Transformations and sentence markedness
著者（英） Kyoko Yoshifuji
journal or
publication title
Core
number 12
page range 43‑67
year 1983‑03‑20
URL http://doi.org/10.14988/pa.2017.0000016407
43
T r a n f o r m a t i o n s a n d
S e n t e n c e M a r k e d n e s s
‑司aムa噌
az d
u
r十i.
i
T n
QU
o y o
' k
k y o
1. Introduction
The objective of the present paper is to explore the interrelationship between transformations and sentence markedness， by comparing a set of sentences derived from a single underlying structure via transforma‑ tions， and proposing a tentative ordering of their markedness degree. In explaining the generation of surface variety from a single deep structure， it seems appropriate to distinguish the formation of the most unmarked sentence from the formation of the rest of marked sentences. In this paper the following two stages of transformational processes are assumed to be involved in the derivation of sentence variants:
Underlying Structure (1st stage) 号
Formation of the Unmarked Declarative Sentence (2nd stage)号
Formation of Marked副SentenceVariants
Our second stage deals with those variants created by such predication‑ internal transformations as Passivization， Raising， etc.. We exclude from
44 Transformations and Sentence Markedness
this stage those produced by what may be called predication‑external (i.e.， pragmatic) transformations， such as Topicalization， Left Dislocation， and Right Dislocation. The discussion proceeds as follows. First we shall consider， by comparing the underlying structure of three represent‑ ative transformationalmodels， how an unmarked declarative sentence is generated， and then go on to discuss in terms of markedness" of applied transformations the formation of other marked sent巴nces.Finally， a principled explanation for their markedness order will be suggested.
2. The First Stage of Derivation
Let us:first note the following set of sentences， which we shall discuss throughout the paper:
( 1 ) a. John believes that Natalie has eaten the cookies.
b. John believes that the cookies have been eaten by Natalie. c. John believes Natalie to have eaten the cookies.
d. John believes the cookies to have been eaten by Natalie. e. That Natalie has eaten the cookies is believed by John. f
. That the cookies have been eaten by Natalie is believed by John.
g. 1t is believed by John that Natalie has eaten the cookies. h. It is believed by John that the cookies have been eaten by
Natalie. i
. Natalie is believed by John to have eaten the cookies. j. The cookies are believed by John to have been eaten by
Natalie.
All the sentences here are assumed to have derived fwm one and the
Transformations and SentenceMarkedness 45 same underlying structure， and， if so， their surface di妊erences are as‑ cribable to the application of transformations. Despite the surface variety， however， sent巴nce (la) should be the most basic of all， since it best reflects the original deep structure configuration in syntactic and semantic terms. 1t is the sentence which is the closest to the underlying structure representation， and the easiest to be generated from it. Provided no particular factors (mostly pragmatic ones) affect the derivational process， sentence (la) is to be produced.
Though we assume (1a) to be the closest realization of the deep structure， the specification of the deep structure representation itself varies among linguistic models. Accordingly， the explanatory power as to how sl.lch an unmarked sentence is formed differs from one model to another. To take Standard Theory， Generative Semantics， and Case Grammar as alternatives， the underlying structl.lre of the sentences in
( 1) will be represented respectively as follows:
( 2 ) Standard Theory S
‑‑‑‑‑‑‑‑‑‑ト下¥
NP AUX VP
J l
￨ / f ¥ ¥ ¥ ohn Pres V Sb ￨ / / / 丁 ¥ ¥ ¥
elieve NP AUX VP .. 1，.
̲ 1 ~\\\ーー
l'Jatahe 上'res V V P
h/\~\
ave eロ V NP/ ¥
eat the cookies
46 Transformations and Sentence Markedness (3) Gen巴rativeSemantics
S 丁一一一一
v
NP NP be1ieve. ) J
they ~え\
V ?¥TP ha
ム n 4
‑‑‑‑‑ト¥
V NP NP . ‑ 1..
eat Natalie cookies ( 4) Case Grammar 1)
~\\\ S
お1. P Pres b
lγ1
elieve JohnJ ; ; :
Me e I 1
O S
P
V
E‑‑1
1 t
the cLkies Na1talie
As we can observe， the Standard Theory model ref!ects the surface word order in its deep structure configuration， whereas Generative Semantics and Case Grammar do not， and so， the deep structure (2) is the closest to (la). This closeness， however， only indicates the shaIIowness of the deep structure. Since the deep structure of Standard Theory is com‑
paratively close to its surface structure， the formation of an unmarked sentence from its deep‑structure is trivially guaranteed. No special rule is needed to create surface subject， object， etc.， for they are directly encoded in the underlying con五guration. The subject， for instance， can be defined simply， in underlying terms， as the NP which is immediately
Transformations and Sentence Markedness 47 dominated by S." In such a model， no explanation can be expected as to how a sentence like (1a) is predicted.
Generative Semantics and Case Grammar， on the other hand， assume an underlying structure that does not reRect surface linear order， for their underlying structure， which is semantic in nature， consists of ar‑ guments (NPs) and associated predicates (Vs) which together make up propositions. Due to the nature of this deep structure， however， both models formulate a set of rules for creating a left‑to‑right linear sequence.
Rules for creating subjects are especial1y necessary， because every English sentence has to have a surface subject.
This subject creation process is cal1ed subjectivalization" in Case Grammar and subject formation" in Generative Semantics， but they are radical1y di妊'erentin nature. 1n Generative Semantics the process of subject formation is relatively automatic， because the underlying predicate of this model encodes relational indices which indicate how the underlying noun phrases are linearly ordered. The lowest S of (3)， for example， actually speci五esthe following underlying word order:
( 5)
S
~ア\\\
Proposition NP; Xj NP; X2 正三二二三』 ぷこ三三』 ぷここここ三‑
X1 eat X2 Natalie the cookies
Hence Subject Formation transformation is to merely move the leftmost NP to the left of the predicate. Simplicity of this operation leads us to think that this is not much different from assuming an underlying subject and a corresponding VP node， which in this model is supposed to be made after the subject is formed. Hence， due to the direct implication
官48 Transformations and Sentence Markedness
of the automatic Subject Formation process， Generative Semantics may be regarded as having an undetlying subject， which might be de五ned as the leftmost NP immediately dominated by S." No notion of subject choice is involved here. As a consequence， Generative Semantics as well as Standard Theory does not provide a natural account of why a certain argument is s己lected as a subject， what kind of properties characterize the status of a subject， and so forth.
With regard to this problem， the Case Grammar approach seems to provide a principled account. Since its deep structure does not involve any direct implication of an underlying subject， it necessarily attaches a deeper significance to the process of subject selection. Consequently， it serves to show what kind of argument is to be realized as a surface
subject.
In Case Grammar， the underlying relationship between NPs and V is specified in terms of the roles" the arguments play with respect to their associated predicates. These roles are. identified with (deep struc‑ ture) cases，" which are taken to be basic in Case Grammar theory. Thus， the case approach is based on the assumption that each argument in a given predicate is assigned one unique case role， which holds a special labeled relation to the predicate. The way of this relational indexing is opposite to that in Generative Semantics. Furthermore， the cases are assumed to exist in a hierarchy， which guides an 0peration of subject selection for sentences containing more than one case category. With a slight revision of the original case notions， C. Fillmore presents the following Case Hierarchy:
( 6) Case Hierarchy
Transformations and Sentence Markedness Agent > Experiencer > 1nstrument >Object >Source >
Goal> Location> Time
49
He claims that the case in a gi ven sentence which occurs first on this list determines what is to be the subject of the sentence in ... the
unmarked' instance." 3)
With these preliminary remarks on Case Grammar in mind， let us turn back to sentence (la) and its underlying configuration (4). Here we can .see why the Experiencer John" rather than the Object with a lower S node is realized as a subject， and why the Agent Natalie"
instead of the Goal the cookies" is selected as a complement subject. This selection owes simply to the subject selection order of the Case Hierarchy， and can be judged as the most unmarked " choice of subject. Hence， sentence (la)， which is created in conformity to the hierarchical order of subject choice， may be regarded as the most unmarked sentence derived from the underlying structure.
What is the Case Hierarchy， then? What does this hierarchical view of subject selection， which allows the formation of such an unmarked sentence as (la)， serve to signify? 1n my view， it serves to provide a prototypical pattern of the subject， and to suggest the ideational nature of the properties of subjects as well as sentences as a whole. Fol1owirig the list， Agency is the most prototypical property of the subjectY That is to say that human minds tend to conceptualize an Agent subject， if there is one， and a concomitant action verb. This agent‑action" pattern is perhaps most deeply rooted in man's cognitive processes.τhe most unmarked sentence may， therefore， be a sentence that is most naturally compatible with human cognition. Of course the Case Hierarchy alone
50 Transformations and Sentence Markedness
cannot handle such an intricate problem as cognition， but it at least presents a good tentative example of what it is like.
Fillmore later developes a notion of Saliency Hierarchy based on such notions as humanness， change， de五niteness，etc. that provides a mapping principle between a perspective" taken in a clause and the structure of the sentence. According to him， something like a SALIENCY HIERARCHY determines what gets foregrounded， and something like a CASE HIERARCHY determines how the foregrounded nominal are assigned grammatical functions." 5) For the present， this explanation， with Case Hierarchy and Salient Hierarchy integrated， seems to be one of the most powerful accounts of how an underlying con五guration comes to form an unmarked sentence， which is the intial stage of sentence production.
3. The Second Stage of Derivation
Up to this point we have seen， in dealing. with the first stage of derivation， that， due to the complete exclusion of the deep‑structure im‑
plication of surface grammatical relations， Case Grammar model， among the three models， provides the most insightful account of how an un‑
marked declarative sentence like (la) is formed. 1t succeeds in offer‑ ing prototypical patterns of unmarked sentences by utilizing the Case Hierarchy that indirectly controls subjectselection.
But， then， a question now arises as to how the rest of the sentences in ( 1) are created. To answer this question， we need to discuss the func‑ tional status of another set of rules， such as Passivization and Raising. In this section， we shall be concerned with the second stage of sentence derivation， in which the unmarked sentence (la) is converted into various
Transformation包andSentence Markedn色ss 51 surface forms， and we shalI particularly consider the surface variants in (1) in terms of markedness. "
Before we go on to examine a tentative markedness ordering of the sentence variants， the notion markedness" must be clarified. This is actually a vague notion used intuitively by many linguists， and its criteria have .not yet been fully specified. We shall， however， adopt the following account of markedness made by T. Givon:
In judging two variant syntactic patterns A and B as to their relati ve basicness，" if one discovers that all the components of A are also found in B but in addition another element， C， is also a component of B but not of A， and thus one considers C as the element which marks B as being distinct from A， then one is justified in considering A the basic， neutral， unmarked type and B the marked type.6J Now， if sentence (la) is taken as basic， all the other sentences are to be considered as transformationally derived. In this respect， transfor‑ mations are a set of choices associated with a giv巴nsensence， and， in my assumption， transformationally related sentences vary ir:t markedness according to the quality and quantity of transformations.
3. 1. Passivizations vs. Raising
As we can see， sentences (lb)一(1j)are a set of sentences generated by the combinations of Passivization and Raising (and additionally， Extraposition and concomitant lt‑replacement). Although Case Grammar can naturally explain an unmarked basic sentence like (la)， it does not provide a detailed account of Passivization and Raising.7> Standard Theory and Generative Semantics provide a better account in this respect. On the basis of the standard account of Passivization and Raising de‑
52 Transforniations and Sentence Markedness
veloped in Standard Theory and Generative Semantics， we can get the following list of transformations applied in order to each sentence in
(1) :
( 7) List of Applied Transformations
i
ofPh恥 加 ca邸 蹴s鎚osmi討p山 向vli同耐zz副ea低amteへ
ntl s i
C吟味:;>KalsmgRa凶 吟叶c:;>i I
Pa.
oおfs路 main 叩s討的i SI
c:;>cxrrapos+
十
十 +
+
+ +
+ 十
+ + +
十 十
+ 十
+
:transformation appliedAlthough we put the total number of transformations for convenience， the mere numeration of transformations applied does not lead to the identi五cationof markedness of the sentences in question. We should also take into account the qualitative status of transformations.
Let us五rstcompare Passivization of a complement S and Raising as are manifested in (la) 一 (ld)~ As we can observe， all the sentences in (la)一(1d) have the same surface subject John" and the predicate believe，" the difference being whether Passivization or Raising is ap‑ plied to the basic sentence (la). There is no doubt that (ld) is the
TransforrIl丘tionsand Sentence Markedness 53 most marked sentence of the four， for it is formed both with Passi vization and Raising. But the markedness order between (lb) and (lc) is hard to determine， because only one of the transformations is applied to each of them.
In my view， (lb) is more marked than (lc) ; that is to say， Passiviza‑ tion is a more marking transformation than Raising. Several reasons can be given in support of this assumption. First， we must note that Passivization is not a single transformation， but is actually comprised of two distInct operations， namely， Agent Postposing and NP Preposing.8l In other words， Passivization involves twoprocesses， .ie.， 1) the demotion of the Agent from the subject position， and 2) the promotion of a non~
agent to subject status. Raising， on the otherhand， involves only one process， i.e.， the promotion of a complement subject to the direct object position of a higher clause. Therefore， if we count Passivization as two different transformations， the mere enumeration of the total number of applied transformations can tell that Passivization is more marking than Raising.
Secondly， from the viewpoint of Case Hierarchy， a derived subject created by Passivization violates the prototypical pattern of subject selection
，
for Passivization is a role咽preservingoperation，
while Raising is not. Even though the patient object (in Fillmore's terms， the NP with the Object case or the Goal case) is moved into a subject position， this derived subject cannot obtain an Agent property， which is， as was noted earlier， the most prototypical ideational pattern of the subject. Hence a derived subject in a passive sentence lacks its prototypical semantic property， i.e.， Agency.Using a multi‑factor concept of subject， E. Keenan suggests a hy‑
54 Transformations and Sentence Markedn邑S8
pothesis concerning anordering of subject properties，的 which may be schematically summarized as in (8):
( 8) The Promotion to Subject Hierarchy (PSH)
Subjec thood
A. Semantic Properties: Agency， autonomous existence， selectional restrictions， etc.
令
B. Behavior and Control Properties: deletion， movement， case changing properties， control of cross‑reference properties， etc.
合
C. Coding Properties: position c:;> case agree‑ ment今verbagreement
In this hierarchy the Coding Properties of the subject (C‑properties) are the easiest to acquire， and the Semantic Properties (A‑properties) the hardest. Therefore， if an NP in a derived sentence is assigned any of the three categories of subject properties， then it is assignedall the lower categories， and the higher an NP is placed in the hierarchy， the more subject like it is. Returning to our example again， we see that the derived passive subject "the cookies" in (lb) has acquired up to the B‑properties but not the A‑properties.
As for Raising， the result will be reverse. To take (lc)， for example， the raised NP Natalie " has obtained a semantic role of the Object case， for it was put into a case frame，" or in other words a perspective，" 10)
of the main verb believe" losing instead its original C‑properties as a complement subject. But even if the raised NP loses the C‑properties， which are in fact the minimum requisite condition to be qualified as a subject， it still retains the B‑properties. In a sentence like John belives Natalie to like herselfj匂lImself，"for instance， it is the raised NP Nata1ie，" not the subject NP John，" that has control of coreference.
Transformations and Sentence Markedness 55 Another B‑property， deletion， is also observable in a sentence like I expect him to come and talk." The supposedly deleted subject of talk"
is not I" but him‑"
Moreover， in addition to B‑properties， raised NPs generally possess A‑properties as wel. 1Selectional restrictions， for example， still hold between a raised NP and its associated to‑infinitive; we cannot say，
*John believes the basket to have eatenthe cookies，" since the verb eat" requires an NP with [十animateJfeature as its subject. Actually， in sentence (lc)， the raised NP Nata!ie" has not lost its original semantic role of the Agent， though it has gained a new semantic role as Object. Natalie" is in this sense placed at the intersection of the case frame for "believe" and that for eat，" thus assuming a double case role， Object and Agent.
These observations lead us to suppose that Passivization is a transfor‑ mation that creates a subject with C‑and B‑properties of the PSH， while Raising 11) is an operation which deprives a complement subjec.t of its original legitimate status of subject. Raising eliminates the C‑properties， but retains the B‑and A司properties. Hence， comparing (lb) and (lc) again with the basic sentence (1a)， we may assume that the object NP
Nata!ie" in (lc) is much closer to its original function as a comple‑
ment subject than a derived complement subject "the cookies" in (lb). Presumably， the prototypical clause pattern of Agent噂Action過Goal，acti‑ vated by the verb eat，" is so firmly fixed in our mind that the violation of this pattern (i.e勺 the violation of the case‑hierarchical ideational pattern)， by means of Passivization， would produce more markedness to our' cognition than asuperficial loss of a complement subject caused by Raising.
56 Transformations and Sentence Markedness
The third argument to attest the markedness" Qf Passivization is related to the functional aspectsof word order. Passivizatron is a clause‑internal operation that induces the order change of NPs， while Raising is an inter"clausal operation whichdoes not a妊ect the. order change of NPs. According to my presentassumption， the left‑to‑right linear order of NPs is functionally more important in English than the preservation of rigid clause boundaries.
The concept of linearity has expressly been elaborated in the theory of Functional Sentence Perspecti ve， developed by Prague School linguists. Especially， their notion of Communicative Dynamism (CD) is useful in explicating the contribution of each sentence element to the development of the communication. The following is its explanation:
The starting point of the theory is the assumption that it is in accordance both with the character of human thought and with the linear character of the sentence that sentence elements follow each otheraccording to the amount (degree) of communicative dynamism (=CD) they convey， starting with the lowest and gradually passing on to the highest.
By the degree of CD carried by a sentence element we understand the extent towhich the sentence element contributes to the devel‑ opment of the communication， to which， as it were， it pushes' the communication forward.12)
According to this definition， the degree of CD increases from the leftmost elementto the rightmost element of the sentence. What is meant exactly by CD is as yet to be clarified， bu.t the notion of CD is perhaps most closely related to the distinction of "given‑new" information， because new information necessarily contributes more to communication than given information. Other functional factors， such as Topic and Fo‑
Transformations and Sentence Markedness 57 cus，l3J are also relevant. The general tendency is that a Topic con‑ stituent is to be identi五ed as given，" carrying the lowest degree of CD， and a Focus constituent as new，" carrying the highest degree of CD. This situation can be diagrammatically presented as fol‑ lows:
( 9) Functional Factors Associated with Word Order clause
‑ ー
right!eft
Then， we can imagine that Passivization， involving the clause句internal interchange of two NPs， changes all these functional factors which are associated with them in an unmarked instance. In sentence (lb)， for example， the cookies，" which initially had the highest degree of CD and Focus， is moved into the slot of Topic. This is really a powerful and marking transformation， with allthe ordering changes of functional factors associated with the moved NPs.
How about Raising? Though this is an inter司clausal transformation， the distribution of CD does not seem to be affected by this operation. Even though CD is basically a clause‑internal notion， a two‑clause sen‑ tence like (la) has the same distributional pattern of CD as a single clause does. Therefore， Raising may lessen the CD degree of the raised NP itself， but it does not induce any order change. of CD among the
58 Transformations and Sentence Markedness
NPs. If we put a tentative number to each position of the sentence to indicate the degree of CD， we shall get the following:
ーi
QU n
‑ ‑
a m
r+ A
o mwD
C C
2 3
CD of complement S 1 2
(la) John believes that Natalie has eaten the cookies total CD ① ② ③ 也
(1b) John believes thatω
ふ
KZEhebeen制 eλ子 育
atalie.① ① ① / ' ④ @
@
(lc) John believes Nautルlie to have eaten the cookies.
① ② ③ ④ ⑤ ①
The raised NP Natiilie" may lose a part of its original CD， but gains instead a higher degree of topicality by being integrated into a nuclear part of a higher clause as its direct object.14) But it does not affect the total distributional order of the functional factors at any rate. Hence Passivization， which causes the permutation of NPs in a single clause， should create greater markedness than Raising， which amalgamates two clauses into one without changing word order.
With aIl the evidence to show the markingness of Passivization in comparison with Raising， we can now posit the following order of sen司 tence markedness among (la)一(ld): (la)く(lc)く(lb)く(ld).
3. 2. Passivizatipn vs. Extraposition
So far， we have been concerned with the sentences (la)一(ld)assuming that Passivization is more marking than Raising. In the sentences in (1)， however， Passivization is applied to both complement and main
Trarisfcirmations and Sentence Markedness 59 clause. If so， which operation would normally yield greater matkedness
?
Let us now compare the passivization of a main clause (A‑Passive) and that of a complement clause (B‑Passive).
From the list of transformations in (7)， it is clear that (lb) is formed with B‑Passive alone， and (le) with A喝Passivealone. 1ntuitively， how‑
ever， (le) should be more marked. This judgement is very natural because a movement operation in a higher cycle will necessarily involve a movement of all the constituents in a lower cycle. While B‑Passive only involves word喝orderdestruction of a complement， A‑Passi ve leads to the destruction of the original word order of the entire sentence. Hence the latter operation is against the important conception of linearity in English， which was mentioned earlier.
If we apply Extraposition (and concomitant It‑replacement) to (le)， the resulting sentence (lg) seems less marked than (le). Counting the applied transformations， therefore， does not correspond to sentence markedness here. From the viewpoint of word司order change， (lg) is apparently less marked than (le)， for it only moved John" from the left of the verb believe" to the right of it. Due to E宜traposition， sentence (lg) maintains a complement that‑clause in its original position even though the main clause is passivized. Because of the similarity of this word order and many other functional factors associated with it， (lg) is cognitively closer to the basic sentence (la) than (le) is. Hence the order of transformations (i.e.， A‑Passive中Extraposition)runs counter to our cogmtion.
1t might rather seem more reasonable in this case to suppose that A‑Passive is a local passivization to convert a sentence like (la) into (lg) without affecting the complement that‑clause. We might further
60 Transformations and Sentence. Markedn巴ss
suppose that Intraposition would apply to produce such a sentence as (le)， instead of assuming A‑Passive to be a global transformation to generate (le) from (la)， and that Extraposition to be applied to produce (lg). But， ifwe prefer the derivational order(la)~(lg) 中 (le) ， we should be able to account for where the expletive it in (lg) comes from. This it cannot be explained in terms of a reduced form of the fact， which Kiparsky and Kiparsky assumed to exist in a deep structure of factivecomplements，m for the verb believe " is not classified as a fac‑ tive predicate. Since we can divide (la) into two sentences (i.e.，Natalie has eaten the cookies" and John believes it勺 it is evident that a that‑clause can be pronominalized into it. But， unless we could assume it to be already existent in the deep structure without the application of such an operation as pronominalization， we would fail to capture the closeness of (la) and (lg)， because the application of more transforma‑ tions is toinvolve their deviation from the original unmarked sentence. Hence， until the expletive it can be adequately explained， the hypothesis of local Passivization and Intraposition will not hold. Then， in the case of (le) and (lg)， we must admit for the present that enumeration of the transformations contradicts the markedness order of the sentences. Now we must resort to other criteria to explain the fact that (le) is more marked than (lg). In addition to the functional principle of word order associated with CD， Topicality， Focus， etc.， what is called the principle of end偽weight"helps to account for the markedness difference between (le) and (lg). The point of this principle is to capture the tendency to reserve the .final position for the more complex parts of a clause or sentence." 16 In general， English has some restrictions on the surface structure ordering of consituents such that the more complex
Transformations and Sent巴nc巴Markedness 61 constituents appear toward the end of the sentence and the less complex constituents appear toward the beginning. Sentences (le) and (lf)， which have a thαt‑clause at the beginning， are apparently against this principle， whereas their extraposed counterparts (lg) and (lh) are not. Thus， we may presume (lg) and (1h) to be less marked than their respective counterparts (le) and (lf)， despite the fact that the latter have one more additional transformation， i.e.， Extraposition.
Furthermore， the global A‑Passive which moves a that‑clause to the sentence‑initial position would also affect the quality of the clause as regards factivity. Usually， the verb believe" takes a non‑factive com‑
plement; hence a proposition expressed in the complement clause in (1a) is asserted but not presupposed to be true by the speaker. However， when the complement is moved to the beginning of the sentence by Passivization， it will be interpreted as factive， that is to say， the moved complement will come to imply the speaker's presupposition of its truth value. Observing this phenomenon， Kiparsky and Kiparsky suggest a general tendency of sentence‑initial clauses to receive factive interpre‑ tation regardless uf types of transformations:
The interpretation can be nonfactive if both passive and extraposi‑ tion have applied， or if neither of them has applied; if only the passive has applied， we get the factive interpretation. This is very hard to state in terms of a condition on transformations. It is much easier to say that the initial position itself of a clause is in such cases associated witha factive sense.17)
Therefore， the closeness of (1g) and (la) as opposed to (le) and (1a) is again attested in terms of this criterion concerning factivity. In fact， the usage of a that・clauseat a sentence‑initial position seems to be limited
62 Transformations and Sentence Markedness
only to special cases in which the speaker wants to make an emphatic assertion fo士therest of the sentence. For example， whereas the content of the that‑clause is asserted in (la)， it is presupposed in (le) and (lf)， and instead a passivized predicate of the main clause . . . is believed by John" is asserted， by being giv巴na higher degree of CD and Focus. In other cases， however， (le) and (lf) should be avoided by reason that they are against the principle of end‑weight.
Hence the tentati ve order of markedness among (1e)一(lh)，when com‑
pared with the unmarked sentence (la)， is as follows: (la)く(lg)く (lh)く(le)く(lf).
4. Markedness Principles and Tentative Ordering of Transformか
tional Variants
Based on what has been discussed so far， we propose some principles of sentence markedness， which can be summarized as in (11):
(11) Princ伊les01 Sentence Markedness 1. Quantity of transformations
Total number of application 2. Quality of transformations
1) Ideational aspects (e妊'ectof transformations)
Degree of Case‑hierarchy violation related to human cognition of prototypes
2) Functional aspects (range of transformations)
Degree of word‑order destruction related to CD， Top‑
icality， Focus， etc. 3. Surface‑structure restrictions
End‑weight， etc.
Transformations and Sentenc己Markedness 63 Up to this point we have been concerned only with a partial ordering of the sentences in (1)， but， by taking these principles as criteria， we can now proceed to consider the total markedness order of all the sen‑ tences. In order to evaluate each principle relatively and to make a synthetic assessment of the interplay of the principles in each of the sentences， we shall tentatively use numerals to indicate the various factors of transformations. But the numerals themsel ves are not to be taken seriously. They merely serve as a device to imply the relative degree of markedness order among the sentences. It must be especially noted that the total suIh of numerals for each sentence is only to be used for estimating the relative markedness order， and is not to indicate the exact degree of markedness.
'Vith these remarks in mind， let us see the results of comparison:
(12) Results of Comparison
￨Number oflmect transformations I transfoofIRtraannsgfe rmations I transfo orof￨To‑‑i mations
Unmarked O O O o
1 O 1
2 1 2
2 2 4
3 1 5 9
3 2 5
4 3 6
5 3 9
2 1 16
Marked 4 3 16
64 Transformations and Sentenc巴M紅 kedness
N. B. 1) The Number of Transformations: Passivization is counted as two distinct operations， and Raising as one. Extraposition is not coun‑ ted for the reasons' mentioned in 3.2.
2) The Effect of Transformations: Case‑hierachy violation of the main clause (by means of Passivization) is counted as one， and that of the complement clause， two， because the demotion of the Agent complement subject Nata!ie" must be cognitively more marked than the demotion of the Experiencer John."
3) The Range of Transformations: we have counted the movement of each element of the sentence in terms of the change of the degree of CD indicated tentatively by numeral ，sas was manifested in (10). Sentence (lc)， for instance， is counted as four， because Natalie，" which was originally in the ④position， is moved to①， and the cookies" is moved from ③ to④.
Even though the exact number of each figure is not quite well‑grounded， the way of calculation proposed in (12) at least serves to show how the markedness of transformationally related sentences can be estimated. Actually， sentence markedness is a complex notion that needs to be evaluated within a wider scope of consideration， and the ultimate justiι cation of markedness order will probably come from the五eldof cognitive psychology. However， by having recourse to the principles of marked‑
ness advocated above， we may as well make a rough calculation of markedness order even within the五eldof linguistics proper.
5. Concluding Remarks
This paper has centered on the discussion of a single set of s邑ntences deri ved from th己 sameunderlying structure. We have shown that the formation of the unmarked sentence can be best explained in accordance with Case Hierarchy， and that the rest of the sentences vary in marked‑
ness according to the three principles we have suggested. Although we have demonstrated th己 intimate relationships between transformations
Transformations and Sentence Markedness 65 and sentence markedness， we could not go Into the question of why a certain transformation is applied to make a sentence more marked. It seems that even predication‑internal transformations like PassivIzation and Raisi:ng are controlled by pragmatic factors. Passivization， for ex‑ ample， may be used to make the patient noun more topical， but the question still remains as to the difference between purely pragmatic topicalizing transformations (e.g.， Topicalization， Left Dislocation， Right Dislocation) and Passivization. The resolution of these issues must await further research in the future.
NOTES
1 This reresentation is based on Charles Fillmore's following two papers: C. Fi1lmore，The Case for Case，" Universals 仇 LinguisticTheory， eds. E. Bach and R. Harms (New Y ork: Holt， Rinehart and i司Tinston，1968)， and C. Fi1lmore， Some Problems for Case Grammr，" Monograph Series on Language and Linguistics， XXIV， R. O'Brien (Washigton D. C.: George‑ town University Press， 1971).
2 This notation follows James McCawley，Where Do Noun Phrases Come From ?，" Semantics， eds. D. Steinberg and L. Jakobovits (London: Cambridge
University Press， 1971)， p. 224.
3 C. Fillmore，Some Problems for Case Grammar，" p. 42.
4 G. Lakoff further analyzes the Agent properties into VOLITION， CONTROL， and PRIMARY RESPONSIBILITY， and suggests that PRIMARY RESPON‑
SIBILITY is the most central Agent property that pairs with subjecthood. See George Lakoff， Linguistic Gestalts，" Papers From the Thirteenth Regional Meeting Chicαgo Linguistic Society， eds. W. Beach et al. (Chicago: Chicago Linguistic Society， 1977)， pp. 250 ff.
5 C. Fillmore，The Case for Case Reopened，" Syntax and Semantics， Vol. VIII， eds. P. Cole and J. Sadock (New York: Academic Press， 1977)， p.80. As for the details of the Saliency Hierarchy， see also C. Fillmore，Topics in Lexical Semantics，" Current Issues in Linguistic Theory， ed. R. W. Cole
66 Transformations and Sentence Markedness (Bloomington: Indiana University Press， 1977)， pp. 102 ff.
6 Talmy Givon， On Understandi・πgGrammar (New Y ork: Academic Press， 1979)， p. 47.
7 Though Passivization is explained as the marked choice of subject that requires transformations to re‑rank the element in the Case Hierarchy， the transformational status of Raising is not discussed at al1 in Case Grammar theory.
8 For the detailed discussions of Passive as a two‑fold transformation， see Peter Culicover， Syntax (New York: Academic Press， 1976)， pp. 160‑172. 9 Edward L. Keenan， ，Towards a Universal De五nitionof subjectγ， Subject
and Topic， ed. C. N. Li (New Y ork: Academic Press， 1976)， p. 324. But my interpretation of the promotion direction of subject properties， which v可，il!be diagramed in (8)， is opposite to his.
10 A case frame，" in Case Grammar terms， is an array of cases which serves to indicate the cas巴notionsconceptual1y present in a sentence. A perspecti ve， "
which is assumed to be assigned by a choice of a particular verb related to an event in question， means a certain way of looking at a cognitive scene activated by an utterarnce. See C. Fillmore， The Case for Case Reopened."
11 What I mean by Raising here is narrowly limited to what Postal cal1s B・verb Raising (Subject‑Object Raising). A‑verb Raising (Subject‑Subject Raising) wil1 not be discussed in this paper. See Paul Postal， On Raising (Cambridge， Mass.: The MIT Press， 1974) for A‑verb Raising and B刊 rb Raising.
12 Jan Firbas， Non‑thematic Subjects in Contemporary English." Travaux linguistiques de Prague， Vol. II， eds. F. Danes et al. (Paris: Academia， 1966)， p. 240.
13 For the definition of Topic and Focus， we shal1 take the position suggested by Simon Dik， Fuπctional Grammar (Dordrecht: Foris Publications， 1978)， p. 130.
14 It has been noted， in fact， that a direct object sometimes assumes a function similar to that of a subject， especial1y in such examples as fol1ows:
( 1) a. Bees were swarming in the garden. b. The garden was swarming with bees.
Transformations and Sent巴nceMarkedness 67 ( 2 ) a. John sprayed paint on the wal l.
b. John sprayed the wal1 with paint.
Just as the subject of (la) the garden" has been suggested to recei ve a HOLISTIC interpretation (i.e.， the interpretation of the whole garden being affeted) whereas the garden" in (la) is PARTITIVE， the direct object of (2b) the wal1" is also interpreted as holistic while the wall" in (2b)， as partitive. The source of this holistic interpretation may be ascribed to the topicality of the NPs. The subject slot is often出ledby an item about which knowledge is being added， but next to the subject， the direct object is likely to assume this function.
15 The postulation of a head noun fact in the deep structure of a factive clause is advocated in Paul Kiparsky and Carol Kiparsky，Fact，" Readings in the Theory of Gran口冗ar，ed. D. Bornstein (Cambridge， Mass.: 百九Tinthrop Publishers， 1976)， pp. 299‑330.
16 Randolph Quirk et al.， A Grammar of Contemporary English (Essex:
Longman， 1972)， p. 943.
17 P. Kiparsky and C. Kiparsky， op. cit.， p. 326.
