13 This suggestion is less of a deus ex machina than it sounds from my brief ..... PRT fall. I understand that my farm is falling (down). These constructions can ...
STEVE
THE
SYNTAX
HARLOW
OF WELSH
SOFT MUTATION*
The Welsh morphophonological phenomenon known as SOFT MUTATION is one which attracts recurrent attention from linguists, b o t h for its phonological properties and for its relationship with morphosyntax. In this paper I review two recent discussions by Lieber and Zwicky which attempt to account for a particular subclass of soft mutation by an appeal to case. I argue that neither successfully accounts for the full range of relevant data and suggest a novel interpretation of the phenomenon which generalises beyond the occurrence of soft mutation in NPs to its occurrence in other categories. Instead of an account founded on a treatment of soft mutation as a manifestation of case, I propose that it can be satisfactorily and elegantly accounted for if we recognise phrasal categories as triggers of phonological phenomena. 1. T H E
PROBLEM
The topic of this paper is a reconsideration of the phenomenon, commonly termed DIRECT OBJECT MUTATION (DOM henceforth), exemplified in (1). (1)
Gwelodd y dyn gi. s a w - P A S T - 3 S the man dog T h e man saw a dog.
The crucial item in (1) is the word gi 'dog', which appears in its soft mutated form. SOFT MUTATION is one of a set of morphophonological changes which affect the initial consonants of Welsh morphemes. 1 Typically, the appearance of a soft-mutated form depends upon the * Earlier versions of this paper were presented at the University of Manchester, the Linguistics Association of Great Britain, Ohio State University and the Kentucky Foreign Language Conference. I am grateful to those audiences for discussion and comments. I am also indebted to Bob Borsley, Connie Cullen, James Fife, John Phillips, Greg Stump and two anonymous reviewers for helpful comments and suggestions. Ultimate responsibility for the results is of course mine. i This process results in the initial consonant of the citation form of a word undergoing the (orthographic) replacements listed in (i). (i)
p---~ b t---~d c~ g
b---~ f d---~ dd g --~ NULL m---~ f rh~r 11---~1
Since I shall have nothing o f substance to say in this paper about how an analysis of soft
Natural Language and Linguistic Theory 7: 289-316, 1989. (~) 1989 Kluwer Academic Publishers. Printed in the Netherlands.
290
STEVE
HARLOW
appearance in the environment of the affected consonant of some triggering item. As a straightforward example, consider (2). (2)a.
Caerdydd
Cardiff b.
i
Gaerdydd
to Cardiff In (2a) we have the word for 'Cardiff' in its unmutated form, with an initial voiceless velar stop. In (2b), where it follows the preposition i 'to', we require the soft-mutated form Gaerdydd with a voiced initial consonant. Examples of this kind are standardly interpreted as the consequehce of the presence of the preposition - i is a soft-mutation trigger. The question that I shall be concerned with can now be specified more precisely: what is the trigger for the soft mutation of gi that we observe in examples like (1)? It is, of course, possible that this particular case of soft mutation is not triggered by any item in the environment, but is rather an inherent property of direct objects. 2 Such an account is, however, hard to sustain in the light of examples like (3). mutation should b e implemented in a theory of phonology, the phonetic realisations of the orthography will not be of any direct relevance. Briefly, p, t, ¢ are voiceless stops, b, d, g are voiced stops, dd is a voiced dental fricative, m is a bilabial nasal, f is a voiced labiodental fricative, tit is a voiceless aspirated alveolar trill, II is a voiceless lateral fricative, r is a voiced alveolar tap and ! is a voiced lateral. For some proposals as to how these, and other, alternations should be accounted for in a phonological analysis and for further details of the Welsh mutation system, see, for example, Sproat (1986). 2 Note that there do in fact seem to be cases of initial consonant mutation in Welsh which are not triggered by some morpheme in the environment. One such example is the soft mutation of feminine adjectives in modifier position. Compare (i) with (ii). (i)
merch dda
girl
good
a good g i r l (ii)
bachgen da
boy
good
a good boy In (i) the adjective, modifying a grammatically feminine noun, appears in the soft-mutated form dda, in contrast to the unmutated masculine form da in (ii). A plausible way of interpreting this contrast would be to state that feminine modifier adjectives are inherently soft-mutated. I am not, therefore, committed to the position that there is a unitary account of the triggering environments for soft mutation. I shall claim that whatever is responsible for DOM in (1) is a sufficient condition for mutation, but not a necessary one. This is worth stating since at least one writer, Fife (1986, p. 181), has mistakenly assumed the contrary. See also Willis (1982) for comprehensive discussion of the complexities of Welsh mutation.
THE
(3)
SYNTAX
OF WELSH
SOFT
MUTATION
291
Mae'r dyn yn gweld ci. is the man P R T see dog The man sees the dog.
In this example the word for 'dog' is still the direct object of a verb (in this case gweld), but here we have the unmutated form ci. The standard interpretation (e.g. Awbery 1976, p. 8) of this contrast between (1) and (3) is that in (1) gi is the direct object of a finite verb gwelodd, whereas in (3) ci is the direct object of a nonfinite verb gweld. 3 The trigger for the soft mutation we observe in (1) is therefore the finite verb. As has been pointed out in papers by Lieber (1983) and Zwicky (1984), this interpretation, if correct, is an embarrassment for attempts to establish restrictive theories of (morpho)phonology. The reason is that, under the interpretation just outlined, since Welsh is a VSO language, the trigger gwelodd and its victim gi are not linearly adjacent. If this statement of the facts of direct object mutation is correct, it will require that phonological theories must permit action at a distance, and, in the worst case, might admit conditioning statements which alli~w a trigger for some phonological process to occur arbitrarily far away from t]he affected segment(s). Zwicky has therefore suggested that (morpho)phonological theories should embody the following constraint.4 (4)
THE TRIGGER CONSTRAINT The trigger determining a rule feature for a morphophonemic rule must be adjacent to the affected word and c-command it. (Zwicky 1984, p. 389)
We have seen that Welsh appears to instantiate a counterexample to the" trigger constraint in that it allows a finite verb to trigger soft mutation on a constituent to which it is nonadjacent. Indeed, additional data suggest that the conditions o n this case of soft mutation actually require a stipulation that the trigger and its victim be nonadjacent. Compare (5a) and (5b). (5)a.
Mae ci mawr yn yr ardd. is dog big in the garden There is a big dog in the garden.
3 Traditionally termed a 'verbal noun'. See note 16 below for further c o m m e n t . 4 Note that (4), as it stands, is a stipulation. It should be preferable for the conditions embodied in the constraint to arise as the c o n s e q u e n c e of a theory of morphophonological rules. I will return to this point briefly in the sequel.
292
(5)b.
STEVE
HARLOW
Mae yn yr ardd gi mawr. is in the garden dog big There is a big dog in the garden.
In (5a) the subject NP ci mawr occurs in immediately postverbal position and, as expected, the initial consonant of ei is unmutated. In (5b) on the other hand we find an instance of inversion, with the subject NP following the PP yn yr ardd. In this case, the initial consonant of the subject NP is required to undergo soft mutation, giving gi. It appears from these examples that the conditions for DOM have to be stated as follows: the initial consonant of an NP undergoes soft mutation if it follows and is not adjacent to a finite verb. 5 Faced with this prima facie violation of the Trigger Constraint, Lieber (1983) and Zwicky (1984) have proposed alternative analyses of this phenomenon, in which they try to show that it is only an apparent counterexample to the adjacency condition and that the restrictive hypothesis encoded in the Trigger Constraint can in fact be sustained. My aim is to show that both Lieber's and Zwicky's proposals encounter severe difficulties, but that there is in fact an alternative construal of the facts of DOM which conforms to the Trigger Constraint. In essence, what I shall propose is that the standard interpretation of DOM, namely the assumption that the triggering element in examples like (1) is the finite verb, is incorrect. 6 I will proceed to this conclusion by first reviewing the proposals made by Lieber and Zwicky. This will serve to draw attention to the difficulties which their analyses encounter and also to broaden the range of data under consideration.
2. A L T E R N A T I V E
ANALYSES
2.1. Lieber 1983 Lieber is concerned with developing an autosegmental account of consonant mutation. 7 She proposes that the facts of mutation can be acs As the examples in (5) show, 'direct object mutation' is in fact a misnomer, since it is not restricted to direct objects. For simplicity's sake, however, I will retain the abbreviation D O M in this paper. 6 My proposal will be, in effect, an elaboration and defence of a suggestion first put forward in Harlow (1981, 238 f.). 7 Note that I am not concerned in this paper with the issue of the adequacy or lack of it of an autosegmentai analysis. I remain agnostic about the correct phonological analysis of soft mutation. For discussion of phonological aspects of Lieber's proposals see Sproat (1986).
T H E S Y N T A X OF W E L S H S O F T M U T A T I O N
293
counted for if we assume that the phonological representation of mutation triggers includes a floating autosegment as a peripheral item (in the case of Welsh, the rightmost). She further requires that this autosegment is linked to the right whenever possible. Thus, example (2b) would work as in (6): (6)
i. [+voice] c ae r d y dd 8
/
V
CV
.........
Note that Lieber's analysis entails at least the adiacency component of the Trigger Constraint. As Lieber points out: Given the mechanics of the autosegmental analysis, autosegmental theory predicts that consonant mutation must be strictly local. If consonant mutation is triggered at all, it must be triggered at the periphery of a morpheme by a contiguous morpheme. Triggers cannot be separated from the initial or final consonant whose mutation they condition - if they were, a violation of the usual autosegmental convention ruling out the crossing of association lines would result. (Lieber 1983, p. 169)
Lieber claims DOM in Welsh is only an apparent counterexample to the adjacency hypothesis. She points out that under the assumptions of Government/Binding theory NPs with phonetic content must be assigned Case. Accepting Stowell's (1981) arguments that Case is assigned under adjacency, she notes that the verb of a Welsh VSO sentence will fail to assign Case to its direct object, since the two positions are not adjacent. Lieber claims that the direct object position in a Welsh VSO sentence will correctly be assigned case "if we assume that Welsh has a rule of preposition insertion, something like the rule of of-insertion that has been proposed for English" (Lieber 1983, p. 173). She therefore proposes that Welsh has a syntactic rule which inserts an empty preposition before an NP which is not adjacent to its governing verb. The syntactic motivation for this preposition is its role as a Case assigner. However, Lieber capitalises on its presence to provide an account of DOM, by proposing that its phonological representation is not entirely empty, but consists of a floating autosegment comprised of the features required for soft mutation. Example (1) thus has, according to Lieber, the structure (7).
8 See Lieber (1983, pp. 170 and 175) for discussion of the phonological contents of this autosegment.
294
STEVE
(7)
HARLOW
S
V
NP
PP
gwelodd
y dyn
P
NP
[+v!ice]
ont]
"
C
V
Lieber observes that this proposal derives additional support from the fact that many 'real' prepositions in Welsh do in fact trigger soft mutation on their objects (such as, for example, the preposition i in (2b)). Note that the lack of soft mutation in (3) also follows from Lieber's analysis, because in such sentences the nonfinite verb gweld is adjacent to its object NP and can thus assign Case to it directly. Since nonfinite verbs do not carry a floating autosegment, there is consequently no mutation effect. Following this brief review of Lieber's proposal, I turn now to some difficulties which it encounters when we come to consider its interaction with Wh-movement. In Wh-constructions, Welsh, like English, allows both pied-piping, as in (8), and preposition stranding, as in (9). (8)
I bwy
(y)
rhoddodd y
to whom P R T gave
dyn yr anrheg t?
the man the present
To whom did the man give the present? (9)
Pwy (y)
rhoddodd y
who P R T gave
dyn yr anrheg iddo
t?
the man the present to-3MS
Who did the man give the present to? Recall that the preposition i 'to' is a soft mutation trigger, and note that when the question-word pwy 'who' immediately follows i, as in (8),
THE SYNTAX OF WELSH SOFT MUTATION
295
its soft mutation form, bwy, appears (as expected). Note also that when the preposition is stranded, as in (9), the question-word takes its unmutated form pwy. Consider now the implications of these constructions for Lieber's analysis. Pied-piping is optional for all overt prepositions in Welsh and the null hypothesis in the light this must be that pied-piping is optional for all prepositions, including empty ones. Consequently, we would expect that a pied-piped empty preposition will continue to trigger soft mutation on the NP which immediately follows it. We therefore predict the following possibilities corresponding to (1) (where 'P' represents Lieber's EMPTY PREPOSITION): 9 (10)
Pwy (PRT) gwelodd y who
saw
dyn P t?
the m a n
Who did the man see? in which the preposition is stranded, and (11)
P bwy (PRT) gwelodd y who
saw
dyn t?
the m a n
Who did the man see? in which the preposition has been pied-piped and has triggered soft mutation on bwy. 10 Unfortunately, this prediction is wrong. The only possibility is a sentence in which the question-word is in its unmutated form. To achieve the correct result we would have to impose some additional condition to block pied-piping of phonetically empty prepositions. Even such a stipulation would still leave a further problem for Lieber's analysis. As is well known, Welsh shows an alternation between two different preverbal particles in sentences involving Wh-movement (dis9 ]~suppress the phonological realisation of the preverbal particle in these examples, giving it simply as PRT for reasons which will become apparent shortly. 10 There is a further problem for Lieber's proposal, drawn to my attention by Richard Sproat. As can be seen from examples (8) and (9), Welsh has a distinction between noninflected and inflected forms of PrePositions, where the inflection is a manifestation of agreement. (See Harlow (1981) for details.) The relevance of this distinction to current concerns is that, in cases of preposition stranding such as (10), only the inflected form of the preposition is possible. For those prepositions which lack an inflected form, such as a/ag 'with', preposition stranding is simply impossible. The requirement for preposition stranding seems to be that the preposition has a phonologically expressed agreement form. The expectation in the case of an empty preposition is that no agreement is possible and that therefore only pied-piping is allowed, with the result that only the incorrect (11) is predicted by Lieber's analysis.
296
STEVE HARLOW
c u s s e d in s o m e d e t a i l in H a r l o w 1981). T h e d e t a i l s of this p h e n o m e n o n a r e c o m p l e x , b u t , for c u r r e n t p u r p o s e s , t h e s i g n i f i c a n t f a c t is t h a t t h e e x t r a c t i o n of a n N P o u t of a P P (the p r e p o s i t i o n - s t r a n d i n g case) also n o r m a l l y r e q u i r e s t h e o c c u r r e n c e of t h e p a r t i c l e y, as (9) illustrates. T h i s m e a n s t h a t , e v e n if a r e s t r i c t i o n c o u l d b e m o t i v a t e d to e x c l u d e p i e d p i p i n g , it w o u l d still n o t r e s o l v e t h e difficulties w h i c h L i e b e r ' s a n a l y s i s e n c o u n t e r s , b e c a u s e we w o u l d e x p e c t t h e p a r t i c l e ( P R T ) i n (10) also to b e y, as in (12). (12)
P w y y g w e l o d d y d y n P t?
A g a i n , this p r e d i c t i o n is i n c o r r e c t ; t h e o n l y g r a m m a t i c a l c o u n t e r p a r t to (1) is (13), w i t h t h e p a r t i c l e a. al (13)
P w y a w e l o d d y d y n ? lz
2.2. Z w i c k y 1 9 8 4 Z w i c k y p r o p o s e s a r e i n t e r p r e t a t i o n of D O M
i n t e r m s of a d i s t i n c t i o n
b e t w e e n a c c u s a t i v e a n d g e n i t i v e case. H i s c l a i m is t h a t a c c u s a t i v e c a s e is r e a l i s e d i n W e l s h b y soft m u t a t i o n , w h i l e g e n i t i v e c a s e is r e a l i s e d s i m p l y
11 It is possible in written varieties of Welsh for a also to occur when there is 'real' preposition-stranding, as illustrated in (i). (i)
Pwy (a) roddodd y dyn yr anrheg iddo t? who P R T gave the man the present to-3MS Who did the man give the present to?
But this does not save Lieber's analysis, because in these examples the y alternative is always available. In the empty preposition cases under discussion, however, a is the only possibility. One way out of this dilemma would be to propose that the empty preposition and the following NP do not form a constituent. This is obviously not an empirically viable conjecture, and has the disadvantage that it introduces another ad hoc element: it would be the only preposition in Welsh which fails to form a constituent with its object, and the support which Lieber's proposal derives from the parallelism with real prepositions would be lost. Bob Borsley and an anonymous reviewer have both suggested that these ditticulties would be circumvented under an analysis in which there is no movement of Wh-phrases, but where instead these are base-generated in their surface positions and movement is of an empty operator. (See Sproat (1985) for an analysis on these lines.) It is not clear to me that this proposal does in fact remove the difficulty, because we still have the problem of determining the permissible distribution of the empty preposition. Given that Welsh has Wh-PPs, as i bwy in (8) demonstrates, we would need a non-ad hoc account of why such PPs cannot contain an empty preposition, a possibility which would again give rise to the ungrammatical (11) by a different route. x2 The particle a in (13) is itself a soft-mutation trigger - hence the appearance of the verbal form welodd.
THE SYNTAX
OF WELSH
SOFT MUTATION
297
as the isolation form of a morpheme. He argues that the distinction between (1) and (3) can be accounted for if we assume that finite verbs assign accusative case to their direct objects, whereas nonfinite verbs assign genitive. 13 Zwicky ensures conformity with the Trigger Constraint by exploiting the device of a 'phantom rule', suggested by Gazdar and Sag (1981) in GENERALIZED PHRASE STRUCTURE GRAMMAR (GPSG). Zwicky suggests that the grammar of Welsh contains rules of the form
(14) (14)
VP[FIN]---) V[FIN] NP[ACC] 1 4
'Phantom rules' are phrase structure rules which have a category on their left-hand side which does not appear on the right-hand side of any other rule in the grammar. Rule (14) is a 'phantom rule' because there are no rules in the grammar of Welsh which expand as ' . . . VP[FIN]...'. Rule (14) does, however, play a role in the syntax of Welsh by functioning as the input to the metarule (15), which induces finite VSO clauses. (15)
VP[F1N] ---)V[FIN] X ~ S[FIN]---) V[FIN] NP X.
The interaction of (14) and (15) ensures that the Trigger Constraint is met at the point at which accusative is assigned, despite the separation of verb and direct object in VSO sentences. It is, however, not clear that Zwicky's analysis achieves what he intends. He assumes that the finite verb in (14) is the trigger for accusative case assignment (and hence for soft mutation), but this property is not in fact explicitly represented in these rules. In GPSG terms it would be more accurate to state that accusative case is assigned by the rule, not the verb. This is an important point because it means that there is no intrinsic relationship between the supposed trigger (the verb) and its victim. Nothing in the version of GPSG which Zwicky was assuming
13 This suggestion is less of a deus ex machina than it sounds from my brief account. Zwicky justifies his proposal concerning genitive case by drawing attention to the similarities between arguments to nonfinite verbs and arguments to nouns. See Zwicky (1984, 391 f.) for details. 14 Zwicky does not actually supply any rules, but from what he says ("VP [ . . . . ] branches into V + (NP)+ (...). For each branching of V P into V + X , there is a corresponding branching of S into (PRT) + V + NP + X; this metarule accounts for the inflected verb constructions." (1984, p. 390)), the rules given in the text seem to be a fair reconstruction of his proposals. The rule for assigning case in nonfinite VPs would be the following: (i)
VP[-FIN]--) V[-FIN] NP[GEN].
298
STEVE H A R L O W
would p r e c l u d e Welsh f r o m having, instead of (14), the rule (16), in which accusative case is assigned to an N P n o n a d j a c e n t to the verb. (16)
V P [ F I N ] --->V [ F I N ] PP N P [ A C C ] .
A l t h o u g h (14) does require that the v e r b and its o b j e c t be sisters (the c - c o m m a n d c o m p o n e n t of the T r i g g e r Constraint), the a d j a c e n c y c o m p o n e n t remains an additional stipulation. F u r t h e r m o r e , in the m o s t r e c e n t version of G P S G ( G a z d a r et al. 1985), the separation of I m m e d i a t e D o m i n a n c e f r o m Linear P r e c e d e n c e rules m e a n s that it is n o t possible, in principle, to refer to a d j a c e n c y of constituents in a rule. Z w i c k y ' s proposals do not, therefore, ensure c o n f o r m i t y with the T r i g g e r Constraint. is In addition, Z w i c k y ' s proposals also e n c o u n t e r similar empirical difficulties to those which beset L i e b e r ' s analysis. It is necessary in a G P S G analysis of u n b o u n d e d d e p e n d e n c i e s to a c c o u n t for the distribution of case m a r k i n g in l a n g u a g e s such as G e r m a n (and m o r e marginally in English), in w h i c h the i n t r o d u c e r of an u n b o u n d e d d e p e n d e n c y inherits the case of the gap to which it corresponds, as in, for example, the G e r m a n sentences (17) and (18). (17)a.
D u hast ihn
gesehen.
you have h i m - A C C seen Y o u saw him. b.
Wen
hast du
gesehen?
w h o - A C C have you seen W h o did y o u see?
(18)a.
D u hast ihm
geholfen.
you have h i m - D A T
helped
Y o u h e l p e d him. is Observe also that Zwicky's suggestions do not extend in any natural way to examples like (5b) with inverted subjects. Additional mechanisms will be needed to handle such sentences. Zwicky also devotes some space to countering the argument that if case is involved it should show up on the head and not the periphery of a phrase (1984, 394 ft.). Even if we admit the validity of his arguments, the fact still remains that his account requires a stipulation that case is realised on the first item in a phrase, rather than, say, the last. His claim that the Control Agreement Principle (Gazdar and Pullum 1982; Gazdar et al. 1985) "predicts that determiners, quantifiers and adjectives will agree in case.., with the noun with which they are in construction", fails to notice that, since most adjectives in Welsh occur in post-head position, appeal to the CAP will make the false prediction that soft mutation will apply across an intervening head noun.
THE SYNTAX OF WELSH SOFT MUTATION
(18)b.
299
Wem hast d u geholfen? w h o - D A T have you helped Who did you help?
Wen is accusative in (17b) because sehen is a verb which requires an accusative object, whereas (18b) has dative were because helien takes a dative object. Indeed, in Gazdar et al. (1985, p. 146) the combination of the Control Agreement and Foot Feature Principles would force such a result. If soft mutation is a realisation of accusative case as Zwicky claims, we would expect that pwy in (13) should have its soft mutation form bwy since it should inherit accusative case in the same way that wen does in the German example (17b), and we have already seen above that this is not the case.
2.3. Further Problems: Soft Mutation of VPs Lieber and Zwicky both assume, in different ways, that soft mutation is in some way linked to the case of NPs. I would now like to consider some data which throw doubt on this assumption. The basic observation is that soft mutation is not in fact restricted to NPs. We can also find numerous examples in which the constituent affected is a VP, 16 as (19) to (22) illustrate. (The first segment of the bracketed words is mutated): (19)
Gall y dyn [ddreifio car]. c a n - P R E S - 3 S the man drive car T h e man can drive a car.
(20)
Penderfynodd y dyn [fynd ar unwaith]. d e c i d e - P A S T - 3 S the man go at once The man decided to go at once.
(21)
Gwnaeth hi [wrthod]. d o - P A S T - 3 S she refuse She refused.
16 There is some dispute over the categorial status of what I am treating as VPs. Traditional discussions of Welsh (Richards 1938; Williams 1959) refer to the 'verbs' in these constructions as berieaw 'verbal nouns'. Awbery (1976, p. 18) analyses the constructions themselves as NPs. However, Sproat (1985, pp. 183-189) provides a number of persuasive arguments that they are projections of nonfinite verbs, which I accept.
300 (22)
STEVE
Dylai'r
HARLOW
bws [gyrraedd am ddeg o'r gloch].
should the bus
arrive
at
ten
o'clock
The bus should arrive at 10 o'clock. The interesting thing about these examples is that the occurrence of soft mutation is exactly parallel to what we have seen in (1) and (3). That is, in VSO sentences we find the initial consonant in both direct object NPs (for example, gi in (1)) and VP complements (e.g. ddreifio in (19)) undergoing soft mutation, whereas in the corresponding periphrastic sentences the corresponding consonants of both object NPs and VP complements occur in unmutated form, as a comparison of (1) and (3) with (19) and (23) demonstrates. (1)
Gwelodd
y
saw-PAST-3S
the m a n dog
dyn gi.
The man saw a dog. (3)
Mae'r is
dyn yn
the m a n P R T
gweld ci. see
dog
The man sees the dog. (19)
Gall
y
dyn [ddreifio car].
can-PRES-3S
the m a n
drive
car
The man can drive a car. (23)
Mae'r
dyn Yn
gallu [dreifio car].
is
man PRT
able
the
drive
car
The man can drive a car. This is a completely general phenomenon and is independent of the particular lexical items involved, as illustrated by a comparison of (20), containing the mutated form fynd, with its periphrastic counterpart (24), containing the unmutated form mynd. (24)
Roedd y was
dyn yn
the m a n P R T
penderfynu [mynd ar unwaith]. decide
go
at once
The man decided to go at once. It is clear that the extension of either Lieber's or Zwicky's analyses to these examples would involve extending the targets of case assignment to VPs. Irrespective of the merits or demerits of such a move, it would appear on superficial inspection as if there is support at least for the
THE
SYNTAX
OF WELSH
SOFT
MUTATION
301
extension of Lieber's empty preposition analysis to these examples. This is because there are indeed instances of phonetically realised prepositionlike morphemes which trigger soft mutation in VPs, as in (25) and (26), where the mutated form gyrraedd is triggered by am and heb. (25)
Mae'r is
dyn am
[gyrraedd am ddeg].
the m a n about
arrive
at
ten
The man wants to arrive at ten. (26)
Mae'r is
dyn heb
[gyrraedd].
the m a n without
arrive
The man has not arrived. In the light of these facts, the suggestion would be that all of the examples in which VPs are subject to soft mutation are instances of the structure in (27). (27)
PP
P
[+voice]
VP
~
... Here the empty preposition, as in Lieber's analysis of (1), contains a floating autosegment which triggers soft mutation on the following verb. This proposal, however, is as untenable as the empty preposition argument for soft mutation in NPs, and for the same reasons. The difficulty in sustaining this idea lies in the fact that prepositions followed by constituents other than NPs obligatorily undergo pied-piping. Sentence (28a), for example, allows only (28b) as its clefted counterpart, not
(28c). (28)a.
Mae'r is
dyn wedi mynd i 'r
the m a n alter go
The man has gone to the town. b.
dref.
to the town
Wedi mynd i'r dref mae'r dyn.
c. *Mynd i'r dref mae'r dyn wedi.
302
STEVE H A R L O W
T h e implications of this for the cases u n d e r consideration should be clear: if these c o n s t r u c t i o n s involve an e m p t y preposition which is a s o f t - m u t a t i o n trigger, it t o o should u n d e r g o pied-piping, resulting in soft m u t a t i o n applying to the f r o n t e d constituent. In fact w h a t h a p p e n s in these c o n s t r u c t i o n s is that the m u t a t i o n effect disappears in f r o n t e d VPs. This c a n be seen f r o m a c o m p a r i s o n of (19) with (29). 17 (29)
Dreifio car (a)
all
drive
c a n the m a n
car P R T
y
dyn.
It is drive a car that the m a n can. T h e conclusions of the discussion so far are (a) that the case-assignm e n t interpretation of soft m u t a t i o n is implausible; (b) that soft m u t a t i o n does in fact show a d e g r e e of locality in that it fails to show up in u n b o u n d e d d e p e n d e n c i e s in the w a y e x p e c t e d f r o m L i e b e r ' s and Z w i c k y ' s analyses.
3. A N A L T E R N A T I V E I N T E R P R E T A T I O N 3.1. T h e Proposal: N P s are S o f t M u t a t i o n Triggers It is clear that neither of the two analyses we have c o n s i d e r e d p r o v i d e s a satisfactory a c c o u n t of D O M . So, w h a t is going on? I suggest that we h a v e b e e n looking in the w r o n g place. Z w i c k y and L i e b e r b o t h assume that soft m u t a t i o n is lexically triggered. T h e answer, I claim, is that, in the cases we h a v e b e e n considering, it is p h r a s a l l y triggered. L e t ' s start
17 One might try to argue that the following examples provide a basis for avoiding this conclusion. (i)
Mac 'r is
dyn yn
dreifio car.
the man P R T drive
car
The man is driving a car. (ii)
Dreifio car mae 'r drive
car is
dyn.
the man
It is driving a car that the man is. In (i) the verb dreifio is preceded by an item yn which marks the progressive. As (ii) shows, this is obligatorily deleted if the VP is clefted. This precedent might be used to argue that the absence of soft mutation in (29) is the consequence of the same deletion process applying to the assumed empty preposition. Sproat (1985, pp. 80-83), however, provides a number of arguments that progressive yn, in contrast to wedi, am and heb, is not a preposition, depriving this proposal of its plausibility.
THE
SYNTAX
OF WELSH
SOFT
MUTATION
303
by looking more carefully at the local environment for DOM. One thing that the cases we have been considering so far have in common is that they all have an NP immediately preceding the soft mutation site, as (30)-(34) show. (30)
Gwelodd [NP y saw
(31)
Gall
[NP Y dyn] ddreifio car. ( - 19) the m a n drive
the m a n go
a t once
Gwnaeth [Np hi] wrthod. (=21) did
(34)
car
Penderfynodd [r~ey dyn] fynd ar unwaith. (=20) decided
(33)
(=1)
the m a n dog
can
(32)
dyn] gi.
she refuse
Dylai [NP 'r should
bws] gyrraedd am ddeg o'r gloch. (=22)
the bus
arrive
at
ten
o'clock
I propose that we adopt the following simple suggestion which accounts for all the cases we have considered so far: 18 (35)
NPs are soft mutation triggers.
More specifically, the presence of an NP immediately preceding the victim is what triggers soft mutation. This immediately gives us the following results: (a) The adjacency component of the Trigger Constraint is maintained. (b) The interactions between soft mutation and unbounded dependencies are correctly predicted. A direct object NP or VP complement which appears in clause-initial position will not mutate because it no longer follows the trigger, i.e. the subject NP. (c) We predict that there will be no distinction between NPs and VPs whatever immediately follows an NP is a legitimate victim, irrespective of category. (d) We also account for the failure of the direct objects and VP complements of nonfinite verbs to undergo soft mutation. In these
18 Recall that it is not m y aim here to provide a phonological analysis of soft mutation. W h a t is implied by (35) is that, whatever m e c h a n i s m is responsible for ensuring that some lexical item, such as a preposition, causes soft mutation of an immediately following consonantal segment, that m e c h a n i s m will suffice to handle D O M , provided 'we assign to NPs the same triggering factor that we assign to le×ical triggers.
304
STEVE HARLOW
constructions (cf. (3)), the NP or V P immediately follows a verb, not an NP. Note also that we account for the contrast between (5a) and (5b), repeated here. (5)a.
M a e ci
is
m a w r yn yr ardd. dog big in the garden
T h e r e is a big dog in the garden. b.
Mae yn [yr
is
in
ardd]
gi
mawr.
the garden dog big
T h e r e is a big dog in the garden. As indicated, (5b) contains an NP immediately preceding the mutated item gi, whereas ei in (5a) immediately follows a nontrigger - a verb. We also account in a straightforward m a n n e r for constructions that we have not considered so far, such as the following example from Jones and T h o m a s (1977, cited in Lieber (1983, p. 174)), on the assumption that y n a is an NP. 19 (36)
M a e [NP yna] fleiddiau yn Rwsia.
is
there wolves
in Russia
T h e r e are wolves in Russia.
3.2. Potential Problems 3.2.1. Barriers to Soft Mutation Despite the predictive power of this proposal, it is not entirely unproblematical. T h e r e are apparent counterexamples to it which require
,9 Support for this assumption comes from the fact that yna can undergo Raising. (i)
Mae yna 'n debyg o fod bleiddiau yn Rwsia. is there PRT likely of be wolves in Russia There are likely to be wolves in Russia.
Flelddiau is the soft mutation form of bleiddiau.
THE SYNTAX OF WELSH SOFT M U T A T I O N
305
discussion. 2° T h e first class of these i n v o l v e s s e n t e n c e s s u c h as the following. (37)
M a e [NP gardd] g a n y
is
dyn.
garden with the m a n
T h e m a n has a g a r d e n . I n (37) the p r e p o s i t i o n g a n takes its isolation form. R u l e (35), as stated c u r r e n t l y , predicts i n c o r r e c t l y that, since gall i m m e d i a t e l y follows the s u b j e c t NP, it s h o u l d u n d e r g o soft m u t a t i o n to give (the u n g r a m m a t i c a l form) *an. T h e r e are, h o w e v e r , i n d e p e n d e n t r e a s o n s for n o t e x p e c t i n g this result. It is o n e of the characteristics of the m u t a t i o n system of W e l s h that c e r t a i n classes of words are systematically e x e m p t f r o m m u t a t i o n effects. P r e p o s i t i o n s are o n e of the c a t e g o r i e s w h i c h are e x e m p t from soft m u t a t i o n . W e c a n d e m o n s t r a t e this with a lexical soft m u t a t i o n t r i g g e r such as the c o n j u n c t i o n n e u ' o r ' . 20 Fife (1986, 181 f. and personal communication) argues that possessive constructions such as (i) provide a direct counterexample to the claim in (35). (i)
ci maer y dref dog mayor the town the dog of the mayor of the town.
Fife claims that "[o]ne place where one NP can directly follow another is in the genitive construction" (1986, p. 181), yet here we have the unmutated form tuner. However, as Borsley (1987, p. 187) points out, "This would be a problem if there were some reason to think that the example involved the following structure: (9)
NP
NP
NP
ci
maer y dref
"The obvious structure, however, is the following: (10)
NP
N
NP
ci
maer y dref
"I conclude, then, that there is no problem here, since the preeeding item is a noun and not an NP." I concur with Borsley's comment (modulo the probable substitution of N' for N in (10)).
306
STEVE HARLOW
(38)a.
te
n e u gofti
tea or b.
coffee
gwyn neu ddu
white or c.
( c i t a t i o n f o r m coJfi)
canu
( c i t a t i o n f o r m du)
black
n e u d d a w n s i o ( c i t a t i o n f o r m dawnsio)
singing or
dancing
N e u d o e s not, h o w e v e r , t r i g g e r soft m u t a t i o n o n p r e p o s i t i o n s : (39)
M a e Sion yn
is
sefyll d a n
y
goeden
John P R T stand under the tree
neu [pp d a n y bont]. or under the bridge J o h n is s t a n d i n g u n d e r t h e t r e e o r u n d e r t h e b r i d g e . (40)
M a e M a i r yn
is
m y n d g a n S i o n n e u [pp g a n E m r y s ] .
M a r y P R T go
with John or
with Emrys
M a r y is g o i n g w i t h J o h n o r with E m r y s . In t h e s e e x a m p l e s the p r e p o s i t i o n s dan ' u n d e r ' a n d g a n ' w i t h ' b o t h a p p e a r in t h e i r c i t a t i o n f o r m s . T h e f a i l u r e of g a n to u n d e r g o soft m u t a t i o n in (37) c a n thus b e s e e n as t h e c o n s e q u e n c e of an i n d e p e n d e n t l y m o t i v a t e d e x e m p t i o n . 21 A s i m i l a r c a s e of o v e r g e n e r a t i o n is i l l u s t r a t e d b y e x a m p l e s s u c h as (41).
21 Historically, Modern Welsh prepositions with voiced initial consonants acquired them as the result of the application of soft mutation. (See, for example, Morris Jones (1931, p. 406).) This 'mutation' has, however, been frozen since Middle Welsh. In the so-called 'literary' language, some prepositions may undergo 'aspirate' mutation, as the following example with gan illustrates. (i)
Yr
oedd dyn a
ehanddo law
wedi gwywo.
PRT was man and with-3MS hand after wither There was a man which had his hand withered. (Matthew, 12:10) It might be argued that this demonstrates that g n has a phonological representation in which its initial consonant is either voiceless or unspecified for voice, and that the form gan is in fact an instance of soft mutation. (See Sproat (19_86) for discussion of an analysis of this kind.) Whether or not such an analysis is justified, the important point is that prepositions are immune to environmentally triggered soft mutation. My position is, as before, that however soft mutation is to be inhibited in examples such as (39)-(40), exactly the same restriction is applicable in cases like (37).
THE
(41)
SYNTAX
OF WELSH
Dwed Sion (y) says
bydd
SOFT MUTATION
ef yn
307
cyrraedd yfory.
John P R T will be he P R T arrive
tomorrow
John says he will be arriving tomorrow. The particle y which introduces the subordinate clause in (41) is optional in Modern Welsh. It might be expected that, when the particle is not present, the verb bydd should undergo soft mutation, because it will then immediately follow the subject NP Sion. This does not happen; the verb preserves its citation form. In this case one cannot claim that finite verbs are exempt from mutation. (See Powers (1986) for discussion of the mutation of finite verbs.) Compare (42), in which the particle fe has triggered mutation to give the form fydd: (42)
Fe
fydd
ef yn
cyrraedd yfory.
P R T will be he P R T arrive
tomorrow
He will be arriving tomorrow. One might be tempted to try arguing that the reason soft mutation fails to apply in (41) is because the particle, although phonetically null, is still syntactically present and absorbs the mutation effect of the NP. The examples in (43) suggest, however, that such an account would be incorrect and, at the same time, provide a basis for a motivated explanation. 22
22 There is a subordinate clause construction in Welsh in which no preverbal particle is possible: (i)
Dywedodd Emrys [rod said
y
dyn yn
hwyr].
Emrys is~was the man P R T late
Emrys said that the man is/was late. In this example, the clause initial verb rod has undergone soft mutation. One might take this as evidence in favour of the suggestion made in the text that empty preverbal particles 'absorb' soft mutation. The argument from (i) is that the impossibility in this case of a preverbal particle means that there can be no empty particle here to act as mutation absorber, and, as a consequence, the verb does indeed undergo soft mutation. Morgan (1952, 236 f.) points out, however, that this mutation has a quite different pattern of occurrence from the other cases that we have been considering. In addition to instances like (i), where foal immediately follows the subject NP of the main clause, we also find examples like (ii), in which | o d occurs when the subordinate clause immediately follows the verb in a periphrastic construction: (ii)
Rwy 'n deall [rod fy nhyddyn yn syrthio]. I am P R T understand is my farm P R T fall I understand that my farm is falling (down).
These constructions can also appear in the soft mutation form in another environment
308
STEVE
(43)a.
HARLOW
Gwn [r~Pi] pwy (a) ddaeth yn 61. know I who P R T came back I know who came back.
b.
Gwn [NPi] pryd daeth y dyn yn ol. know I when came the m a n back I know when the man came back.
c.
Gwn [ ~ i ] p a m d a e t h y dyn y n o l . know I why came the man back I know why the man came back.
d.
Arweiniodd Lloyd-George [Nv ymgyrch benderfynol] led L - G. attack determined pan syrthiodd llywodraeth Roseberry. when fell government R . Lloyd-George led a determined attack when Roseberry's government fell.
In all of these examples the consonant immediately following an NP fails to undergo mutation, but there is no evidence here for an empty element which would serve to block the transmission of soft mutation. The crucial factor is that these examples all involve clause-initial consonants, and there is independent evidence that clauses are, like PPs, immune to soft mutation triggered from outside "the clause. This point can again be demonstrated with the conjunction neu 'or'. (44)a.
Gwrandewch ar y radio neu [s byddwch yn dawel]. listen to the radio or be P R T quiet Listen to the radio or be quiet.
b.
Awn ni yfory neu Is pan ddaw 'r meddyg]. will go we tomorrow or when comes the doctor We will go tomorrow or when the doctor comes.
which does not trigger it in the typical case - after verbs in the impersonal passive. (See below in the text for further discussion of impersonal passives.) In M o d e r n Welsh there is considerable variation as to w h e t h e r the m u t a t e d form fad or the u n m u t a t e d form b a d occurs in these constructions, s o m e speakers using f a d in all cases, others using lind throughout. It s e e m s to be the case that tod in these examples is simply a lexicalised form of the verb. (Note that this occurrence of lind/fad is not to be identified with the h o m o p h o n o u s item which occurs in nonfinite VPs. T h e latter behaves with respect to D O M exactly as expected.)
THE
SYNTAX
OF
WELSH
SOFT
MUTATION
309
The italicised clause-initial items fail to undergo soft mutation, even though they are immediately preceded by neu. The lack of soft mutation in (41) and (43) can consequently be seen to be the consequence of an independently motivated restriction on the application of soft mutation. 3.2.2. Soft Mutation and Missing NPs A further issue which requires discussion is the interaction of DOM with the gaps in Wh-constructions. As can be seen from (45), a direct object must undergo soft mutation if the subject NP is a Wh-gap. (45)
Pwy welodd _ _ who saw
gi? dog
Who saw a dog? The most straightforward way of accounting for this kind of example is to make an ontological commitment to empty NPs. Such a commitment is in fact made by a range of alternative syntactic theories. The Projection Principle of Government Binding Theory (Chomsky 1981, 38 ft.) entails the existence of an NP in this position, and recent work in GPSG has espoused an analysis of Slash Termination which also assunlles phonologically empty NPs in such constructions (Gazdar et al. 1985, 142 f.). Indeed, one might take the position that this kind of syntax/phonology interaction supports the case for the postulation of empty NPs. I conclude that there is nothing particularly problematical about examples such as (45). A more interesting question is raised by sentences like (46). (46)
Gwelodd (hi) gi. saw she dog She saw a dog.
In examples like these Welsh allows an optional pronominal subject. Whether the pronoun is present or not, DOM is obligatory. In Harlow (1981; 1983) I propose analyses of this phenomenon in which there is an empty subject. More recently McCloskey and Hale (1984), in a discussion of similar constructions in Irish, have proposed that the empty NP is to be identified as "little pro". If one accepts the argnments put forward in Harlow (1981) and if the arguments that McCloskey and Hale put forward can be reconstructed for Welsh, the case for an empty NP in these examples with missing pronominal subjects is quite strong. The behaviour of such examples with respect to DOM stands in striking contrast to so-called IMPERSONALPASSIVES like (47).
310
(47)
STEVE
Gwelir
HARLOW
ci.
s e e - I M P E R S dog
Someone sees a dog. Here ci is in its unmutated form. On the basis of (47) it would be plausible to argue that ci is in subject position, but this is not a possible analysis of (48) where the direct object appears in its normal position. (48)
Yr
ydys
yn
gweld ci.
P R T i s - I M P E R S P R T see
dog
Someone is seeing a dog. The general pattern of impersonal passives suggests that nonsubject complements simply occupy in their normal syntactic positions and that ei in (47) is in fact the direct object of the verb. So far in this paper I have attempted to maintain a position of neutrality with respect to syntactic theories (in contrast to Lieber and Zwicky) in the hope that the arguments I have been putting forward and the interpretation of DOM that I have been advocating are robust enough to survive translation into different frameworks. I believe that this neutrality can be maintained with respect to the evidence provided by impersonal passives. Since subjects in impersonal sentences are obligatorily absent, in contrast to the optionality of subjects in examples like (46), there would seem to be no reason in a framework like GPSG to posit the synactic presence of an NP in these constructions at all. What would be required is to ensure the appropriate semantic interpretation for such sentences, perhaps along the lines sketched in Dowty (1982, p. 92). In Government/Binding theory, however, such a move is not open. The Extended Projection Principle (Chomsky 1981, p. 10) will require that these sentences have a subject. Interestingly, Stenson (1989), in an extended investigation of Irish impersonal constructions, argues that the missing NP is an instance of "big PRO". Caution should be exercised in extrapolating from Irish to Welsh (and Welsh impersonals have received very little systematic attention), but if Stenson's result can be carried over to Welsh, it would be of considerable interest because it is in precisely those positions in which PRO would appear that Welsh fails to exhibit DOM effects. To illustrate this point, refer to examples (19), (23), (20) and (24), repeated here. (49)
Gall
y
dyn [ ddreifio car].
c a n - P R E S - 3 S the man drive
The man can drive a car.
car
( = 19)
THE
(50)
SYNTAX
Mae 'r is
OF WELSH
dyn yn
SOFT
MUTATION
311
gallu [dreifio car]. (=23)
the m a n P R T
able
drive
car
The man can drive a car. (51)
Penderfynodd
y
dyn [fynd ar unwaith]. (= 20)
decide-PAST-3S
the m a n
go
at once
The man decided to go at once. (52)
Roedd y was
dyn yn
penderfynu [mynd ar unwaith]. (--24)
the m a n P R T
decide
go
at once
The man decided to go at once. Both Harlow (1981) and Sproat (1985) argue that in D-structure Welsh is configurationally identical to English. That is, following Sproat (1985, p. 174), the D-structure of a Welsh sentence is as follows: (53)
I"
N"
I'
INFL
V"
The VSO order of Welsh fnite clauses is the result of verb fronting. (See Sproat (1985, pp. 198-210) for details.) If these arguments are correct, the consequence is that the bracketed subparts of (49)-(52) will have the same structure at all levels of representation as other nonfinite clauses that of (53). This means that (49)-(52) are actually of the form in (54). (54)a. b. c. d.
Gall y dyn [PRO ddreifio car]. (=49) Mae'r dyn yn gallu [PRO dreifio car]. (= 50) Penderfynodd y dyn [PRO fynd ar unwaith]. (=51) Roedd y dyn yn penderfynu [PRO mynd ar unwaith]. (= 52)
As can be seen from (54b) and (54d), the presence of PRO does not trigger DOM. Indeed, as (54a) and (54c) make clear, PRO is completely transparent to DOM. 23 However this transparency of PRO is to be accounted for, the main point for my argument is that, under a Govern-
23 Borsley (1984), who also discusses this phenomenon, uses it as part of an argument against analysesinvolving PRO.
312
STEVE HARLOW
ment/Binding analysis of the subject of impersonals as P R O , the effect of this P R O on soft mutation is identical to the effect of P R O on soft mutation in other, m o r e standard positions. I conclude that, although there are matters which require further investigation, none of the examples discussed in this section provides the basis for a rejection of the generalisation in (35).
4 . POSSIBLE EXTENSIONS Let us now return to a previous example. As we saw, the generalisation that NPs trigger soft mutation of the initial consonant of an immediately following constituent accounts for the D O M effect on gi in (55). (55)
Mae [pp yn [NP yr
is
in
ardd]] gi
mawr. (= 5b)
the garden dog big
T h e r e is a big dog in the garden. T h e r e is, however, the possibility that soft mutation of gi is not the result of the preceding NP, but rather that PPs too are soft mutation triggers. 24 T h e r e is no way of distinguishing these two alternatives, since Welsh does not permit intransitive PPs in positions which would allow us to test the different predictions. W e might however speculate still further that if there is a possibility that both NPs and PPs are soft mutation triggers, this may be an instance of a m o r e general p h e n o m e n o n - that expressed in (56). (56)
All phrasal categories are soft mutation triggers.
T h e motivation for suggesting (56) comes from a consideration of examples like those in (57)-(59). (57)a.
Fe
aeth ef i
'r
gem
ddoe.
P R T went he to the game yesterday H e went to the g a m e yesterday. b.
Ddoe,
fe
aeth ef i
'r
gem.
yesterday P R T went he to the game It was yesterday that he went to the game. 24 It is desirable from a theoretical point of view to explore this possibility further. Note that, if it is the NP which is the soft-mutation trigger in (55), this would entail a violation of the Trigger Constraint, since the NP does not c-command its victim gi. The PP node on the other hand does. I am grateful to an anonymous reviewer for reminding me of the significance of this point.
THE
(58)a.
SYNTAX
OF WELSH
SOFT MUTATION
313
Fe a 'r plant i 'r ysgol bob dydd. P R T go the children to the school every day The children go to school every day.
b.
Bob dydd, fe a 'r plant i 'r ysgol. every day P R T go the children to the school It's every day that the children go to school.
(59)a.
Fe ddaethon nhw i fyw yma ddwy flynedd yn oi. P R T came they to live here two years ago They came to live here two years ago.
b.
Ddwy flynedd yn ol, fe ddaethon nhw i fyw yma. two years ago P R T came they to live here It is two years ago that they came to live here.
In these sentences we have examples of NPs (in boldface) whose initial consonant is in soft-mutation form, but which exhibit a new pattern of distribution. This pattern is characteristic o f adjunct NPs, such as temporal or locative expressions, which differ.from our previous examples in that they maintain their soft-mutation forms even in the focus position of cleft sentences - (57b)-(59b). Furthermore, they also appear in the soft-mutation form in periphrastic constructions such as (60), in contrast to the other kinds of constituents discussed above. (60)
Mae ef yn mynd bob dydd. is he P R T go every day He goes every day.
The invariance of these expressions suggests strongly that the soft mutation that occurs here is the result of phonological reanalysis. That is, soft mutation is an intrinsic property of adjunct NPs. 25 There remains, though, the question of why these adverbial NPs should have undergone this reanalysis. Rule (56) provides the basis for a speculative answer. We have seen that these expressions occur in three different constructions: (a) VSO clauses, (b) periphrastic constructions and (c) the focus position of clefts. In a VSO clause, adjunct expressions will always occur in a position which follows a phrasal constituent (the subject NP, direct object NP or a subcategorised PP). On the basis of
25 Zwicky (1986) makes the same point.
314
STEVE HARLOW
such sentences, therefore, there will be an absence of evidence for a nonmutated form for such expressions. In periphrastic constructions, the same result would obtain if the verb and its argument NPs form a constituent, as has been repeatedly suggested in the literature (Jackendoff 1977; Gazdar et al. 1985, p. 166). Thus if (60) has either of the structures in (61), the adjunct bob dydd will again always be in a position in which soft mutation is required. (61)a.
S
V
NP
VP
ADJUNCT-NP
mae
ef
y n myfid
bob dydd
b.
S
V
mae
NP
ef
VP
VP
ADJUNCT-NP
yn mynd
bob dydd
I have no convincing evidence to offer in support of either of the alternatives in (61), but, if either is correct, the consequence is that adjunct NPs will again always appear in a position in which the softmutation form is required. The net effect of these distributional properties is that the majority of occurrences of such adverbial expressions will be in the soft-mutation form. The only position in which the unmutated form could legitimately appear would be in the focus position of a cleft sentence. If the actual frequency of occurrence in speech of clefts is lower than that of the other two construction types (something which would require empirical confirmation), it is likely that the pressure of this evidence would have led children acquiring Welsh to opt for the simpler hypothesis that soft mutation is a marker of the function of these NPs instead of a marker of a structural relationship. 26 26 H o n e s t y compels me to confess that there is one construction in which m y proposals fare no better that any of their predecessors: coordinate constructions. O n e would reasonably
T H E S Y N T A X OF WELSH SOFT M U T A T I O N
315
5. C O N C L U S I O N A c k n o w l e d g i n g t h e f r a n k l y s p e c u l a t i v e n a t u r e of the last s e c t i o n , I h a v e n o w r e a c h e d t h e p o s i t i o n w h e r e we c a n s u m up the results of t h e p r e c e d i n g discussion. I h a v e t r i e d to show in this p a p e r t h a t existing a c c o u n t s of D O M fail to a c c o u n t for the full r a n g e of e m p i r i c a l facts. I h a v e p r o p o s e d an a l t e r n a t i v e a c c o u n t of the p h e n o m e n o n whic]h p r o v i d e s a g e n e r a l s o l u t i o n to a w i d e r a n g e of W e l s h m u t a t i o n effects. M y c o n c l u s i o n is t h a t t h e T r i g g e r C o n s t r a i n t d o e s in f a c t h o l d for w h a t a p p e a r s to b e a n o n l o c a l m u t a t i o n . T h e final j u d g e m e n t c o n c e r n i n g t h e v a l i d i t y of m y p r o p o s a l still to s o m e e x t e n t d e p e n d s o n f u r t h e r inv e s t i g a t i o n of the s y n t a x of W e l s h . H o w e v e r , o n the basis of the e v i d e n c e t h a t I h a v e r e v i e w e d h e r e it fares b e t t e r t h a n its c o m p e t i t o r s . A m o r a l of this s t o r y is t h a t c a r e n e e d s to b e e x e r c i s e d in a c c e p t i n g d a t a at f a c e v a l u e . T h e i n t e r p r e t a t i o n w h i c h I h a v e a t t e m p t e d to p l a c e o n the facts of D O M in W e l s h m e a n s t h a t the T r i g g e r C o n s t r a i n t is still a realiLstic t a r g e t to a i m for in t h e c o n s t r u c t i o n of p h o n o l o g i c a l t h e o r i e s . 27
REFERENCES Awbery, Gwenllian: 1976, The Syntax of Welsh, Cambridge University Press, Cambridge. Borsley, Robert: 1984, 'VP Complements: Evidence from Welsh', Journal of Linguistics 20, 277-302. --: 1987, 'A Note on "Traditional Treatments of Welsh"', Journal of Linguistics 23, 185-190. Chomsky, Noam: 1981, Lectures on Government and Binding, Foris, Dordrecht. Dowty, David: 1982, 'Grammatical Relations and Montague Grammar', in Pauline Jacobson and Geoffrey Pullum (eds.), The Nature of Syntactic Representation, Reidel, Dordrecht, pp. 79-130. Fife, James: 1986, 'Additional Facts About Welsh VPs', Journal of Linguistics :22, 179-186. Gazdar, Gerald and Geoffrey Pullum: 1982, Generalised Phrase Structure Grammar: a Theoretical Synopsis, Indiana University Linguistics Club, Bloomington, Indiana. expect under any account of DOM which is based on case, that all of the NPs in a coordinate construction would have the same case and hence would all undergo soft mutation. This is not, however, what happens. Instead, only the initial item in the coordinate structure undergoes mutation: (i)
Bwytais i fara, menyn a ate
chaws.
I bread butter and cheese
I ate bread, butter and cheese. Fara is the soft mutation form of bara, whereas menyn is a citation form. My ]proposal also
predicts that menyn should be in its mutated form, since on any reasonable assumption about the structure of coordinate constructions menyn is immediately preceded and c-commanded by the NP dominating rata. I have no plausible account of this exception. 27 But see Stump (1988) for a more robust counterexample to the Trigger Constraint from Breton.
316
STEVE HARLOW
Gazdar, Gerald and Ivan Sag: 1981, 'Passives and Reflexives in Phrase Structure Grammar', in J. A. G. Groenendijk, T. M. V. Janssen and M. B. J. Stockhoff (eds.), Formal Methods in the Study of Language, Part I, Amsterdam Mathematisch Centrum, Amsterdam, pp. 131-152. Gazdar, Gerald, Ewan Klein, Geoffrey Pullum and Ivan Sag: 1985, Generalized Phrase Structure Grammar, Harvard University Press, Caml~ridge, Mass. Harlow, Stephen: 1981, 'Government and Relativisation in Celtic', in Frank Heny (ed.), Binding and Filtering, Croom Helm, London, pp. 213-254. •: 1983, 'Celtic Relatives', York Papers in Linguistics 10, 77-121. Jackendoff, Ray: 1977, X-bar Syntax: A Study of Phrase Structure, MIT Press, Cambridge, Mass. Jones, Morris and Alan Thomas: 1977, The Welsh Language: Studies in its Syntax and Semantics, University of Wales Press, Cardiff. Lieber, Rochelle: 1983, 'New Developments in Autosegmental Phonology: Consonant Mutation', in M. Barlow, D. Flickinger and M. Wescoat (eds.), Proceedings of the West Coast Conference on Formal Linguistics, vol. 2, Stanford Linguistics Association, Stanford, pp. 165-175. McCloskey, James and Kenneth Hale: 1984, 'On the Syntax of Person-Number Inflection in Modern Irish', Natural Language and Linguistic Theory 1, 487-533. Morgan, T. J.: 1952, Y Treigliadau a'u Cystrawen, University of Wales Press, Cardiff. Morris Jones, J.: 1931, A Welsh Grammar, Historical and Comparative, Clarendon Press, Oxford. Powers, Joyce: 1986, 'Soft Mutation on Verbs in Welsh', paper presented to the Kentucky Foreign Language Conference, Lexington, Kentucky. Richards, Morris, 1938, Cystrawen y Frawddeg Gymraeg, University of Wales Press, Cardiff. Sproat, Richard: 1985, 'Welsh Syntax and VSO Structure', Natural Language and Linguistic Theory 3, 173-216. .: 1986, 'More on Celtic Consonant Mutation: a Case for a Non-autosegmental Analysis of CM in Welsh', paper presented to the Kentucky Foreign Language Conference, Lexington, Kentucky. Stenson, Nancy: 1989, 'Irish Autonomous Impersonals', this volume. Stowell, Timothy: 1981, The Origins of Phrase Structure, unpublished Ph.D. dissertation, MIT. Stump, Greg: 1988, 'Non-local Spirantization in Breton', Journal of Linguistics 24, 457481. Williams, S. J.: 1959, Elfennau Gramadeg Cymraeg, University of Wales Press, Cardiff. Willis, Penny: 1982, The Initial Consonant Mutations in the Brythonic Celtic Languages, unpublished Ph.D. dissertation, City University of New York. Zwicky, Arnold: 1984, 'Welsh Soft Mutation and the Case of Object NPs', in J. Drogo, V. Mishra and D. Testen (eds.), Papers of the 20th Regional Meeting of the Chicago Linguistic Society, Chicago, pp. 387-402. --: 1986, 'The General Case: Basic Form Versus Default Form', Proceedings of the 12th Meeting of the Berkeley Linguistics Society, Berkeley, pp. 305-314. Received 30 December 1986 Revised 16 July 1988 Department of Language and Linguistic Science University of York Heslington, York Y01 5DD Great Britain