesis, Sanders' Invariant Ordering Constraint. This will be seen better in the next section, regarding Bach's next argument.
The example given by Bach is 7 , coming from 8: Sentence 7 is said to be a violation of the constraint, and "illformed"; the word "ungrammatical" is not used. I mention this, since many persons find this sentence and others given by Postal as violations of his constraint to be okay, or at least no worse than, say, clustered negatives, and extra-embeddings, which are complex, but not ill-formed in the sense 'ungrammatical'. Sentences 5 and 6 , then, are therefore also consistent with invariant ordering; 6 is supposed to come from 5 , by inversion of cruel to Harry and though he expects me to be.
Notice how reasonable it is that such inversions be disallowed. This is just what invariant ordering requires. Consider Postal's example Who did Joan talk about his illness to? As Postal says, both can to who. Clearly it would be inefficient for the grammar to first move to who to the right, then extract who and move it to the left.
Instead the grammar is directed to put things where they belong, the first time. A global rule interpreting the input and output of wh-movement treats a symptom of the problem only. And note that the example above Postal's 16b is difficult to interpret whether coreferentiality between Who and his is assumed or not.
The problem is the 'stranded' preposition.
Many of the sentences said to violate the constraint on wh-word movement seem all right, if complex, e. The questions, such as Postal's 2a , above, and 7 are more difficult to interpret for reasons discussed by Chomsky Others are problematic for reasons unrelated to the proposed constraint on wh-word movement, such as Who did Joan talk about his illness preposition.
This sentence is bad pragmatically. It has an embedded passive sentence, and the function of the passive, topicalization, doesn't often exist for embedded sentences. Consistency of topicalization would require either The tigers, which a t e. Similarly with the example Who did Joan claim his mistress had been visited by? The first step is pragmatically unmotivated.
Another example starred by Postal is: An e n g i n e e r X who X Mary claimed his X mother refuses to let out after dark arrived 10a. This sentence would be a lot better if it didn't have the 'heavy' modifier on the subject. The subject of a sentence is ordinarily a topic, and thus unlikely to be appropriate with such extensive modification. Notice how much better is a sentence, structurally also a violation of Postal's constraint, when we shorten the relative clause modifier of the subject and extend the verb phrase focus: That little boy whom Mary claims his mother pampers is nevertheless knocking on our front down there in the rain door!
Among other things, her hypothesis accounts for the difference in placement of main stress and high pitch on examples like these": This argument against non-linear deep structure totally relies on the claim that stress assignment can be done, at least in general, along the lines of Bresnan's analysis. That is, stress is assigned as a function of syntactic structure. But a description along these lines can only be in statistical terms" Bolinger We can take sentences 15 and 16 to illustrate this point. Bresnan's analysis will place main stresses as in these sentences.
But what then of sentences 17 and 18 , which have the same structure, but different stress? One who would maintain Bresnan's analysis might wish to claim that 17 and 18 , unlike 15 and 16 , are 'contrastive', or in some sense special, marked, deserving special treatment. But while this may be true for 17 , it is doubtfully so for 18 , which seems just as probable a sentence as And, with Bol inger, we would question as adequate a grammar which merely accounts for most of the sentences of the language, and leaves as exceptions innumerable quite common others.
Again the foundation of the argument against non-linear deep structure proves faulty. Tai argued that Chinese is in deep structure an SOV language, the relative clause is after its head, and pronominalization is then always forward. The argument for linear deep structure in all this is supposed to be a resulting simplification of the pronominalization rule, which operates only forward. The rule which, after pronominalization, would invert head nouns and relative clauses is said to be universal for SOV languages Bach ; Tai It is important to note that this 'universal rule of relative clause preposing in SOV languages', as not part of the grammar of specific languages, does not itself succeed in simplifying the grammar of Chinese as an SOV language, since Chinese does have some post-posed clauses Tai and footnote 7.
These would have to be derived by a language-specific blocking of the proposed universal rule. The argument from Chinese as an SOV language for deep linear order Bach's argument; Tai did not consider non-linear alternatives depends on the claimed simplification of pronominalization. Let us compare the relevant parts of the two grammars of Chinese, that with deep SOV order, and that without deep order. Since the language does have some post-posed relatives, this possible advantage is immediately neutralized in Chinese by the necessity to block some applications of the universal rule.
But this sort of advantage, arguing from universals, can be claimed for the analysis in which Chinese deep structure is non-linear. For pronominalization is universally into a relative clause, on whichever side of the head noun it stands. In Turkish, it is just those clauses that have clause-initial complementizers ki that can extrapose. Clauses with preposed complementizers can extrapose; clauses with postposed complementizers cannot.
Bach then says of "grammars with unordered structures and ordering rules", that "since extraposition affects not just order but also dominance relations, it must be a regrouping rule ordered before any order has been defined on the structures. The rule cannot be constrained to operate on clauses containing inparticular but preposed complementizers.
As Bach says, "Surely, this is turning things on their head. Unfortunately for this argument against such grammars, such a 'separation of levels' requirement has not been suggested. Sanders' Invariant Ordering Constraint, which Bach specifically refers to as "the most precise hypothesis about unordered base structures" Bach What this means is that the complementizer can certainly be ordered before its clause prior to the movement of the clause, with its preposed complementizer, in extraposition. Bach's sixth argument is useful since it forces us to clarify the meaning of the notion 'non-linear deep structure'.
Probably because of the vagueness in the notion 'deep structure' itself, the former term has been often misunderstood. In Section 5 I wrote, "full generalizations about the linear order of elements can only be captured at the post-transformational level. The wider, higher level of the relationship of the clause, with its complementizer, to the rest of the sentence, can still undergo at least the transformation of extraposition. Simply put, the point about linearity of deep structure is whether transformations change linearity, and the answer to this is that they do not, or at least there is no evidence of this.
This means that the selection of a particular complementizer would be tantamount to linearizing it. This seems reasonable, since complementizers are grammatical morphemes with no intrinsic selectability. Selection of grammatical morphemes may ordinarily be dependent on linearization, rather than vice versa.
If the problem of misunderstandings is great this will explain how the interdependence came about in language evolution via natural selection, much as humans are generally believed to have evolved upright stance. For discussion of the issue, see Sampson On the other hand, if preposed complementizers make extrapositions easier to comprehend, they are reasonably a factor in learning extrapositions, and hence available for incorporation as a condition on the synchronic extraposition rule.
The differentiation of functional factors which control diachronic evolution only, and those which contribute to synchronic development is sure to be an issue for future research. We have examined Bach's six arguments against non-linearity in deep structure, and found each of these wanting. If it is, shouldn't there be better evidence of this claim than the six arguments discussed above?
Since Bach himself did not find any of these strong enough to include in his section of counter-arguments, I have not considered it necessary to reply to them here. Anyway, they all seem to me also faulty. For example, arguments on pp. According to Bach p. Elsewhere we read that "There is something inherently linear in language. This paper will be concerned with the manner in which these ordering specifications are assigned by grammars and with some of the metatheoretical constraints on ordering which are required for the adequate explanation of natural language data.
I am indebted to Andreas Koutsoudas for his valuable comments and criticism, and have also benefitted from the discussion of various ordering problems with Emmon Bach, Robert Lefkowitz, Gunter Schaarschmidt, and James Tai. This paper has previously appeared in Papers in Linguistics 2, We would like to thank Anthony Vanek and Linguistic Research, inc. SANDERS the claim that the most general and most revealing grammatical theories are precisely those which assume that all terminal phonetic elements are ordered and that all linguistic ordering relations are derivationally predictable and invariant.
These constraints all appear to be independently reasonable and well-motivated. The first relation is symmetric, or commutative, and has the rough interpretation 'is cognitively or psychologically associated with'. We may call this relation grouping, or co-constituency, and symbolize it by a comma between bracketed arguments. I know of no evidence to the contrary, and there actually appear to be a number of very strong arguments for making equivalence an explicit condition for the wellformedness of all derivations. Each ordering rule expresses an empirical generalization about the order of the constituents of some grammatical construction.
Thus, for example, rule 4 expresses the generalization that in constructions consisting of a determiner and a noun it is always the case e. It should also be noted that ordering rules are defined here in such a way that ordering relations can be assigned only to sister immediate constituents of the same construction. Like other rules of grammar, therefore, each ordering rule is empirically justifiable in terms of its specific contributions to the general grammatical mapping function that specifies the proper pairings between the interpretable semantic representations of linguistic objects and their interpretable phonetic representations.
This general system, with its two relations and one inter-relational mapping schema, is sufficient for the reduction of all possible theories of ordering. We are thus able now to discuss the alternative assumptions of these theories in an explicit and notationally uniform manner. Concerning the predictability of ordering relations, there are only three logically-possible alternatives: It goes without saying that there is no natural language which has this property, and even an artificial language 4 of this sort appears to be inconceivable.
The possibility of complete prediction has in fact been explicitly demonstrated already, both for invariant-order grammars by Hockett, and for variable-order ones by Chomsky, However, since it is possible to predict almost anything, given the right assumptions and initial conditions, we will assume here that the only interesting questions about the prediction of ordering relations involve not the possibility of prediction, but rather its explanatory significance. We will take up these questions shortly. For theories which observe neither of the completeness constraints, semantic and phonetic representations are relationally nondistinct, both being strings of elements that are partially ordered and partially grouped.
For theories which observe both of these constraints, these two modes of terminal representation are mutually exclusive not only in their constituent property or categorial elements but also in the relations which hold between them. There appears to be no presently known evidence against either of the two completeness constraints. Concerning the derivational variability of ordering relations, finally, there are only two possible assumptions: The latter assumption allows for the possibility of rules effecting the derivational reordering, or permutation, of constituents; the former precludes any such rules.
By translating these alternatives into the common metalanguage of derivational ordering, we have already shown, in effect, that, in terms of ordinary descriptive capacity at least, any fact which can be accounted for by a theory that assumes unpredictable, variable, or incomplete ordering can also be accounted for by a more restricted derivational theory which observes both the completeness and invariance constraints.
We turn now to some of the positive evidence suggesting that these constraints are not only possible principles of grammar but also necessary ones. The Phonetic Completeness Constraint implies that some such prediction can be made about every constituent of every terminal phonetic representation. SANDERS The Invariant Order Constraint implies simply that all predictions about ordering must be true -- in other words, that contradictory orderings can never be assigned to the same structure, and that each ordering rule must be consistent with all other rules and theorems of the grammar that includes it.
However, even if we ignore their superior restrictive and motivating power and assume that the burden of proof rests equally on all possible hypotheses about ordering, it would still be possible to demonstrate the truth of the completeness and invariance constraints by showing that there are some facts about languages which can be adequately explained only by theories which are inconsistent with their contradictories. With respect to semantic completeness and invariance at least, there appear to be a number of facts of this sort. Thus if, in contradiction to the Semantic Completeness Constraint, some orderings were unpredictable in natural language, then it would be possible that some grammar might generate or accept a pair of pre-transformational structures which are related to each other in the manner of 7a and 7b.
This strongly suggests that it is a fact about natural languages that they do not have distinct underlying constructions which differ only in the order of their constituents. Since ordering is predictable from the properties of constituents and their grouping relations, it is logically possible that ordering rules might apply at any or all levels at which such properties and relations are specified. For simplicity, though, we will consider the only available choices to be the mutually-contradictory hypotheses of Deep Ordering and Surface Ordering.
The Surface Ordering Hypothesis asserts that ordering rules are applied only subsequently to the application of other transformations, and that the latter are thus defined on structures that are wholly unordered. The basis for choice here is quite straightforward, and the relevant evidence is abundant and varied. The conclusions following from this evidence, moreover, appear to be consistently the same, namely that the hypothesis of Surface Ordering is correct and that the hypothesis of Deep Ordering is false. Moreover, Hockett's demonstration of the complete or at least near-complete predictability of orderings from superficial groupings appears to have gone wholly unchallenged thus far; nor, it seems, has any real effort been made to justify any postulation of pre-superficial ordering, or to show how an observed regularity in superficial ordering might be explained by any theory that permits such postulations.
But, apart from all of this, there is even more important evidence which can be brought to bear on the question of Deep vs. Surface Ordering, evidence which shows that Surface Ordering is necessary for the expression of generalizations not only about order but also about many of the other significant properties and relations of linguistic objects.
In order to falsify the Surface Ordering Hypothesis it would be necessary merely to show that if some transformation is assumed to depend on a prior ordering of constituents a more general explanation of some facts results than if the contrary is assumed. I know of no explicit ar5 guments to this effect , and I will attempt to demonstrate shortly that what would appear to be the strongest implicit one is actually invalid. It would appear that Chomsky's very brief and inconclusive remarks on ordering Chomsky, One rather simple example should suffice to illustrate what appears to be the general situation here.
Thus, consider the fact that in English all superficial subjects are ordered before their superficial verbs -this being true, of course, regardless of the underlying properties or constituency relations of the subject. Thus English has alternations of the active-passive type.
SANDERS follow in this grammar from different sets of rules -- in one case from a pretransformational ordering rule, in the other from that rule in conjunction with some reordering rule. Thus the only way that a Deep Order grammar might hope to prevent such loss of generality with respect to alternation sets of this type would be by the postulation of a base order for the set which differs from the surface order of any of its members. But any such treatment would clearly be arbitrary, excessively complex, and grossly unnatural, since it would require an ad hoc ordering rule and two ad hoc reordering rules to do exactly the same work that a single entirely natural surface ordering rule can do.
Moreover, any attempt to extend this approach to accomodate facts about other constructions in the language would quickly lead to the postulation of pretransformational structures whose orderings are wholly unrelated to those of the surface structures which are derived from them, with the further result that the base orders required to preserve generalizations about some alternation sets will be incompatible with those required to preserve generalizations about other sets.
However, such a constraint, which could be justified only in grammars without rules of formation cf. The fact that natural languages of this sort are apparently non-existent thus provides further empirical support for the particular theory of derivational ordering which has been proposed here, since this theory requires that all ordering rules must be applied at a level of representation at which raised and non-raised subjects are completely non-distinct -- the level of syntactic surface structure -- and thereby explains why they are always governed by exactly the same principles of ordering.
To bring this principle to bear on alternations of the active-passive type, however, it is necessary to factor 9 into its constituent processes of Subject Lowering and Nonsubject Raising, where the latter could then be reduced, along with 10 to a single general process of nominal raising. Postal, ; Sanders, , a. Thus, while X may include subordinate nominals, it may not have any nominal as one of its immediate constituents.
In the light of the existing evidence, in fact, it would appear that the assumption that any major principles of grammar are applicable to ordered structures has no significant logical or empirical basis at all. If the Surface Ordering Hypothesis is correct, as appears to be the case thus far, then this is a fact which itself requires some explanation.
Deskriptive Deutsche Syntax - Lineare Und Hierarchische Struktur (German Edition) [Vanessa Ruhrmann] on www.farmersmarketmusic.com *FREE* shipping on qualifying offers. Buy Deskriptive Deutsche Syntax - Lineare Und Hierarchische Struktur by Vanessa Kindle Edition . Paperback: 24 pages; Publisher: GRIN Verlag (11 Oct. ); Language: German; ISBN ; ISBN
The three principles of ordering that have been proposed here all play a role in facilitating the expression of true generalizations about the ordering of constituents in languages, and help to motivate and direct the continued search for higher levels of explanation in all areas of grammar. It is clear, though, that the invariance principle has a much wider range of presently-testable implications than the others, and a much greater capacity to restrict the set of possible grammars and the set of possible explanations of particular linguistic data.
On the basis of present evidence, moreover, it is also by far the most wellsubstantiated, I believe, and thus can be most strongly maintained as a 8 Among other things, rule 15 serves to explain why in English — unlike German, for example — verbs always precede their particles regardless of whether the particle is superficially raised or not. For further discussion of the extraposition and ordering of particles and nominals, see Sanders b.
I assumed there that these arguments would have to involve the existence of one or the other of the following situations. The first situation, which depends on the acceptance of rule-specific conditions on the relative order of application of transformations, would involve the necessary application of some rule to different orderings of the same construction at different levels of derivation. Arguments suggesting the necessity of this reduction, such as those presented in Sanders , would thus be superfluous here.
Ross in his seminal paper on "Gapping and the Order of Constituents" a. In this paper Ross addresses himself to the explanation of two very 10 about verbally-reduced coordinations interesting related observations in natural language. Some of the particular explanatory values of the latter assumption are illustrated in Sanders and Tai with respect to the treatment of certain facts about coordination, topicalization, dislocation, and relativization in various languages.
These observations, of course, constitute only a very small and quite arbitrary sample of the relevant facts about coordination in natural languages. SANDERS First, he reports that, while various languages exhibit one or more of the superficial reduction patterns 16a , 16b , and 16c , it is apparently the case that no language has reductions of the form 16d. On this basis, four language12 types are found: However, we will not be concerned here with the restricted nature of Ross's chosen domain of explanation, or with the viability of his observations and assumptions within the larger and more natural domain of coordination reduction in general.
There is also a fifth type, of course, consisting of languages such as Chinese, Hausa, Lebanese Arabic and Thai, in which there are no verballyreduced coordinations at all. These languages differ systematically from those of the other four types in a number of other respects as well, these differences being explainable, as proposed in Sanders and Tai , by their observance of a general Immediate Dominance Condition on identity deletion.
Furthermore, if the reported facts about Quechua are correct, then this language establishes the existence of yet a sixth distinct type, which differs from the other five by having all four of the reduction patterns of Ross's directionally-restricted verbal deletion rule, which he refers to as a "gapping" rule, can also directly account for two of the three reduction patterns of Russian and for one of the two patterns of Hindi, since Russian has both reduced and unreduced coordinations of both the English- and Japanese-types, while one of the Hindi reduction patterns and all of its unreduced clauses are of the Japanese-type.
The problem then is to find some way of accounting for the occurrence of the third Hindi-type: This is illustrated , for example, by the following types of well-formed reductions in English: To solve this problem, Ross makes a number of additional assump14 tions , the most important being 1 that prior to the applicability of reduction the clauses of Japanese are orderd SOV and the clauses of of the other three languages are ordered SVO; 2 that the general rule of verbal reduction can be applied to these ordered structures, yielding, in accordance with its assumed directionality restriction, the reductions SO and SOV in Japanese and SVO and SO in the other languages; 3 that there is a reordering rule, optional in Russian, conditionally 15 obligatory in Hindi, and universally non-reversible, which effects Serious objections could be raised against each one of these assumptions.
In the context of the present discussion, however, it is fortunately unnecessary to do this. Alternatively, the rule of object preposing could be assumed to be unconditionally optional and unordered with respect to verbal reduction, provided that it is also assumed that there is a general verb-final well-formedness condition on the surface clauses of languages of this type. In the context of variableorder grammar, the latter treatment would appear to be preferable, since it allows for the differentiation of Ross's four language-types in terms of only two general principles: The basic ordering problems for languages of the Hindi-type are not explicitly dealt with in Ross's study, and he assumes there that there are actually two distinct rules which effect preposings of objects in these languages, one being optional, pre-reductive, and shared by the Russian-type, the other being obligatory, post-reductive, and unique to the Hindi-type.
Moreover, he also assumes that the first of these preposings can be subsumed under a more general rule of" Scrambling", which optionally effects a random reordering and regrouping of the major constituents of clauses. However, this Scrambling rule, which Ross has proposed b in order to account for certain facts about languages such as the variety of literary Latin used in the odes of Horace, is neither empirically nor metatheoretically justifiable, some of the reasons for this having been noted by Ross himself b: Actually, this appears to be only a special case of a much more general principle which applies to all superficial alternations in the order of nominal constituents, and which seems capable of also explaining the apparently universal non-occurrence of right topicalizations in natural language.
For additional examples of such general principles for the prediction of directions of inference, see Sanders b. This is because the reduction rule applies here only to unordered structures, and because each of the ordering rules applies to sister clauses and sister phrases alike, and to the constituents of ordinary simple sentences as well as to those of reduced and unreduced coordinations. Among other things, therefore, this theory provides a very natural explanation of Russian: Some of the products of this deletion transformation will fall within the domains of certain general rules or metarules of associativity, which will justify appropriate regroupings of their constituents.
For a detailed discussion of these and other possible principles of coordination reduction, see Tai This provides further evidence in support of the assumption that verbal reduction is only one of the many intrinsically disjunctive subrules of a single general rule of coordination reduction.
With respect to the Russian-type rules in 19 , it should be noted that any negative variable of the form -X stands for any string which is not analyzable as X. The general relationship between clause- and phrase-ordering is most readily illustrated here by the case of verb-object ordering in English and Japanese.
The principle of uniform analysis is illustrated by the analysis of a verbally-reduced coordination with respect to the first ordering rules of English and Japanese given in 20 , by the three possible analyses of this structure with respect to the first ordering rule of Russian given in 21 , and by its two possible analyses with respect to the first ordering rule of Hindi given in Each of these distinctively analyzed representations will be mapped by the appropriate first ordering rule for its language-type into a distinct particular sets of language-specific ordering rules. This rule will apply to the unordered products of verbal reduction 18 to yield, in full conformity with the uniform analysis condition, all four of the ordered reduction patterns of 16 , as well as a large number of the other well-formed clause- and phrase-orderings of Quechua.
Moreover, the operant rule here is, in fact, simply the universal maximally-unmarked rule of Alternative Ordering, a rule which assigns a random order to each pair of constituents which remains unordered subsequent to the application of every one of the more marked, or more fully specified, ordering rules of any given grammar.
Thus a Quechua-type language would differ from the four types discussed by Ross simply by its possession of a smaller number of marked rules for the ordering of clause constituents. The important point here, though, is that the principles of unordered reduction, and the more general principles of completeness and invariance are all consistent with all possible degrees of freedom in the ordering of the constituents of superficial constructions, and that these principles are capable of providing highly general explanations of the facts about languages at any and all levels of the scale of relative freedom.
The second ordering rule for each language-type will then map the appropriate partially-ordered structures into the correct fully-ordered representations of the gram21 matical verbally-reduced coordinations of that language-type. This theory makes no postulations of reordering and is thus completely consistent with the Invariant Order Constraint. Ross's treatment of verbal reduction appears to achieve the highest level of explanatory generalization that is possible for any theory which assumes that ordering relations are assigned before rather than after the application of deletion transformations.
Thus it is not only the 21 In one of the three derivations for languages of the Russian-type, the sequential application of rules i and ii yields a structure in which one pair of constituents is still grouped rather than ordered. However, since all ordering rules are obligatory, and since this grouping satisfies the structural-description of rule i , this rule will properly apply again here to complete the assignment of correct orderings to all groupings. SANDERS case that the given facts about verbal reduction fail to falsify the Invariant Order Constraint, but also that they can be fully accounted for, it seems, only by means of theories which are fully consistent this principle.
In the light of the existing evidence, each of the three proposed constraints on constituent ordering would thus appear to be not only a possible universal principle of grammar, but also a necessary one. The common approach to this problem has been to propose A preliminary and rather truncated version of the theory presented in this paper appears in Peterson ; since that time it has undergone some revision and considerable expansion.