ON THE NATURE OF WELSH UNBOUNDED DEPENDENCIESprivaterborsley/Welshuds.pdf · ON THE NATURE OF WELSH UNBOUNDED DEPENDENCIES Robert D. Borsley ... Literary Welsh, but Colloquial Welsh

ON THE NATURE OF WELSH UNBOUNDED DEPENDENCIES

Robert D. Borsley

University of Essex

Abstract

Welsh has a variety of unbounded dependency constructions which involve both gaps and

resumptive pronouns (RPs). Gaps and RPs differ in their distribution but otherwise are

quite similar. This suggests that they involve the same mechanism. Within a

transformational approach it suggests that both involve movement, a position which is

difficult to implement. Within HPSG it suggests that both should be analyzed as

realizations of the SLASH feature. It is not difficult to develop an analysis of this kind. It

is possible to propose an analysis in which RPs are associated with the SLASH feature

and are also the ordinary pronouns which they appear to be.

Keywords: Welsh; Unbounded dependencies; Resumptive pronouns; HPSG

1. Introduction

An important feature of many languages are unbounded dependency (UD) constructions

containing a gap or in some languages a resumptive pronoun (RP) and some distinctive

higher structure. Among Welsh UD constructions are wh-interrogatives, relative clauses,

cleft sentences, free relatives and complements of adjectives like hawdd ‘easy’. The

following illustrate:

(1) a. Pwy welest ti? (wh-interrogative)

who saw.2SG you. SG

‘Who did you see?’

b. y dyn [welest ti] (relative clause)

the man saw.2SG you. SG

‘the man you saw’

c. Rhiannon welest ti. (cleft sentence)

Rhiannon saw.2SG you.SG

‘It was Rhiannon that you saw’

d. Mi wna’ i [beth bynnag (a) wnei di]. (free relative)

PRT do.FUT.1SG I what ever PRT do.FUT.2SG you.SG

‘I will do whatever you do.’

e. Mae Carys yn hawdd [i Ifor ei gweld] . (hawdd complement)

is Carys PRED easy to Ifor 3SGF see

‘Carys is easy for Ifor to see.’

In some cases, the construction contains a filler, a constituent in a non-argument position

with most of the properties of the gap or RP. In other cases there is no visible filler. Wh-

interrogatives obviously have a filler, and Borsley (2008) argues that this is also the case

in free relatives. However, Borsley argues that the initial constituent in cleft sentences is

2

not a filler. If this is right, they have no visible filler. There is also no visible filler in

relative clauses and hawdd complements1. Whether or not they have a visible filler is a

major way in which UD constructions vary. They vary in a number of other ways, and

Borsley (2008) argues that Head-driven Phrase Structure Grammar (HPSG) is well

equipped to accommodate the differences. Here, however, I want to focus on the

similarities. The main one is that they involve both gaps and RPs.

There has been considerable discussion of Welsh UD constructions within various

versions of transformational grammar (see, for example, Hendrick 1988, Rouveret 1994,

2002, Sadler 1988, and Willis 2000, 2011). However, there are many unresolved issues.

There has been very little discussion of Welsh UD constructions within non-

transformational frameworks.2 Here I will seek to establish the most important facts and

to develop an analysis within HPSG. I will also comment on the implication of the facts

for transformational approaches.

In any language which allows both gaps and RPs, a central question is: how similar or

how different are they? I will show that they differ in their distribution but otherwise are

quite similar. This suggests that they involve the same mechanism. Within a

transformational approach, it suggests that both involve movement, while in HPSG, it

suggests that both involve the SLASH feature. I will argue that it is difficult to combine

movement with RPs. However, it is not difficult to develop an analysis within HPSG

which treats RPs as a realization of the SLASH feature. SLASH makes information about

gaps available higher in the structure, but there is no necessary connection between

SLASH and gaps, and it is quite easy to associate SLASH with a pronoun. On the

analysis that I will develop, RPs introduce a complication into the grammar. I will argue

that this is not a weakness.

Most work on Welsh UD constructions has focused on Literary Welsh, including inter

alia Hendrick (1988), Rouveret (1994, 2002), and Sadler (1988). This is rather surprising

since the native language of all speakers is a form of Colloquial Welsh and where

speakers have some command of Literary Welsh it is the result of formal teaching. In

many respects Literary Welsh reflects earlier forms of the language, but it has also been

influenced by prescriptive grammarians. In other words, it is a rather artificial system.

For this reason, I will not focus on the literary language. Rather I will follow Willis

(2001, 2011) and Borsley, Tallerman and Willis (2007) and focus on Colloquial Welsh.

The paper is organized as follows. In section 2, I provide a theory-neutral description

of the facts. In section 3, I consider transformational approaches to the data. Then, in

section 4, I develop an analysis within HPSG. Finally, in section 5, I provide some

concluding remarks.

1 Adjunct relative clauses allow the wh-words lle ‘where’, pryd ‘when’ and pam ‘why’ as fillers. The

following from Borsley, Tallerman and Willis (2007: chapter 4) illustrates the first of these:

(i) yr ardal lle gafodd ei fagu

the district where get.PAST.3SG 3SGM raise

‘the district where he was brought up’

2 Harlow (1983) outlined an analysis of Literary Welsh relative clauses within Generalized Phrase

Structure Grammar.

3

2. The Data

The facts involving gaps and RPs are quite complex. In this section, I will first consider

their basic distribution. Then I will highlight some important similarities between them.

Finally, I will look at the possible relevance of island effects in this area. I will argue that

islands do not distinguish between gaps and RPs.

2.1. The distribution of gaps and resumptive pronouns (RPs)

Before we consider the distribution of gaps and RPs, we need to say something about the

behaviour of pronouns and non-pronominal NPs. Pronouns, including RPs, are associated

with agreement in a number of positions. In each case it is also possible to have the

agreement with no visible pronoun. However, there is evidence from mutation (Borsley

1999) and agreement (Borsley 2009) that there is a phonologically null pronoun in this

situation. Non-pronominal NPs do not trigger agreement in the way that pronouns do, and

we will see that nominal gaps generally behave like non-pronominal NPs. As a result it is

not too hard to distinguish between true gaps and phonologically null RPs. I will say

more about this when we consider the various nominal positions.

We begin with the highest subject position. Welsh, like the other Celtic languages, is

a VSO language with verb-subject order in finite clauses. Finite verbs agree with a

following pronominal subject, which may be phonologically null, as the following

illustrate:

(2) a. Enillodd (e/ hi) y wobr.

win.PAST.3SG he she the prize

‘He/she won the prize.’

b. Enillon (nhw) y wobr.

win.PAST.3PL they the prize

‘They won the prize.’

With a non-pronominal subject, singular or plural, the third person singular form of the verb

appears.3

(3) Enillodd y myfyriwr/ myfyrwyr y wobr.

win.PAST.3SG the student students the prize

‘The student(s) won the prize.’

(4) *Enillon y myfywwyr y wobr.

win.PAST.3PL the students the prize

‘They won the prize.’ 3 There are two views one might take of this datum. One might assume that finite verbs have a form with

no agreement features with a non-pronominal subject but that it looks just like the third person singular

form. This approach would make finite verbs like other heads considered below which have a form with no

agreement features. Alternatively, one might assume that a special mechanism ensures that a finite verb

with a non-pronominal subject has the same agreement features as a finite verb with a third person singular

pronominal subject.

4

As one might expect, only a gap is possible in the highest subject position. We have

examples like the following, where the gap is indicated by ‘___’:

(5) Pa fyfyrwyr enillodd ___ y wobr?

which students win.PAST.3SG the prize

‘Which students won the prize?’

Here the gap is presumably third person plural, but the verb is third person singular. It is

clear, then, that it is not a phonologically null RP but a real gap behaving like a non-

pronominal NP. The following shows that neither an overt RP nor a null RP is possible in

the highest subject position:

(6) *Pa fyfyrwyr enillon (nhw) y wobr?

which students win.PAST.3PL they the prize

‘Which students won the prize?’

Next we can consider the highest object position. There is no agreement associated

with the object of a finite verb. Hence there is no difference between pronouns and non-

pronominal NPs in this position and no possibility of a null RP. The following show that

only a gap and not an RP is possible in this position.

(7) a. Beth welest ti ___?

what see.2SG you

‘What did you see?’

b. *Beth welest ti fo?

what see.2SG you he

‘What did you see?’

The data are not so clear, but it seems that embedded subject and object positions

also allow gaps but not RPs in the colloquial language. Both positions allow RPs in

Literary Welsh, but Colloquial Welsh is different. Consider first the following examples

from Willis (2001).

(8) a. Pa lyfrau wyt ti ’n meddwl oedd ___ yn addas?

which books be.PRES.2SG you PROG think be.IMPF.3SG PRED suitable

‘Which books do you think were suitable?’

b. *Pa lyfrau wyt ti ’n meddwl oedden (nhw)

which books be.PRES.2SG you PROG think be.IMPF.3PL they

’n addas?

PRED suitable


The gap in (8a) like that in (5) is presumably third person plural, but the verb is again

third person singular. (8b) with an overt RP or a null RP is unacceptable. Willis suggests

that such examples are unacceptable for an independent reason. However, as discussed in

5

section 2.2, it is clear that his explanation is untenable. Willis notes that some examples

with an RP in an embedded subject are more acceptable than this. I assume this is a

reflection either of the fact that RPs are acceptable in this position in Literary Welsh or of

the fact that RPs are generally more acceptable in embedded contexts (see e.g. Erteschik-

Shir 1992 and Alexopoulou and Keller 2007). As for embedded objects, a gap is again

fine, but an RP is quite marginal in the colloquial language:

(9) y llyfrau yr wyt ti ’n meddwl [y darllenai

the books PRT be.PRES.2SG you PROG think PRT read.COND.3SG

Megan ___]

Megan

‘the books you think Megan would read’

(10) ??y llyfrau yr wyt ti ’n meddwl [y darllenai

the books PRT be.PRES.2SG you PROG think PRT read.COND.3SG

Megan nhw]

Megan they

‘the books you think Megan would read’

Embedded objects seem more acceptable than object gaps which are not embedded, but

this is not surprising since, as we have noted, RPs are generally more acceptable in

embedded contexts.

I turn next to object of a non-finite verb. Non-finite verbs show agreement in the form

of a clitic with a pronominal object.4

(11) a. Gwnaeth Emrys ei weld o.

do.PAST.3SG Emrys 3SGM see he

‘Emrys saw him.’

b. Gwnaeth Emrys ei gweld hi.

do.PAST.3SG Emrys 3SGF see she

‘Emrys saw her.’

c. Gwnaeth Emrys eu gweld nhw.

do.PAST.3SG Emrys 3PL see they

‘Emrys saw them.’

The third person singular masculine and feminine clitics are identical in form but differ in

their mutation effects. The former triggers soft mutation and is responsible here for the

loss of initial /g/ in (11a). The latter triggers aspirate mutation, which is not seen here

because it only effects /p/, /t/ and /k/. There is no clitic with a non-pronominal object, as

the following show:

(12) Gwnaeth Emrys weld y bachgen/bechgyn.

do.PAST.3SG Emrys see the boy boys

‘Emrys saw the boy/boys.’

4 Non-finite verbs are traditionally known as verb-nouns. This label is potentially misleading. See Borsley

(1993, 1997) and Borsley Tallerman and Willis (2007: 3.1) for relevant discussion.

6

(13) a. *Gwnaeth Emrys ei weld y bachgen.

do.PAST.3SG Emrys 3SGM see the boy

b. *Gwnaeth Emrys eu gweld y bechgyn.

do.PAST.3SG Emrys 3PL see the boys

Consider now a relevant wh-interrogative:

(14) Beth ydych chi ’n ei fwyta ___?

what be.PRES.2PL you PROG 3SGM eat

‘What are you eating?’

Here, the gap is associated with agreement in the form of a clitic, which triggers soft

mutation on the following verb, whose basic form is bwyta. This might suggest that the

gap in an example like (14) is really a phonologically null RP, and this is the conclusion

that a number of researchers have reached (see Awbery 1977, Sadler 1988 and Rouveret

2002: 124). There are, however, reasons for doubting that this is right. First, as

emphasized in Willis (2000: 545), an overt RP is not possible in this position:

(15) *Beth ydych chi ’n ei fwyta o?

what be.PRES.2PL you PROG 3SGM eat he

‘What are you eating?’

As we will see below, an overt RP is possible in positions which clearly contain an RP.

Second, as noted in Borsley, Tallerman and Willis (2007: 114), Colloquial Welsh allows

a third person singular masculine clitic with a feminine wh-phrase such as pa gath ‘which

cat’. Thus, we have not just (16) but also (17).

(16) Pa gath ydych chi ’n ei phrynu ___?

which cat be.PRES.2PL you PROG 3SGF buy

‘Which cat are you buying?’

(17) Pa gath ydych chi ’n ei brynu ___?

which cat be.PRES.2PL you PROG 3SGM buy

It is also possible to have a third person singular masculine clitic when the wh-phrase is

plural. Thus, instead of (18), (19) may occur.

(18) Pa lyfre ydych chi ’n eu prynu ___?

which books be.PRES.2PL you PROG 3PL buy

‘Which books are you buying?’

(19) Pa lyfre ydych chi ’n ei brynu ___?

which books be.PRES.2PL you PROG 3SGM buy

It is possible also to have no clitic and just a soft mutated verb:

7

(20) Pa gath ydych chi ’n brynu ___?

which cat be.PRES.2PL you PROG buy

‘Which cat are you buying?’

(21) Pa lyfre ydych chi ’n brynu ___?

which books be.PRES.2PL you PROG buy

‘Which books are you buying?’

It is also possible although more colloquial to have an unmutated verb:

(22) Pa lyfre ydych chi ’n prynu ___?

which books be.PRES.2PL you PROG buy

(23) Pa gath ydych chi ’n prynu ___?

which cat be.PRES.2PL you PROG buy

Again this is very different from what we find in positions which clearly contain an RP. I

will assume, then, with Willis (2011) that we have a true gap here. I will propose in

section 4.5 that the appearance of clitics in these examples is the result of a special

constraint.

As one might expect, we have gaps and not RPs in non-nominal positions, especially

adverbial positions, as in (24), and PP argument of an adjective, as in (25).5

(24) a. Sut gwyddost/ wyddost ti hyn ___?

how know.PRES.2SG you DEM

‘How do you know that?’

b. Pryd cest/gest ti dy benblwydd ___?

when get.PAST.2SG you 2SG birthday

‘When did you have your birthday?’

(25) Am beth mae Gwyn yn siwr ___?

about what be.PRES.3SG Gwyn PRED certain

‘About what is Gwyn certain?’

We do not have gaps as PP arguments of nouns. Rather than (26a), we have (26b), with a

complex NP filler.

(26) a. *Am bwy wyt ti ’n darllen llyfr ___?

about what be.PRES.2SG you PROG read book

‘About what are you reading a book?’

b. Llyfr am bwy wyt ti ’n ei ddarllen ___?

book about what be.PRES.2SG you PROG 3SGM read

‘A book about what are you reading?’

We turn now to positions where only an RP is possible. The first is prepositional

object position. Most prepositions agree with a following pronominal object, which may be

phonologically null. The following illustrate:

5 Adverbial wh-words such as sut and pryd are sometimes followed by soft mutation and sometimes not in

Colloquial Welsh.

8

(27) a. arno (fo)

on.3SGM he

‘on him’

b. arni (hi)

on.3SGF she

‘on her’

c. arnyn (nhw)

on.3PL they

‘on them’

Unlike finite verbs, inflecting prepositions have separate masculine and feminine third

person singular forms. With a non-pronominal object the basic uninflected form of the

preposition, ar in the present case, appears. Thus, we have (28) and not (29).

(28) ar y bachgen/ yr eneth/ y bechgyn

on the boy the girl the boys

‘on the boy/the girl/the boys’

(29) a. *arno ’r bachgen

on.3SGM the boy

b. *arni ’r eneth

on.3SGF the girl

c. *arnyn y bechgyn

on.3PL the boys

Consider now a relevant relative clause:

(30) y dyn werthodd Ieuan y ceffyl iddo (fo)

the man sell.PAST.3SG Ieuan the horse to.3SGM he

‘the man that Ieuan sold the horse to’

Here, we have an RP which may be phonologically null or overt.6 We also find wh-

interrogatives with an RP in prepositional object position (Borsley, Tallerman and Willis

2007: 115), but it is the norm when the object of a preposition is questioned for the whole

PP to be fronted. An RP that is object of an inflected preposition must also agree with the

antecedent in a relative clause or the wh-phrase in a wh-interrogative. There is no

possibility of a masculine RP and masculine agreement if the antecedent or wh-phrase is

feminine. The following illustrate for relative clauses:

(31) a. y ferch werthodd Ieuan y ceffyl iddi (hi)

the woman sell.PAST.3SG Ieuan the horse to.3SGF she

‘the woman that Ieuan sold the horse to’

b. *y ferch werthodd Ieuan y ceffyl iddo (fo)

the woman sell.PAST.3SG Ieuan the horse to.3SGM he

6 Overt RPs in positions associated with agreement are viewed as unacceptable in Literary Welsh

(Rouveret 1994: 381). However, it is clear that they are acceptable in the colloquial language.

9

Thus, the situation here is quite different from that when object of a non-finite verb is

questioned or relativized.7

Since nominal gaps appear to be non-pronominal, we would expect a preposition with

a gap as its object to be uninflected. In other words, we would expect examples like the

following:

(32) *y dyn werthodd Ieuan y ceffyl i ___

the man sell.PAST.3SG Ieuan the horse to

‘the man that Ieuan sold the horse to’

Such examples are ungrammatical except in very colloquial forms of Welsh.

The second position in which only an RP is possible is the possessor position within

an NP. When a noun is followed by a pronominal possessor, it is preceded by an agreeing

clitic, as the following illustrate:

(33) a. ei dad o

3SGM father he

‘his father’

b. ei thad hi

3SGF father she

‘her father’

c. eu tad nhw

3PL father they

‘their father’

Again we see that the third person singular masculine and feminine clitics are identical in

form but differ in their mutation effects. There is no clitic with a non-pronominal

possessor, as the following show:

(34) tad y bachgen/ bechgyn

father the boy boys

‘the boy’s/boys’ father’

(35) a. *ei dad y bachgen

3SGM father the boy

b. *ei thad yr eneth

3SGF father the girl

c. *eu tad y bechgyn

3PL father the boys

7 With prepositions which do not show agreement, a pronoun, including an RP, must be overt. Here is a

relevant example:

(i) y bêl mae o ’n chwarae efo hi

the ball bePRES.3SG he PROG play with she

‘the ball that he is playing with’

10

Consider now a relevant relative clause:

(36) y dyn weles i ei chwaer (o)

the man see.PAST.1SG I 3SGM sister he

‘the man whose sister I saw’

Here, we have an RP, which may be null or overt. It must agree with the antecedent in a

relative clause, and there is no possibility of a masculine RP and masculine agreement if

the antecedent is feminine, as the following show:

(37) a. y ferch weles i ei thad (hi)

the woman see.PAST.1SG I 3SGF father she

‘the woman whose father I saw

b. *y ferch weles i ei dad (o)

the woman see.PAST.1SG I 3SGM father he

Again, the situation here is quite different from that when object of a non-finite verb is

questioned or relativized.

Assuming that nominal gaps are non-pronominal, we would expect a noun with a

gap as a possessor to have no clitic. In other words, we would expect examples like the

following:

(38) *y dyn weles i chwaer ___

the man see.PAST.1SG I sister

‘the man whose sister I saw’

Such examples are completely ungrammatical.

The facts are not entirely straightforward, but it seems reasonable to assume that the

distribution of gaps and resumptive pronouns in Colloquial Welsh is as follows:

Gap RP

Subject Yes No

Object of finite verb Yes No

Embedded subject Yes No (except in a

literary register)

Embedded object a finite verb Yes No (except in a

literary register)

Object of non-finite verb Yes No

Adjunct Yes No

PP dependent of an adjective Yes No

PP dependent of a noun No No

Object of preposition No (except in very

colloquial registers)

Yes

Possessor No Yes

Table 1: The distribution of gaps and resumptive pronouns

11

It seems, then, that the situation in Colloquial Welsh is as follows:

Gaps are possible as arguments of verbs and adjectives and in adverbial positions.

RPs are possible as arguments of prepositions and nouns.

In section 4, I will show how this situation can be accommodated in an HPSG analysis.

There is one further matter that we should say something about here. As noted in the

introduction, much work on Welsh UD constructions has focused on Literary Welsh, and

it has been much concerned with two particles a and y, which are a feature of these

constructions in the literary language. The following illustrate:

(39) Pwy a gafodd ___ y wobr?

who PRT get.PAST.3SG the prize

‘Who got the prize?’

(40) y dyn y gwerthodd Ieuan y ceffyl iddo

the man PRT sell.PAST.3SG Ieuan the horse to.3SGM

‘the woman that Ieuan sold the horse to’

It has been assumed that a appears when the highest subject or object position is involved

in a UD and y when any other position is involved in a UD, and analyses have been

developed to accommodate this fact. However, as explained in Willis (2001: section 3),

this is essentially a mid twentieth century prescriptive rule, found, for example, in

Watkins (1961). It does not occur in earlier work, e.g. Richards (1938), based on the

usage of the Bible and nineteenth- and early twentieth-century literature, which allows a

more widely. In the circumstances, it is not really clear why the rule has had the attention

that it has from theoreticians. However, in the present context the important point is that

the two particles do not generally occur in the colloquial language. Since it is the

colloquial language that I am concerned with here, I will say no more about them.8

2.2. Some similarities between gaps and resumptive pronouns

If the preceding discussion is sound, gaps and RPs appear in disjoint sets of environments

in Colloquial Welsh. In this section I will show that they are similar in some important

ways.

One similarity between gaps and RPs involves certain restrictions on tense. A notable

feature of Welsh is that present forms of bod ‘be’ and for some speakers imperfect forms as

well do not appear in affirmative complement clauses. Thus, (41) is ungrammatical, and for

some speakers (42) is too.

8 A full account of UDs will need to ensure that the verb undergoes soft mutation in a relative clause and

after some wh-elements in a wh-interrogative (pwy ‘who’, beth ‘what’, but not. lle ‘where’). There is no

obvious difficulty here, but I will not propose an account in the present paper.

12

(41) *Mae Aled yn credu [y mae Elen yn darllen

be.PRES.3SG Aled PROG believe PRT be.PRES.3SG Elen PROG read

y llyfr].

the book

‘Aled believes that Elen is reading the book.’

(42) %Mae Aled yn credu [roedd Elen yn darllen y llyfr].

be.PRES.3SG Aled PROG believe be.IMPF.3SG Elen PROG read the book

‘Aled believes that Elen was reading the book.’

Instead of these forms, what looks like the non-finite form bod appears. Thus, the

grammatical counterpart of (41) and (42) is (43).9

(43) Mae Aled yn credu [bod Elen yn darllen y llyfr].

be.PRES.3SG Aled PROG believe be Elen PROG read the book

‘Aled believes that Elen is/was reading the book.’

What is important in the present context is that the ban on the present and imperfect forms of

bod is nullified by a UD. Thus, both the following are fine:

(44) Beth mae Aled yn credu [y mae Elen yn ei

what be.PRES.3SG Aled PROG believe PRT be.PRES.3SG Elen PROG 3SGM

ddarllen ___]?

read

‘What does Aled believe that Elen is reading?’

(45) Beth mae Aled yn credu [roedd Elen yn

what be.PRES.3SG Aled PROG believe be.IMPF.3SG Elen PROG

ei ddarllen ___]?

3SGM read

‘What does Aled believe that Elen was reading?’

Willis (2000: 556) suggests that it is only UDs involving a gap that have this effect. He

cites the following example as evidence that UDs involving an RP do not nullify the ban:

(46) *Pa lyfrau wyt ti ’n meddwl oedden (nhw)

which books be.PRES.3SG you.SG PROG think be.IMPF.3SG they

’n addas?

PRED suitable


Notice, however, that this has an RP in an embedded subject position. I suggested earlier

that RPs are barred from this position in the colloquial language. I suggest that it is this

and not the ban on the imperfect of bod that is responsible for the status of this example.

Consider instead the following examples:

9 Tallerman (1998) and Borsley, Tallerman and Willis (2007: 3.3) show that there is evidence that bod-initial

clauses are probably finite, but this is not particularly important in the present context.

13

(47) y llyfr mae pawb yn dweud mae / roedd Mair

the book be.PRES everyone PROG say be.PRES.3SG be.IMPF.3SG Mair

yn sôn amdano fe

PROG talk about.3SGM he

‘the book that everyone says Mair is/was taking about’

(48) y dyn mae pawb yn dweud mae / roedd

the man be.PRES.3SG everyone PROG say be.PRES.3SG be.IMPF.3SG

ei dad o ’n glyfar

3SGM father he PRED clever

‘the man whose father everyone says is/was clever’

These examples have RPs in positions in which they are unproblematic, prepositional

object position and possessor position, respectively. In both cases the RP is inside a

complement clause where the verb is the present or imperfect tense of bod. Hence, they

show clearly that UDs with an RP nullify the ban on the present and imperfect of bod just

as much as UDs with a gap do.

A second similarity, highlighted by Willis (2011), involves non-finite verbs that

appear between the top and the bottom of a UD. We saw in section 2.1 that a non-finite

verb is preceded by a clitic if its object is questioned. We also noted that it is possible to

have a third person singular masculine clitic when the wh-phrase is feminine or plural. It

is also possible to have a third person singular masculine clitic with a higher non-finite

verb, as the following from Willis (2011) illustrates:

(49) Beth wyt ti ’n (ei) feddwl bod hyn yn (ei)

what be.PRES.2SG you PROG 3SGM think be this PROG 3SGM

olygu ___?

mean

‘What do you think this means?’

Here the object of a non-finite verb in a subordinate clause is being questioned and the

verb is mutated and optionally preceded by a third person singular masculine clitic. The

non-finite verb in the main clause is also mutated and optionally preceded by a clitic.

Consider now the following example also from Willis (2011):

(50) y llyfr roedd pawb yn (ei) feddwl oedd Mair

the book be.IMPF.3SG everyone PROG 3SGM think be.IMPF.3SG Mair

yn sôn amdano fe

PROG talk about.3SGM he

‘the book that everyone thought that Mair was talking about’

Here the object of a preposition in a subordinate clause is being relativized and we have

an overt RP. Again we have a non-finite verb in the higher clause and again it is soft

mutated and optionally preceded by a clitic. In other words, we have exactly the same

effects on a higher non-finite verb as in (49).

14

A further similarity involves coordinate structures. It has been well known since Ross

(1967) that UDs are subject to the Coordinate Structure Constraint, which essentially says

that a UD may not affect one conjunct of a coordinate structure unless it affects the

other(s), in which case it is commonly referred to as an across-the-board dependency.10 In

the case of Welsh, it rules out (51) while allowing (52).

(51) *y dyn [welais i ___ a gwelaist tithau Megan]

the man see.PAST.1SG I and see.PAST.2SG you Megan

*‘the man that I saw and you saw Megan’

(52) y dyn [welais i ___ a gwelaist tithau ___ hefyd]

the man see.PAST.1SG I and see.PAST.2SG you too

‘the man that I saw and you saw too’

(52) has a gap in both clauses. Consider now the following:

(53) y dyn [welais i ___ a soniais amdano fo]

the man see.PAST.1SG I and talk.PAST.1SG about.3SGM he

‘the man that I saw and talked about’

(54) y dyn [welais i ___ ac oeddwn i ’n nabod ei dad o]

the man see.PAST.1SG I and be.IMPF.1SG I PROG know 3SGM father he

‘the man who I saw and whose father I knew’

These examples have a gap in the first clause and an RP in the second. It seems, then, that

gaps and RPs have the same status as far as the Coordinate Structure Constraint is

concerned.11

It seems, then, that there are three important similarities between gaps and RPs. They

both nullify the ban on present and imperfect forms of bod in an affirmative complement

clause, they both allow mutation and an optional clitic on a higher non-finite verb, and they

behave in the same way with respect to the Coordinate Structure Constraint. Any analysis

must accommodate these similarities.

2.3. Islands: a further difference between gaps and resumptive pronouns?

So far we have seen that gaps and RPs appear in disjoint sets of environments but are

similar in a number of important ways. It has often been suggested that RPs unlike gaps

allow violations of island constraints. For example, Borsley, Tallerman and Willis (2007:

146) claim that ‘[t]he resumptive strategy may also be used freely to void many island

effects’. Clearly this is something that we need to look into.

10 Work by Goldsmith (1985), Lakoff (1986), and Kehler (2002) has shown that the Constraint only applies

when the conjuncts are parallel in certain ways. However, this is not particularly important in the present

context.

11 Similar examples are found in Swedish (Zaenen, Engdahl and Maling 1981) and Hebrew (Sells 1984,

Vaillette 2002).

15

The first point to emphasize here is that the facts are not very clearcut. Borsley,

Tallerman and Willis (2007: 147) consider the following example:

(55) y rhai hynny y gwnawn [ymdrech deg i ’w cuddio ___]

the ones DEM PRT make.FUT.1PL effort fair to 3PL hide

â gwên-wneud wrth sôn am rywun neu rywbeth diflas

with smile-make by talk about someone or something boring

‘those that we make a fair attempt to hide with a put-on smile by talking

about someone or something boring.’

This example involves the relativization of the object of a non-finite verb inside a

complex NP consisting of a noun and clausal complement, which is marked by brackets.

Borsley, Tallerman and Willis suggest that the non-finite verb is followed by a null RP.

Recall, however, that we argued in section 2.1 that object of a non-finite verb is a position

in which only a true gap can appear and not an RP. Part of the evidence for this comes

from the fact that an overt pronoun is not generally possible in this position. Bob Morris

Jones informs me that an overt pronoun is not very good in an example like this. It looks,

then, as if such examples probably involve not an RP but a gap. If so, what they suggest

is that certain types of complex NP may contain a gap.

Some further evidence for this position comes from the following example from

Tallerman (1983: 201) discussed by Borsley, Tallerman and Willis (2007: 147):

(56) Dyma ’r dyn y credodd Dafydd [y si [y gwelodd

here-is the man PRT believe.PAST.3SG Dafydd the rumour PRT see.PAST.3SG

Mair (o)]].

Mair he

‘Here’s the man who David believed the rumour that Mair saw.’

Here we have the relativization of the object of a finite verb inside a complex NP

consisting of a noun and clausal complement. Notice that the pronoun is marked as

optional. Tallerman comments that whether it is present or absent ‘appears to make little

or no difference to the acceptability of such sentences to native speakers’. This suggests

that a gap is possible within some complex NPs since there is no possibility of a

phonologically null RP here. I suggested earlier that RPs are ungrammatical as object of a

finite verb. I also noted, however, that RPs are generally more acceptable in embedded

contexts. I suggest that this is what is responsible for the status of (56) when it contains

an RP. As we might expect, similar examples with an RP in a standard RP position are

also acceptable. Here is an example:

(57) Dyma ’r dyn y credodd Dafydd [y si [y cest

here-is the man PRT believe.PAST.3SG Dafydd the rumour PRT get.PAST.2SG

ti ’r llythyr ’na ganddo (fo)]].

you the letter DEM with.3SGM him

‘Here’s the man who David believed the rumour that you got that letter from.’

16

It looks, then, as if both gaps and RPs are fairly acceptable within a complex NP

consisting of a noun and clausal complement.

Borsley, Tallerman and Willis (2007: 148) also consider the following example,

adapted from Tallerman (1983: 198):

(58) *Dyma ’r ffenest darais i [’r bachgen [dorrodd ____ hi ddoe]].

that-is the window hit.PAST.1SG I the boy break.PAST.3SG she yesterday

*‘That’s the window that I hit the boy who broke it yesterday.’

This unquestionably contains an RP, the third person singular feminine pronoun hi,

reflecting the fact that the antecedent ffenest is a feminine noun. Like (57), (58) involves

a complex NP. However, whereas (57) contains a complement clause (58) contains a

relative clause. This presumably accounts for their different status. As one might expect,

an example like (58) but with a gap instead of the RP is also bad.

(59) *Dyma ’r ffenest darais i [’r bachgen [dorrodd ____ ___ ddoe]].

that-is the window hit.PAST.1SG I the boy break.PAST.3SG yesterday

*‘That’s the window that I hit the boy who broke it yesterday.’

Thus, it seems that neither a gap nor an RP is acceptable inside a relative clause.

It seems, then, that both gaps and RPs are possible inside the clausal complement of

a noun but that both are impossible inside a relative clause. Hence, it is not obvious that

there are any differences between gaps and RPs with respect to islands. It is worth adding

that if we did find some differences between RPs and gaps in this area, it would not

necessarily follow that the grammar needs to treat them differently. It has been argued

e.g. by Kluender (1998), Levine and Hukari (2006), Hofmeister and Sag (2010), and

Hofmeister, Staum Casasanto, and Sag (in press) that island phenomena are a processing

matter. If this is right, any differences would not necessitate differences in syntactic

analysis.

There is no doubt more to be said here, but there appears to be no clear evidence from

island phenomena for a fundamental difference between gaps and RPs. It seems, then,

that they are broadly similar, the main difference being in their local environment, as

spelled out at the end of section 2.1.

3. TG approaches

The main goal of the present paper is to demonstrate that HPSG can provide a fairly

straightforward analysis of the main facts about Welsh UDs set out in the preceding

sections. However, before I turn to HPSG, I want to say something about the possible

implications of the data for transformational approaches. A satisfactory analysis needs to

capture the similarities between gaps and RPs highlighted in 2.2. These suggest that the

two types of UD involve the same mechanism. The question, then, is whether this is

possible within a transformational approach. One point to note at the outset is that there is

very large body of potentially relevant transformational literature. There is no possibility

of discussing all of it in detail. I will concentrate on work that seems to be particularly

relevant.

17

Within transformational work UDs which involve a gap are standardly analyzed as the

product of A-movement, movement to Spec CP. Movement leaves behind a copy, which

is deleted in PF. Hence, a typical UD construction has the following form, where ‘XP’

represents the deleted copy:

(60) CP

XP C

C TP

XP

Where there is no visible moved constituent, an invisible moved constituent is assumed.

UDs with an RP are commonly seen as involving no movement. However, as Willis

(2011) notes, the similarities between UDs with a gap and UD’s with an RP which we

have highlighted suggest that the latter should also be analysed in terms of movement.

But before we accept this. there are certain other positions that we should consider.

Although UDs with a gap are standardly analyzed as the product of A-movement,

Adger and Ramchand (2005) argue that Scots Gaelic UDs with a gap do not involve

movement. They propose that the gap is a phonologically empty pronoun and that is

associated with some binding element by the Minimalist operation Agree. Whatever may

be the case in Scots Gaelic, this approach seems untenable in Welsh. Firstly, as we have

seen, there is a contrast in Welsh between nominal gaps and RPs. The latter behave like

ordinary pronouns, but the former do not. Hence the idea that nominal gaps are really

pronouns would involve the postulation of a special type of pronoun with surprising

properties. Secondly, gaps in Welsh are not always nominal, but may also be

prepositional. This is what we have in the wh-interrogatives in (61) corresponding to the

declaratives in (62).

(61) a. Am beth wyt ti ’n chwilio ___?

about what be.PRES.2SG you PROG search

‘What are you searching for?’

b. Ar beth wyt ti ’n dibynnu ___?

on what be.PRES.2SG you PROG depend

‘What are you depending on?’

(62) a. Dw i ’n chwilio am lyfr.

be.PRES.1SG I PROG search for book

‘I am searching for a book’

b. Dw i ’n dibynnu ar ffrind.

be.PRES.2SG I PROG depend on friend

‘I am depending on a friend’

18

There is no possibility of claiming that these gaps are phonologically empty pronouns.

Hence, it is fairly clear that Welsh UDs with a gap must be analysed in terms of

movement within a transformational approach.

One might accept that UDs with gaps are the result of movement and that there are

important similarities between such UDs and UDs with RPs but still reject the view that

the latter involve movement. One might propose that they are similar for some other

reason. This is essentially the position of Rouveret (2008). He proposes that what the two

types of example have in common is Agree. Gaps involve both Agree and movement and

RPs involve just Agree. If Agree operates in the same way in the two sorts of example, it

might be possible to claim that it is responsible for the similarities that they show.

Agree involves a probe, which is a head, and a goal, which the head c-commands. At

least normally, the probe has an uninterpretable feature (or a set of such features) and the

goal has a matching interpretable feature (or features). The goal also has some

uninterpretable feature, which renders it ‘active’, i.e. capable of undergoing Agree. Agree

deletes the uninterpretable features of probe and goal (often after they have triggered

some morphological effect). The most straightforward instance of Agree is responsible

for subject-verb agreement. Here the probe is T and the goal is the subject DP. T has

uninterpretable phi features and the DP has interpretable phi features. It also has an

uninterpretable Case feature rendering it active.

Focusing on relative clauses, Rouveret proposes that Agree links the C that heads the

relative with the RP via one or more intermediate positions. The simplest examples take

the following form, where v is the light verb which is a central feature of Minimalist

analyses of clause structure, ‘u-‘ marks an uninterpretable feature, and ‘i-‘ marks an

interpretable feature:

(63) [ C [ … v … pronoun … ]]

u-phi u-phi i-phi

i-Rel u-Rel u-Rel

The C has uninterpretable phi features and an interpretable Rel feature, the v has

uninterpretable phi features and an interpretable Rel feature, and the RP has interpretable

phi features and an uninterpretable Rel feature. Agree applies twice, first to the v and the

RP and then to the C and the v. Thus, the intermediate element v is first a probe and then

a goal. An example in which the RP is in a subordinate clause will take the following

form:

(64) [ C [ … v … [ C [ … v … pronoun … ]]]]

u-phi u-phi u-phi u-phi i-phi

i-Rel u-Rel u-Rel u-Rel u-Rel

Here we have three intermediate elements and Agree applies four times.

How satisfactory is this approach? One point to note is that there are a number of

respects in which the way Agree operates here differs from the way it operates in the case

of subject-verb agreement. A minor difference is that in the final instance of Agree, the

probe has not only an uninterpretable feature but also an interpretable feature. More

importantly, in all instances of Agree except the first, the goal has no interpretable

19

feature. Moreover, in a complex example there are instances of Agree where neither

probe nor goal has any interpretable features. A consequence of this is that

uninterpretable features on an intermediate element must not be deleted when it is the

probe in an instance of Agree since they must be available when it is the goal in the next

instance of Agree. It seems, then, that Agree is being ‘stretched’ quite a lot.

A second point to emphasize is that this is only an approach to relative clauses. What

about other UD constructions, e.g. wh-interrogatives? One might think that essentially the

same analysis but with a feature Q instead of the feature Rel could be proposed here. In

other words, one might propose analyses the following form:

(65) [ C [ … v … pronoun … ]]

u-phi u-phi i-phi

i-Q u-Q u-Q

But while this might seem plausible for wh-interrogatives with a resumptive pronoun, it

cannot be right for wh-interrogatives involving a gap if they are the result of movement.

On a movement analysis, the most deeply-embedded goal here is the wh-phrase, and this

must surely have an interpretable Q feature, as Chomsky (2000: 128) suggests. Thus, we

have the following situation:

(66) [ C [ … v … wh-phrase … ]]

i-Q

Where the wh-phrase is nominal, it will presumably have phi features, but, as emphasized

earlier, there are non-nominal wh-phrases. These presumably do not have phi features.

Hence, if the movement is preceded by Agree, it will have to involve quite different

features here from those proposed for relative clauses.

A further important problem arises from McCloskey’s (2002: 192) observation that

RPs universally look just like ordinary pronouns. As Asudeh (2004) points out, this casts

doubt on any analysis which treats RPs as special pronouns distinct in some way from

ordinary pronouns, and McCloskey (2006) argues that ‘there can be no syntactic feature

which distinguishes RPs from ‘ordinary’ pronouns’. An analysis in which RPs have an

uninterpretable Rel feature or an uninterpretable Q feature is a fairly clear example of an

analysis of this kind.

It is worth adding that Chomsky (2008) argues that A-movement is not associated

with Agree. He proposes that it is triggered by a so-called edge feature and is not

preceded by Agree. Obviously, if UDs with gaps do not involve Agree. there is no

possibility of claiming that Agree is what UDs with a gap and UDs with an RP have in

common.

Thus, it seems unlikely that the similarities between UDs with gaps and UDs with

RPs can be attributed to Agree. Therefore, from now on, I will assume with Willis (2011)

that a transformational approach that seeks to capture the similarities between UDs with a

gap and UDs with an RP needs to assume that both involve movement.

In pre-minimalist transformational work, movement leaves a phonologically empty

trace, and within Minimalism, it leaves a copy which is deleted in PF. Thus, within both

approaches, one expects there to be a gap when movement has applied, and there is a

20

prima facie problem if there is no gap to be seen. However, there are certain ways in

which one might try to address the problem.

Within Minimalism one might suggest that under certain circumstances the copy is

not deleted but is somehow converted into a pronoun. A version of this approach is

sketched in McCloskey (2006).12 He considers the following English example:

(67) I wonder which word they are not sure how it’s spelled.

Here the moved constituent is which word, which on standard Minimalist assumptions is

a DP headed by which. McCloskey suggests an analysis in which the complement word is

deleted and which is realized as it as a result of the deletion of its ‘WH-feature’.

McCloskey appears to assume that the head of a nominal constituent that has

undergone A-movement is distinguished from a pronoun by a single feature and that

when this is deleted a pronoun automatically results. A variety of issues arise here.

Firstly, it is not clear what a WH feature is. As Sag (2010: 491-3) notes, different UD

constructions involve different sets of wh-words, and many UD constructions do not

involve a visible wh-word. Hence, it is not clear that there is any independently motivated

feature that is characteristic of all the constituents that undergo A-movement. Secondly,

it is not clear what ensures that some feature is deleted. It is widely assumed that an

example like the following from German involves a copy which has not been deleted in

an intermediate position.

(68) Wen meinst du wen sie liebt? (Nunes 2001: 311)

whom believe you whom she loves

‘Who do you think that she loves?’

It is not clear why a copy could not remain unmodified in the original position of the

moved constituent. Thirdly, in the case of Welsh, it is fairly clear that more than just the

deletion of some feature is required. There is evidence from data like the following that

interrogative expressions, both simple and complex, are non-pronominal:

(69) a. i bwy b. *iddo bwy

to who to.3SGM who

‘to whom’

(70) a. i ba fachgen b. *iddo ba fachgen

to which boy to.3SGM which boy

‘to which boy’

These show that pwy ‘who’ and pa fachgen ‘which boy’ do not trigger agreement on a

preceding preposition in the way that a pronoun does. On the face of it, this means that

the head of the copy must be changed from [-PRO] to [+PRO] in PF.

12 A similar approach was advocated in Pesetsky (1998). McCloskey (pc) emphasizes that this is not an

approach he favours.

21

An obvious question to ask is whether there is any evidence that Welsh RPs are only

[+PRO] in PF. On the face of it, there could be. As we have emphasized, pronouns but

not non-pronominal NPs trigger agreement in Welsh. For Minimalism, agreement is a

product of Agree. Accepting this view of agreement, the position that Welsh RPs only

become [+PRO] in PF predicts that they should not trigger agreement in the same way as

ordinary pronouns. In fact, however, as we have seen, they trigger agreement just like

ordinary pronouns. One might conclude from this that agreement in Welsh applies not in

the syntax but in PF after the conversion of copies to pronouns.13 So the data may not be

a problem. However, it looks as if there is no reason to think that Welsh RPs are [-PRO]

at some point in the derivation. Hence, this feature of the proposed analysis seems quite

dubious.

Perhaps the most serious objection to this analysis is that like Rouveret’s it treats

RPs as special pronouns distinct from ordinary pronouns. An analysis in which RPs are

the result of the modification in PF of DPs which may be non-pronominal is a very clear

example of an analysis of this kind.14

Thus, it seems that an approach in which RPs are the realization of copies left by

movement faces a variety of problems. It looks, then, as if a transformational approach

needs to find some other way to combine movement with RPs. The obvious suggestion is

that the movement occurs from somewhere near the RP. Given the complexity of

structures assumed within transformational work, it is not hard to propose an analysis of

this kind, and a number of possibilities have been outlined in the literature.

As noted above, Willis (2011) concludes that there are important similarities between

UDs with gaps and UDs with RPs and hence that the latter must be able to involve

movement. He focuses on examples where the RP is object of a preposition such as the

following:

(71) Beth wyt ti ’n chwarae efo fo?

what be.PRES.2SG you PROG play with it

‘What are you playing with?’

He proposes that a PP containing an RP as its object may have a coindexed operator in its

specifier position, which undergoes A-movement. Within this approach, (71) has the

following analysis:

13 The idea that agreement applies in PF would be rather like the position proposed within HPSG in

Borsley (2009), in which it involves constraints on order domains. See section 4.4.

14 This point is made in Asudeh (2011). One might think that the analysis also violates the Inclusiveness

Condition, which, as formulated in Chomsky (2001, pp.2-3) ‘bars introduction of new elements (features)

in the course of a derivation’. However, it appears that this is only meant to apply to the syntax proper and

not to syntactic processes in PF.

22

(72) CP

DPi C'

beth wyt ti’n chwarae PP

DPi P

P DPi

beth

efo fo

Presumably, he would also propose that a DP containing an RP could have a coindexed

operator in its specifier position. On this analysis RPs are ordinary pronouns. Hence, it is

immune to what is arguably the most important objection to an analysis in which RPs are

the realization of copies left by movement. However, a question arises about the specifier

positions which it requires. There does not seem to be any independent motivation for

these positions. In English, what Culicover (1999) calls sluice-stranding, exemplified by

the following, seems to provide some support for a Spec PP position.

(73) a. Who with?

b. What about?

c. Who to?

Welsh does not have examples like this. It is also not clear what ensures that an

expression is only merged in these specifier positions if there is nearby pronoun that it is

coindexed with. At the very least, then, the analysis needs further work.

A different way of combining movement with an RP is proposed in Aoun, Choueiri

and Hornstein’s (2001) discussion of Lebanese Arabic. They suggest that a moved

constituent may originate as the specifier of an RP (2001: 285). Applying this approach to

Welsh, we have the following analysis for (71):

(74) CP

DP C'


P DP

DP D

efo beth fo

23

Whatever may the case in Lebanese Arabic, this analysis, like Willis’s, involves a

structure which is not independently motivated in Welsh. The following, in which an

ordinary DP is left adjoined to a pronoun, is unacceptable:

(75) *Wyt ti ’n chwarae efo ’r bêl hi?

be.PRES.2SG you PROG play with the ball she

‘Are you playing with the ball it?’

This analysis also makes RPs unlike ordinary pronouns. They may not be special in their

feature makeup but they are special in the context in which they appear. Thus, this

analysis faces the same objection as an analysis in which RPs are the realization of copies

left by movement.

Another way of combining movement with an RP is developed in Boeckx (2003).

Boeckx adopts the fairly common assumption that ordinary pronouns are intransitive

determiners. He then proposes that RPs are transitive determiners whose NP complement

has been fronted. In other words they are stranded determiners. On this analysis, (71) has

the analysis in (76).

(76) CP

NP C'


P DP

D NP

efo fo beth

This approach is problematic in a number of ways. Pronouns with a complement do not

occur elsewhere in Welsh. Hence, like the two preceding analyses, it involves a structure

which is not independently motivated. It also treats RPs as special pronouns since they

appear in structures in which ordinary pronouns do not appear. Moreover, given the

reasonable assumption that the complement-selection properties of lexical items are

reflected in their feature-makeup, they will also be special in this way. Hence, this

analysis too faces the same objection as an analysis in which RPs are the realization of

copies left by movement.

Further problems arise from the fact that a nominal filler is a DP when it is associated

with a gap but an NP when it is associated with an RP. The fillers look the same in the

two cases. In the case of wh-interrogatives, we have pwy ‘who’, beth ‘what’ or pa

‘which’ in both cases. Such elements are commonly analyzed as determiners. Boeckx

must assume either that they are sometimes determiners and sometimes elements within

NP or that they are always elements within NP. Neither position seems very attractive.

24

An example like (53), repeated here for convenience as (77), seems particularly

problematic for this approach.

(77) y dyn welais i ___ a soniais amdano fo]

the man see.PAST.1SG I and talk.PAST.1SG about.3SGM he

‘the man that I saw and talked about’

On Boeckx’s analysis there is a gap in both conjuncts but the gaps are of different

categories, a DP in the first conjunct and an NP in the second conjunct. It is not really

clear how an example with a gap of the same kind in the two conjuncts, such as (52),

repeated here as (78), should be handled within a transformational approach.

(78) y dyn [welais i ___ a gwelaist tithau ___ hefyd]

the man see.PAST.1SG I and see.PAST.2SG you too

‘the man that I saw and you saw too’

However, an example like (77) seems even more problematic given the proposed

analysis.15

It seems that it is very difficult to assimilate RPs to gaps within a movement-based

approach. The approaches of McCloskey, Aoun, Choueiri and Hornstein and Boeckx face

serious problems. Willis’s approach seems less problematic, but at least it needs more

work. I will show in the next section that it is quite easy to assimilate RPs to gaps within

HPSG.

4. An HPSG analysis

We can turn now to HPSG. As I have emphasized, the similarities between gaps and RPs

in Welsh suggest that they involve the same mechanism. There are various possible

approaches to RPs within HPSG. For example, in his work on Hebrew and Irish, Vaillette

(2000, 2002) argues that RPs in these languages should be analysed as the realization of a

special feature, which he calls RESUMP. However, given the similarities between gaps

and RPs, the latter should involve the same mechanism as the former. This is the SLASH

feature.

I will begin by outlining the standard HPSG approach to UDs. I will then consider the

nature of gaps. Then, I will outline an HPSG analysis in which RPs are a realization of

the SLASH feature. Finally, I will discuss some further constraints.

15 Rouveret (1994: chapter 5) proposes analyses in which there is movement from prepositional object

positions associated with agreement and possessor positions. However, his concern is with Literary Welsh,

in which overt RPs are unacceptable in these positions (as noted in fn. 5). Hence he is not concerned with

the problem that I am focusing on here.

25

4.1. The SLASH approach to UDs

HPSG is a monostratal theory, in which the syntactic structure of a sentence is a single

relatively simple constituent structure. Hence, there are no movement processes as in the

various forms of transformational grammar, and what are seen as moved constituents in

transformational work only ever occupy their superficial position. HPSG is also a

constraint-based theory, in which a grammar consists of a set of word and phrase types,

and a set of constraints to which they are subject. The constraints are implicational

statements, saying that if a linguistic object has some property or properties then it must

have some other property or properties.

As indicated above, the HPSG approach to UDs involves the feature SLASH. This

makes information about gaps available higher in the structure. However, as we will see,

there is no necessary connection between SLASH and gaps. There are two different

views of gaps in the HPSG literature. In some HPSG work, e.g. Pollard and Sag (1994),

Levine and Hukari (2007), they are analyzed as empty categories. In other work, e.g.

Bouma, Malouf and Sag (2001), Ginzburg and Sag (2000), they are members of the

ARG-ST (ARGUMENT-STRUCTURE) list of some head with no counterpart in

constituent structure. We will consider which of the two approaches is appropriate for

Welsh in 4.2. In both approaches, gaps have the following feature makeup:

(79)

{[1]} SLASH

[1] LOCAL

The LOCAL feature encodes most but not all of the syntactic and semantic properties of

an expression. As (79) makes clear, the SLASH feature is not part of the value of

LOCAL. Nor is the WH feature used in the analysis of wh-interrogatives. In (79) the

value of LOCAL is also the single member of the set which is the value of SLASH.

The standard approach to the distribution of SLASH is ‘head-driven’. It involves the

following assumptions:

(80) a. The SLASH value of a head is normally the same as that of its arguments.

b. The SLASH value of phrase is normally the same as that of its head.

The latter is ensured either by the SLASH Inheritance Principle (Bouma, Malouf and Sag

2001) or the Generalized Head Feature Principle (Ginzburg and Sag 2000).16 The former

is ensured by the SLASH Amalgamation Principle, which we can formulate as follows,

following Ginzburg and Sag (2000: 199):

(81) SLASH Amalgamation Principle

16 The former requires a phrase and its head to have the same value for SLASH. The latter is much more

general and requires a phrase and its head to have the same syntactic and semantic properties. However, it

is a default constraint which can be overridden by other constraints.

26

word /

{[n]}] [SLASH..., ],{[1]} SLASH[ ST-ARG

[n]} ... {[1] SLASH

This relates the SLASH value of a word to the SLASH values of the members of its

ARG-ST list. Essentially the Principle says that by default any local feature structure in

the SLASH value of a word must appear in the SLASH value of an associated argument

or adjunct and any local feature structure in the SLASH value of an argument or adjunct

must appear in the SLASH value of the associated word. The SLASH Amalgamation

Principle is a default constraint to accommodate examples like the following:

(82) This book is easy [for anyone to read]

Here, the adjective easy takes an infinitival complement with a non-empty SLASH

feature but should not have a non-empty SLASH feature itself since the complement is

the top of the dependency. The same situation arises with the Welsh example in (1e).

Most arguments are [SLASH {}]. Hence a UD typically involves structures of the

following form:

(83) [SLASH {[1]}]

HD-DTR

... {[1]}], [SLASH ... ST-ARG

{[1]} SLASH

...

...

For HPSG, all aspects of linguistic expressions including their internal structure are

analyzed in terms of features. However, it is convenient to use the traditional tree format

to represent constituent structure.

A somewhat different account is necessary for coordinate structures in the middle of a

UD since they are assumed to be headless. However, I will not consider what form this

should take.

At the top of the dependency, there are a number of possibilities, as noted in the

discussion after (1) in the Introduction. Where there is a filler, we have a head-filler-

phrase subject to the following constraint.

(84)

head-fill-ph

]2[ DTR-HD

{[1]}] [2][SLASH ],[1] LOC[ DTRS

{} SLASH

This requires a head-filler phrase to be SLASH {} and to have a head daughter and a non-

head daughter whose LOCAL value is the local feature structure within the value of

SLASH on the head daughter. The constraint licenses structures of the following form:

27

(85) S

[SLASH {}]

HD-DTR

[1]XP S

[SLASH {[1]}]

I will not discuss the other possibilities that arise at the top of the dependency.

We can now illustrate this approach with the following simple English wh-

interrogative:

(86) Who did Kim talk to?

Ignoring the ARG-ST feature and the HD-DTR annotations and assuming that gaps are

empty categories, this will have a representation of the following form:

(87) S

[SLASH {}]

[1]NP S

[SLASH {[1]}]

V NP VP

[SLASH {[1]}] [SLASH {[1]}]

V PP

[SLASH {[1]}] [SLASH {[1]}]

P [1]NP

[SLASH {[1]}] [SLASH {[1]}]

Who did Kim talk to e

This has three instances of a head with the same value for SLASH as an argument: the

two Vs and the P. This highlights the head-driven nature of the HPSG approach.

There are two properties of the SLASH analysis that are crucial in the present context.

Firstly, as emphasized at the outset, there is no necessary connection with gaps. This

means that there is no difficulty in principle in extending the analysis to RPs. Secondly,

the analysis is head-driven. It follows from this that it is not difficult to capture the fact

that gaps appear with some heads and RPs with others.

28

4.2. Nature of gaps

Before I consider how RPs should be accommodated I want to consider which of the two

approaches to gaps is appropriate for Welsh. There is in fact quite strong evidence in

Welsh that they should be analyzed as empty categories.

One type of evidence comes from mutation, the systems of word-initial consonant

alternations which are an important feature of Welsh and the other Celtic languages.

Particularly notable is the mutation of objects of finite verbs, as e.g. in (88), in which the

mutated object is in bold and the basic form in brackets:

.

(88) Prynodd Emrys geffyl. (ceffyl)

buy.PAST.3SG Emrys horse

‘Emrys bought a horse.’

A number of researchers (Harlow 1989, Borsley and Tallerman 1996, Borsley 1999,

Tallerman 2006, 2009) have argued that this is one instance of mutation triggered by an

immediately preceding phrase. The following illustrate other instances of this kind of

mutation:

(89) Ddechreuodd Emrys [ddarllen y llyfr] (darllen)

begin.PAST.3SG Emrys read the book

‘Emrys began to read the book.’

(90) a. Mae chwant mynd adref arna’ i.

be.PRES.3SG desire go home on.1SG I

‘I desire to go home.’

b. Mae chwant arna’ i fynd adref. (mynd)

be.PRES.3SG desire on.1SG I go home

‘I desire to go home.’

(91) a. Mae Emrys wedi rhoi darlun o Gwyn i Megan.

be.PRES.3SG Emrys PERF give picture of Gwyn to Megan

‘Emrys has given to Megan a picture of Gwyn.’

b. Mae Emrys wedi rhoi i Megan ddarlun o Gwyn. (darlun)

be.PRES.3SG Emrys PERF give to Megan picture of Gwyn

‘Emrys has given to Megan a picture of Gwyn.’

(92) a. Mae ci yn yr ardd.

be.PRES.3SG dog in the garden

‘A dog is in the garden.’

b. Mae yn yr ardd gi. (ci)

be.PRES.3SG in the garden dog

‘A dog is in the garden.’

In (89) a non-finite complement shows mutation. (90a) has a subject containing an

abstract noun and an infinitival complement, while in (90b) the complement is extraposed

29

and it is mutated. (91a) contains an NP and a PP complement in that order, while (91b) is

essentially an example of heavy-NP-shift with the NP following the PP. In the former the

NP is unmutated, while in the latter it is mutated. Finally, (92a) shows the normal

subject-complement order, and (92b) shows a marked complement-subject order. In the

former the subject is unmutated, but in the latter it is mutated. In the present context, the

crucial point is that objects of finite verbs are mutated when there is a gap in subject

position. The following illustrates:

(93) Pwy brynodd ___ geffyl? (ceffyl)

who buy.PAST.3SG horse

‘Who bought a horse?’

If the mutation of objects of finite verbs is triggered by an immediately preceding phrase,

then the object here must be immediately preceded by a phrase just like the object in (88).

Thus, examples like (93) provide quite strong evidence that gaps should be analyzed as

empty categories.

Further evidence that gaps are empty categories comes from agreement. Borsley

(2009) argues that agreement in Welsh involves a head and an immediately following

pronoun. This may be a finite verb agreeing with a subject, a preposition agreeing with an

object, a non-finite verb agreeing with an object, or a noun agreeing with a possessor. In

the first two cases agreement takes the form of a suffix, while in the other two it takes the

form of a clitic. Here are some relevant examples:

(94) a. Gwelon nhw ddraig.

see.PAST.3PL they dragon

‘They saw a dragon.’

b. arnyn nhw

on.3PL they

‘on them’

c. Gwnaeth Emrys eu gweld nhw.

do.PAST.3SG Emrys 3PL see they

‘Emrys saw them.’

d. eu tad nhw

3PL father they

‘their father’

Consider now the following:

(95) Pwy welodd ___ ti?

who see.PAST.3SG you.SG

‘Who saw you?’

Here, the filler pwy is associated with a subject gap. If agreement is with an immediately

following pronoun, there must be an empty category in subject position since otherwise

the verb would be immediately followed by the object ti and would agree with it, giving

(96).

30

(96) Pwy welest ti ___?

who see. PAST.2SG you.SG

‘Who did you see?’

This is perfectly grammatical, but it can only have an object gap, as the translation makes

clear.

It is fairly clear, then, that UD gaps should be analyzed as empty categories. I will

assume following Bouma, Malouf and Sag (2001) that they are the realization of special

gap-synsem objects. I will also assume that they are required to be phonologically empty

and that nominal gaps are required to be non-pronominal. The following constraints will

do this:

(97) a. [gap] [PHON <>]

b.

noun

gap

HEAD [CONTENT npro]

If nominal gaps are non-pronominal, they will be associated with a non-pronominal

SLASH value. This suggests that a gap will never be associated with a pronominal filler.

A cleft sentence such as the following looks problematic here:

(98) Nhw welodd ___ ddraig.

they see.PAST.3SG dragon

‘It was they that saw a dragon.’

However, Borsley (2008, 2011) argues that the focused constituent in a cleft sentence is

not a filler, partly on the basis of examples like (98) and partly on the basis of examples like

the following:

(99) Fi welodd ___ ddraig.

I see.PAST.3SG dragon

‘It was I that saw a dragon.’

Here the focused constituent is a first person pronoun but the gap is clearly not first person. I

think, then, that the fact that nominal gaps are associated with a non-pronominal SLASH

value is unproblematic.

It seems, then, that there are good reasons for analysing Welsh UD gaps as empty

categories. However, the analysis that I will develop below will not depend in any major

way on this assumption. It would involve the same constraints if the assumption was

abandoned.

31

4.3. The analysis

As emphasized above, there is no necessary connection between the SLASH feature and

gaps. Hence, it is not too difficult to extend the SLASH analysis to RPs. Essentially, we

need to allow gaps under some circumstances and RPs under others. To do this, we need

to replace the SLASH Amalgamation Principle, which is designed to handle a language

with just gaps, with something more complex. It is not too difficult to do this.

As noted at the end of 2.1, Welsh UDs have gaps as arguments of verbs and adjectives

and in adverbial positions and RPs as arguments of prepositions and nouns. I will assume

with Ginzburg and Sag (2000: 168, fn.2) that adverbial elements are optional extra

complements of verbs. Hence adverbial gaps are gaps that are arguments of verbs. I will

also assume with Kluender (1998), Levine and Hukari (2006), and Hofmeister and Sag

(2010) that island phenomena are a processing matter and hence that, as far as the

grammar is concerned, a slashed head can always be associated with a constituent

containing a gap or RP. Thus, I assume that verbs or adjectives with a non-empty SLASH

value have an argument which is either a gap or a constituent containing a gap or RP,

while prepositions and nouns have an argument which is either an RP or a constituent

containing a gap or RP.

Within this approach, we will have structures of the following form with verbs and

adjectives:

(100)

{[2]} SLASH

[1] HEAD adj verb

HD-DTR

...[3]... ST-ARG

{[2]} SLASH

[1] HEAD

...

{[2]}] SLASH][3[

...

Here, the slashed argument is either a gap or constituent containing a gap or an RP and

given the assumption that adverbial elements are optional extra complements of verbs,

the argument may be an adverbial. Essentially, we have the situation that the SLASH

Amalgamation Principle is designed to handle. We have a single type of structure here

because gaps and constituents containing a gap or an RP both have a non-empty SLASH

value and thus form a natural class.

With prepositions and nouns we will have two rather different types of structure.

Firstly, we will have structures of the following form:

32

(101)

{[2]} SLASH

[1] HEAD noun prep

HD-DTR

...[3]... ST-ARG

{[2]} SLASH

[1] HEAD

...

{[2]} SLASH]3[

canon

...

Here the slashed argument is identified as canon(ical), i.e. as one that is not a gap. This

means that it can only be a constituent containing a gap or an RP. Secondly we will have

structures of the following form:

(102)

[3]]} [INDEX:{[2]NP SLASH

[1] HEAD noun prep

HD-DTR

...[4]... ST-ARG

{[2]} SLASH

[1] HEAD

...

[3]] [INDEX:[4]NP ppro

...

Here we have not a slashed argument but an argument which is a pronoun coindexed with

the value of SLASH. This will only be possible where the value of SLASH is nominal.

We have two different structures with prepositions and nouns because RPs and

constituents containing a gap or an RP are not a natural class. Thus, the situation is more

complex with prepositions and nouns than with verbs and adjectives. I will argue below

that this is not a weakness of the analysis.

Before I formulate appropriate constraints, there is one further empirical point to note.

This is that it seems that Welsh does not have parasitic gaps.17 One might suppose that

there is a parasitic gap after the verb ddarllen in the following example:

(103) Dyna ’r adroddiad dw i wedi ei daflu ___ i ffwrdd

there-is the report be.PRES.1SG I PERF 3SGM throw away

[heb ei ddarllen ___].

without 3SGM read

‘There is the report that I throw away without reading.’

It is clear, however, that this is not a true gap but a phonologically null pronoun. It is

possible to have an overt pronoun, as the following shows:

17 I am grateful to David Willis for help with the data here.

33

(104) Dyna ’r adroddiad dw i wedi ei daflu ___ i ffwrdd


[heb ei ddarllen o].

without 3SGM read he

‘There’s the report which I threw away without reading.’

Now consider the following:

(105) *Dyna ’r adroddiad dw i wedi ei daflu ___ i ffwrdd


[heb ddarllen ___].

without read

Here, ddarllen has no clitic. A phonologically null pronoun is only possible when

agreement of some kind is present. Thus, the object here can only be a gap, and not a null

pronoun. However, this example is ungrammatical. This suggests rather strongly that

Welsh does not have parasitic gaps.

The absence of parasitic gaps means that only a single member of any ARG-ST list

may be a gap or RP or contain a gap or RP. Thus, we have the following situations when

a head has a non-empty SLASH value, as follows:

(106) a. If a verb or adjective has a non-empty SLASH value then one of its

argument has the same non-empty SLASH value.

b. If a preposition or noun has a non-empty SLASH value then one of its

arguments is canonical argument with the same non-empty SLASH value

or a pronoun coindexed with the SLASH value.

These are essentially informal constraints. We can formalize them as follows:

(107)

{[1]} SLASH

HEAD adj verb [ARG-ST L2 <[SLASH {[1]}]> L3]

Li = list([SLASH {}])

(108)

[2]]} {[1][INDEX SLASH

HEAD prepnoun [ARG-ST L3 < NP:ppro[INDEX [2]]

{[1]} SLASH

canon> L4]

Li = list([SLASH {}])

34

(108) involves a disjunction and is thus more complex than (107). This reflects the fact

that there are two types of structure with prepositions and nouns but just one with verbs

and adjectives. In both constraints, the stipulation about Li ensures that parasitic gaps do

not occur.

A full analysis must also ensure that a head with a slashed argument is itself slashed

in normal circumstances. The following constraint does this:

(109) [ARG-ST L1 <[SLASH {[1]}]> L2] / [SLASH {[1]}]

This is a default constraint as indicated by ‘/’. The constraint will be overridden in the

Welsh counterpart of an English ‘tough’ sentence, where an adjective takes an infinitival

complement with a non-empty SLASH feature. A relevant example in (1e), repeated here

as (110):

(110) Mae Carys yn hawdd [i Ifor ei gweld ___].

be.PRES.3SG Carys PRED easy to Ifor 3SGF see

‘Carys is easy for Ifor to see.’

The adjective hawdd takes an infinitival complement with a non-empty SLASH value,

but hawdd itself should not have a non-empty SLASH value since the complement is the

top of the dependency. It is not necessary to ensure that any head that has an argument

which is a pronoun has a coindexed SLASH value since obviously not all pronouns are

RPs.

The constraints in (107), (108) and (109) will together replace the SLASH

Amalgamation Principle. They are more complex than the SLASH Amalgamation

Principle, but that is because the data is more complex than that with which SLASH

Amalgamation Principle is concerned.

The analysis gives appropriate structures for the full range of examples discussed in

2.1. For example, the clause following the wh-phrase in (1a) and (7a) will have the

structure in (111) with a gap in object position.

(111)

{[2]NP} SLASH

[1] HEAD verb

HD-DTR

[4] [3], ST-ARG

{[2]} SLASH

[1] HEAD

NP]3[

{[2]} SLASH

[2] LOCAL[4]

welest ti e

The PP in (30) will have the structure in (112) with an RP as object.

35

(112)

[3]]} [INDEX:{[2]NP SLASH

[1] HEAD prep

HD-DTR

[4] ST-ARG

{[2]} SLASH

[1] HEAD

[3]] [INDEX:[4]NP ppro

iddo fo

Since both gaps and RPs are associated with the SLASH feature within this analysis,

the similarities between gaps and RPs documented in 2.2 are unsurprising. Given the

standard HPSG approach to UDs outlined above, any verb in the path of a UD will have a

non-empty SLASH value, whether it involves a gap or an RP. We can assume that the

ban on present tense forms of bod is nullified whenever it is [SLASH {NP}]. This will

allow both (44) and (45) with a gap and (47) and (48) with an RP. In much the same way

we can assume that mutation and a third person singular masculine clitic may appear

when a non-finite verb is [SLASH {NP}]. This accounts both for (49) with a gap and (50)

with an RP. Finally, both conjuncts with a gap and conjuncts with a RP are [SLASH

{NP}]. Hence, the coordinate structures in (53) and (54) will be just like that in (52). If

we adopted Vaillette’s (2000, 2002) view that RPs are the realization of a separate

feature, which he calls RESUMP, the phenomena discussed in section 3 would be

surprising and would require more complex analyses referring to both SLASH and

RESUMP.

I emphasized above that we have a more complex situation with nouns and

prepositions, which allow RPs as an argument, than with verbs and adjectives, which

allow gaps. Is the weakness of the proposed analysis? I think not. It means that RPs

complicate the grammar, but this accounts for the fact that they only occur in positions

which are relatively inaccessible. In these positions, they facilitate sentence

comprehension (Hofmeister and Norcliffe 2012) and hence the complexity is justified. In

positions which are more accessible, there is no reason to employ RPs and so they are

avoided.

A further point that is worth making is that it would not be difficult to modify the

analysis to accommodate more literary or more colloquial varieties. We noted in 2.1 that

embedded subject and object RPs are possible in the literary language. To allow this we

would need to restrict the constraint in (102) to root verbs and to propose a second

constraint on non-root verbs allowing its argument to be an RP. We also noted in section

2.1 that very colloquial registers allow a gap in prepositional object position. To allow

this we would need to restrict the constraint in (103) to nouns and to propose a separate

constraint on prepositions allowing a gap instead of or in addition to an RP. A variety of

Welsh which allows both a gap and an RP in some position is also no problem for this

36

approach. It will, however, be more complex than the kind of Colloquial Welsh that is the

focus of this paper since it will have not just two sorts of argument positions but three:

one allowing gaps, one allowing RPs, and one allowing both. Hence, it is not surprising if

gaps and RPs appear in disjoint sets of environments for many speakers.

To conclude this section I want to briefly consider an alternative analysis in which

RPs are slashed elements, presumably of the following form:

(113)

[1]]} [INDEX:{NP SLASH

[1]] [INDEX:NP LOCAL ppro

Such an analysis violates McCloskey’s principle that ‘there can be no syntactic feature

which distinguishes RPs from ‘ordinary’ pronouns’. Nevertheless it is of some interest to

compare it with the analysis developed above.

Within such an analysis the constraint on verbs and adjectives would be more

complex. It would be necessary to stipulate that the slashed constituent cannot be a

pronoun. On the other hand, the constraint on prepositions and nouns would be simpler.

One could stipulate that they must have a slashed argument which is a canonical

expression. This would exclude gaps but allow RPs and constituents containing a gap or

RP. It looks, then, as if an analysis that treats RPs as slashed elements would be of

roughly equal complexity to the analysis developed above. However, it has the major

disadvantage that it violates McCloskey’s principle. It also has the weakness that it treats

RPs as no more complex than gaps. Hence, it is not clear within this analysis why RPs

should be confined to relatively inaccessible positions. Thus, we have two important

objections to such an analysis.

4.4. Some further constraints

We now have a fairly full account of Welsh UDs. However, there are certain other

matters that we need to deal with. We noted in 2.1 that non-pronominal NPs do not

trigger agreement in the way that pronouns do, and that nominal gaps generally behave

like non-pronominal NPs, and in 5.2 we proposed a constraint that ensures that nominal

gaps are non-pronominal. However, we saw in section 2.1 that when a non-finite verb has

a gap as its object, it may have a preceding clitic, which agrees with the gap (and hence

with the filler if there is one) or is third person singular masculine. We also saw in section

2.3 that a non-finite verb may bear a third person singular masculine clitic when it has a

complement which contains a gap or an RP. In this section I will consider how these facts

should be accommodated.

Borsley (2009) proposes that Welsh agreement is the product of constraints on order

domains, which are the value of the feature DOM and are also the locus of constraints on

word order. He proposes that Welsh agreement, whether in the form of an inflection or a

clitic, is the realization of an AGR feature, whose value is an index (with the features

PERS, NUMB and GEND) or none when there is no agreement, and formulates the

following constraints:

37

(114) [DOM <[AGR [1]], …>] / [1] = none

(115)

>… ,

[2] INDEX CONT

NP CAT

LOC [1]], [AGR< DOM ppro , …>] [1] = [2]

The constraint in (114) is a default constraint and accounts for all the situations in which

agreeing words do not show agreement.18 It is overridden by (115), which ensures that an

agreeing element shows agreement with an immediately following pronoun.19 This, of

course, includes an RP. There is no reason why (114) should not be overridden by other

constraints.

For non-finite verbs with a gap as their object we can propose the following

constraint:

(116)

... ,

[2]] [INDEX CONT

NP CAT LOC

,[1]] AGR V[ DOM

gap

inf,

[1] = [[2] [3rd, sing, masc] none]

This allows three different values for the AGR feature on the non-finite verb. The first is

that it agrees with the gap. The second is that it is third singular masculine. I assume that

this gives rise both to cases where there is a third singular masculine clitic and to

examples where there is just mutation. The third is that it is none. This gives rise to

examples with neither a clitic nor mutation. The last possibility is quite colloquial, and

we might exclude it if we are concentrating on a more neutral variety.

For non-finite verbs with a complement which contains a gap or an RP we can

propose the following:

18 This will include including finite verbs followed by a subject gap if they are seen as forms with no

agreement features. If they are seen as really third person singular forms, they will require a special

constraint.

19 The agreement marking does not immediately precede the pronoun when it takes the form of a clitic.

However, on the analysis developed in Borsley (2009), the agreeing element is the non-finite verb or noun

to which the clitic is attached and this immediately precede the pronoun.

38

(117)

... ,

{[]} SLASH ,[1]] AGR ,V[ DOM

canoninf [1] = [[3rd, sing, masc]

none]

This allows just two values for the AGR feature on the non-finite verb. It is either third

singular masculine or it is none. In this case examples with neither a clitic nor mutation

are not particularly colloquial. So there is no reason here to exclude the value none.

There is at least one other matter that an analysis of Welsh UDs needs to deal with,

namely the way they nullify the ban on the present and imperfect of bod in affirmative

complement clauses. I will not attempt to provide a formal account here. However, it

should be clear that there is no problem. We just need to ensure that present and

imperfect forms of bod are realized by the non-finite form bod under certain conditions

and one of the conditions is when they are [SLASH {}].

5. Concluding remarks

In this paper I have investigated Welsh UDs and especially the behaviour of gaps and

RPs. I have shown that they differ in their distribution but that otherwise they are quite

similar. Within a transformational approach this suggests that both involve movement.

Within HPSG it suggests that both should be analyzed as realizations of the SLASH

feature. I have outlined an analysis in which the SLASH Amalgamation Principle

proposed for various languages which do not have RPs is replaced by three constraints,

one of which allows an RP to appear as the sister of a preposition or noun with a non-

empty SLASH value. Within this analysis RPs are ordinary pronouns and hence it is not

surprising that they look like ordinary pronouns.

It is perhaps worth emphasizing that I do not assume that the approach to RPs that I

have developed here is appropriate for all languages. It may well be that they are

sufficiently different from gaps in some languages that they should involve a different

mechanism, either a different NONLOCAL feature such as Vaillette’s (2000, 2002)

RESUMP or something else. However, as we have seen, the fact that gaps and RPs

appear in different local contexts does not entail that they involve different mechanisms.

Nor does the fact that RPs are more acceptable than gaps within islands given the view

that islands are a processing matter. It may well be, then, that a SLASH analysis of RPs is

appropriate in many languages.

There is naturally more to be said about Welsh UDs. There is some more data that

needs to be considered. Borsley, Tallerman and Willis (2007: 146) note that an RP

appears in subject position in non-finite subordinate clauses introduced by what looks like

the preposition i ‘to’, ‘for’, which Borsley (1986) analyzes as a prepositional

complementizer. The following illustrates:

(118) y bobl hoffen ni iddyn nhw ddod

the people like.COND.1PL we to.3PL they come

39

‘the people that we’d like to come’

This is easy to accommodate if we assume with Borsley (1999) that i takes the following

subject and predicate as two separate complements and has the following category:

(119)

VP NP, COMPS

SUBJ

HEAD comp-prep

Assuming this category, we will predict the RP in (118) if we add prep-comp to the

possible values of HEAD in (108).

It seems that there is one more position in which some speakers allow an RP.

According to Tallerman (1983), some speakers accept examples in which an RP is one

conjunct of a coordinate structure. She gives the following examples:

(120) Dyma ’r dyn y siaradais i gydag o a Siôn y bore

here-is the man PRT speak.PAST.1SG I with he and Siôn the morning

’ma.

here

‘Here’s the man that I spoke to him and Siôn this morning.’

(121) Dyma ’r dyn y gwelais i Mair ac o ddoe.

here-is the man PRT see.PAST.1SG I Mair and he yesterday

‘Here’s the man that I saw Mair and him yesterday.’

It is clear that not all speakers accept such examples, but for those that do they need to be

accommodated. This should not be too difficult. Assuming that RPs are ordinary

pronouns, all we need to do is to license coordinate structures where the mother is

[SLASH {NP}] and where one of the conjuncts is a pronoun coindexed with the SLASH

value. This will require a special subtype of coordinate structure, but it doesn’t seem to

pose any particular problems.

As an anonymous referee has emphasized, there are also questions about the

interpretive properties of the two types of UDs. In particular, it is important to consider

how far they show reconstruction, how far, that is, a filler behaves as if it were in the

position of the gap or RP. This is a major concern of Rouveret (2008). Rouveret claims

inter alia that UDs with a gap show reconstruction with respect to condition C but that

UDs with an RP do not. He considers inter alia the following examples (where I use his

glosses):

(122) Yn ddiweddar dygwyd [darlun o Siôn [a roddasai i Mair]]

recently was-stolen picture of Siôn Rel had-given to Mair

‘Recently was stolen a picture of Siôn which he had given to Mair.’

40

(123) Yn ddiweddar dygwyd [darlun o Siôn [yr oedd ef wedi

recently was-stolen picture of Siôn that was he Perf

ei roddi i Mair]]

it give to Mair

‘Recently was stolen a picture of Siôn which he had given to Mair.’

He claims that Sion and the unexpressed subject of roddasai cannot be coreferential in

(122) but that Sion and ef can be corereferential in (123). In other words, he suggests that

the antecedent darlun o Siôn behaves as if it were in the position relativized in (122) but

not in (123). In (122) the position relativized is object of a finite verb, whereas in (123) it

is object of a non-finite verb. Rouveret assumes that the former involves a gap and the

latter an RP. I argued earlier with Willis (2000) that both positions involve a gap in

Colloquial Welsh. On this analysis, one might expect the same possibilities for

coreference in the two examples. Interestingly, speakers that I have consulted feel that

coreference is possible in both. However, it is quite difficult to get clear judgments about

such matters. This suggests that this is an area where the traditional methodology of

informal investigations of the judgments of a few speakers is insufficient and should be

supplemented by some carefully designed experimental work with larger numbers of

speakers. I leave this to future research.

Acknowledgements

This paper is based on talks given at the Fifteenth Welsh Syntax Seminar at Gregynog,

Wales, July 7-8, 2008, at the Sixteenth Welsh Syntax Seminar at Gregynog, Wales, July

6-7, 2009, at the 17th International Conference on HPSG10, in Paris, France, July 7-10,

2010, and at the Celtic Workshop, University of Berkeley, September 7-8, 2012. I have

benefited from the comments of audience members on all four occasions and especially

from those of Danièle Godard and Bob Levine. I have also benefited from the comments

of three referees, and I am also grateful to Bob Morris Jones, David Willis and Emyr

Davies for help with the data. Some of the ideas in the paper were previously presented in

Borsley (2010). I alone am responsible for what appears here.

REFERENCES

Alexopoulou, T., Keller, F., 2007. Locality, cyclicity and resumption: at the interface

between grammar and the human parser. Language 83, 110-160.

Aoun, J., L., Choueiri, L., Hornstein, N., 2001. Resumption, movement, and derivational

economy. Linguistic Inquiry 32, 371-403.

Asudeh, A. 2004. Resumption as Resource Management. Ph.D. dissertation, Stanford

University.

Asudeh, A. 2011. Towards a unified theory of resumption. In: A. Rouveret, A. (Ed.),

Resumptive Pronouns at the Interfaces. Amsterdam: John Benjamins, pp. 121-187.

Awbery, G., 1977. A transformational view of Welsh relative clauses. Bulletin of the

Board of Celtic Studies 27. 155–206.

41

Boeckx, C., 2003. Islands and Chains. John Benjamins, Amsterdam.

Borsley, R. D., 1986. Prepositional complementizers in Welsh. Journal of Linguistics 20,

277-302.

Borsley, R. D., 1993. On So-called Verb-Nouns in Welsh. Journal of Celtic Linguistics 2:

35-64.

Borsley, R. D., 1996. On a nominal analysis of Welsh verb-nouns. In: Ahlqvist, A.,

Capková, V. (Eds.), Dán do Oide: Essays in Memory of Conn R. Ó Cléirigh,

Institiúid Teangeolaíochta Éireann, Dublin, pp. 39-47.

Borsley, R. D., 1999. Mutation and constituent structure in Welsh. Lingua 109, 263-300.

Borsley, R. D., 2008. On some Welsh unbounded dependency constructions. Essex

Research Reports in Linguistics 57.4, 1-21

Borsley, R. D., 2009. On the superficiality of Welsh agreement. Natural Language and

Linguistic Theory 27, 225-265.

Borsley, R. D. 2010, An HPSG approach to Welsh unbounded dependencies, in S. Müller

(ed.), Proceedings of the HPSG10 Conference, Stanford: CSLI Publications, 47-67.

Borsley, R. D., Tallerman, M.O., 1996. Phrases and soft mutation in Welsh. Journal of

Celtic Linguistics 5, 1-33.

Borsley, R. D., Tallerman, M.O., Willis, D., 2007. The Syntax of Welsh. Cambridge

University Press, Cambridge.

Bouma, G., Malouf, R., Sag, I. A., 2001. Satisfying constraints on extraction and

adjunction. Natural Language and Linguistic Theory 19, 1-65.

Chomsky, N. 2000. Minimalist inquiries: the framework. In: Martin, R., Michaels, D.,

Uriagereka, J. (Eds.), Step by Step: Essays on Minimalist Syntax in Honor of

Howard Lasnik, MIT Press, Cambridge, Mass, pp. 89-155.

Chomsky, N. 2008. On phases. In: Freidin, R., Otero, C. P., Zubizaretta, M.-L. (Eds),

Foundational Issues in Linguistic Theory. MIT Press, Cambridge, Mass, pp. 133-

166.

Culicover, P. W. 1999. Syntactic Nuts: Hard Cases, Syntactic Theory and Language

Acquisition. Oxford University Press, Oxford.

Erteschik-Shir, N., 1992, Resumptive pronouns in islands. In: Goodluck, H., Rochemont,

M., (Eds.) Island Constraints: Theory, Acquisition and Processing. Kluwer,

Dordrecht, pp. 89–108

Ginzburg, J., Sag, I. A., 2000. Interrogative Investigations: The Form, Meaning and Use

of English Interrogatives. CSLI Publications, Stanford.

Goldsmith, J., 1985. A principled exception to the coordinate structure constraint. CLS

21, 133-143.

Harlow, S. J., 1983. Celtic relatives. York Papers in Linguistics 10, 77–121.

Harlow, S. J., 1989. Syntax of Welsh soft mutation. Natural Language and Linguistic

Theory 7, 289-316.

Hendrick, R., 1988. Anaphora in Celtic and Universal Grammar. Kluwer, Dordrecht.

Hofmeister, P., Norcliffe, E., 2013. Does resumption facilitate sentence comprehension?

In: Hofmeister, P., Norcliffe, E. (Eds), The Core and the Periphery: Data-driven

Perspectives on Syntax Inspired by Ivan A. Sag. CSLI Publications, Stanford.

Hofmeister, P., Sag, I. A., 2010. Cognitive constraints and island effects. Language 86,

366-415.

42

Hofmeister, P., Staum Casasanto, L., Sag, I. A., in press. Islands in the grammar?

Standards of evidence. In: Sprouse, J., Hornstein, N. (Eds.), Experimental Syntax and

Island Effects.

Kehler, A., 2002. Coherence, Reference and the Theory of Grammar. CSLI Publications,

Stanford.

Kluender, R., 1998. On the distinction between strong and weak islands: A processing

perspective. In: Culicover, P. W., McNally, L. (Eds.), Syntax and Semantics 29.

Academic Press, New York, pp. 241-279.

Lakoff, G., 1986. Frame semantic control of the coordinate structure constraint. CLS 22,

Part 2: Papers from the Parasession on Pragmatics and Grammatical Theory, 152-

167.

Levine, R. D., Hukari, T. E., 2006. The Unity of Unbounded Dependency Constructions.

CSLI Publications, Stanford.

McCloskey, J., 2002. Resumption, successive cyclicity and the locality of operations. in.

S. D. Epstein and T. D. Seeley (Eds.), Derivation and Explanation in the Minimalist

Program, Blackwell, Oxford, pp. 184-226.

McCloskey, J., 2006. Resumption. In: Everaert M., van Riemsdijk, H. (Eds.), The

Blackwell Companion to Syntax. Blackwell, Oxford, pp. 94–117.

Nunes, J., 2001. Sideward movement. Linguistic Inquiry 32, 303-344.

Pesetsky, D., 1998. Some optimality principles of sentence pronunciation. In: Barbosa,

P., Fox, D., Hagstrom, P., McGinnis, M., Pesetsky, D. (Eds.), Is the Best Good

Enough. MIT Press, Cambridge, Mass., pp. 337–383.

Pollard, C., Sag, I. A., 1994. Head-driven Phrase Structure Grammar. University of Chicago

Press, Chicago.

Richards, M., 1938. Cystrawen y Frawddeg Gymraeg. Gwasg Prifysgol Cymru, Cardiff.

Ross, J. R., 1967. Constraints on Variables in Syntax. PhD dissertation, MIT.

Rouveret, A., 1994. Syntaxe du gallois: principes généraux et typologie. CNRS Éditions,

Paris.

Rouveret, A., 2002. How are resumptive pronouns linked to the periphery?, In: Pica, P.,

Rooryck, J. (Eds.), Linguistic Variation Yearbook: Volume 2. Benjamins.

Amsterdam, pp. 123–84.

Rouveret, A., 2008. Phasal agreement and reconstruction. In: Freidin, R., Otero, C. P.,

Zubizarreta, M-L. (Eds.), Foundational Issues in Linguistic Theory: Essays in Honor

of Jean-Roger Vergnaud, pp. 167-195.

Sadler, L., 1988. Welsh Syntax: A Government-Binding Approach. Croom Helm,

London.

Sells, P., 1984. Syntax and Semantics of Resumptive Pronouns. Ph. D. thesis, UMass

Amherst.

Sag, I. A., 2010. English filler-gap constructions. Language 86, 486-545.

Tallerman, M. O., 1983. Island constraints in Welsh. York Papers in Linguistics 10, 197–

204.

Tallerman, M. O, 1998. The uniform case-licensing of subjects in Welsh. The Linguistic

Review 15, 69-133.

Tallerman, M. O., 2006. The syntax of Welsh ‘direct object mutation’ revisited. Lingua

116, 1750-1776.

43

Tallerman, M. O., 2009. Phrase structure vs. dependency: the analysis of Welsh syntactic

soft mutation. Journal of Linguistics 45, 167-201.

Vaillette, N., 2000. Hebrew relative clauses in HPSG. Proceedings of the 7th

International Conference on Head-Driven Phrase Structure Grammar, CSLI

Publications. 305-324.

Vaillette, N., 2002. Irish gaps and resumptive pronouns in HPSG. Proceedings of the 8th

International Conference on Head-Driven Phrase Structure Grammar, CSLI

Publications. 284-299.

Watkins, T. A., 1961. Ieithyddiaeth. Gwasg Prifysgol Cymru, Cardiff.

Willis, D. W. E.. 2000. On the distribution of resumptive pronouns and wh-trace in

Welsh. Journal of Linguistics 36, 531-573.

Willis, D. W. E., 2011. The limits of resumption in Welsh wh-dependencies. In:

Rouveret, A. (Ed.), Resumptive pronouns at the Interfaces, Benjamins, Amsterdam,

pp. 189–222.

Zaenen, A., Engdahl, E., Maling, J., 1981. Resumptive pronouns can be syntactically

bound. Linguistic Inquiry 12, 679-692.

ON THE NATURE OF WELSH UNBOUNDED DEPENDENCIESprivaterborsley/Welshuds.pdf · ON THE NATURE OF WELSH UNBOUNDED DEPENDENCIES Robert D. Borsley ... Literary Welsh, but Colloquial Welsh

Documents