The Formation of Relative Clauses in Colloquial Indonesian: Evidence for NonSubject Relativization J

1
The Formation of Relative Clauses in Colloquial
IndonesianEvidence for Non-Subject
RelativizationJan 03, 2004

• Peter Cole, Gabriella Hermon,
• and Yassir Tjung
• University of Delaware

2
Goals of the Talk

• To investigate which grammatical position(s) can
  be relativized in the restrictive relative clause
  (RC) in Indonesian. Specifically, does the
  so-called subject constraint on relativization
  hold in Indonesian?
• Naturalistic and experimental data from adult and
  children speaking colloquial Indonesian indicate
  that the subject constraint on relativization
  does not appear to be operative in colloquial
  Indonesian to the same extent as in the standard
  variety.

3
Relative Clauses (RCs)

• The RC construction consists of two components
• a. The head noun
• b. The restricting clause
• The semantic function of the head noun is to
  establish a set of entities or the domain of
  relativization.
• The semantic function of the restricting clause
  is to identify a subset of the domain.
• (Keenan and Comrie 1977).

4
Examples (1) The girl that John loves is the
governors daughter. the NP, the girl, represents
the domain of the relativization, which is then
narrowed down to the only entity that can
satisfy the condition expressed by the
restricting clause that John loves. We shall
refer to the NP the girl as the head and to
the (empty) position following the verb loves as
the relativized NP.
5
Indonesian We shall regard the RC construction
in Indonesian to be one involving a. head noun
(overt or non-overt) b. yang-restricting
clause c. a missing (relativized NP) inside the
RC Example of RC with overt head (2) Orang
yang nyium Siti (itu) ditangkap polisi. orang
represents the domain of relativization, and the
restricting clause yang nyium Siti identifies the
subset of the domain. Example of RC without
overt head (3) yang nyium Siti (itu) ditangkap
polisi.
6

• The standard analysis for the RC
  construction
• The head noun, orang, is base-generated in the
  matrix clause
• The relativized NP (what Comrie 2003 calls the
  notional head and others label a GAP) which is
  coreferential with the head noun is missing.
• There may be a NULL OP moving to spec CP from
  the relativized position
• (4) Orangi CP OP i yang IP ti nyium Siti
  (itu)
• ditangkap polisi.

7
Two views regarding which grammatical positions
can be relativized a. The subject-only view
(e.g. Dardjowidjojo 1973, Sie 1988, Sneddon
1996) The strategy for relativizing with a gap
is only available for the subject position. b.
The non-subject view (e.g. Chung 1976, Cole and
Hermon 1998, Musgrave 2001) The strategy for
relativizing is not restricted solely to the
subject position inside the yang-clause. The two
views seem contradictory.
8

• Our claim
• The apparent contradiction in the literature is
  due to various authors examining different
  registers of the language.
• The subject constraint holds in the standard
  language (SI).
• It has been relaxed in colloquial Indonesian,
  which has a more liberal grammar.Our data
  comes from the colloquial variety spoken in the
  Jakarta area (CJI). In this dialect, both
  subjects and non-subjects can be relativized.
  However, even in CJI there is a strong preference
  or tendency for relativizing.

9
Relativization on subject position is OK in
Standard Indonesian (SI) (5) RC with
meng-prefix and missing subject orang yang
mencium dia (itu) ditangkap polisi. (6) RC
with di-prefix and missing subject orang yang
dicium (oleh) dia (itu) ditangkap
polisi.(7) RC with bare verb (passive semu) and
missing subject orang yang dia cium (itu)
ditangkap polisi.Relativization on direct
objects is ungrammatical in SI (8) orang
yang dia mencium (itu) ditangkap polisi.
10
The Accessibility Hierarchy (Keenan and Comrie
1977) (9) Subject gt Direct Object gt Indirect
Object gt Object of Preposition gt Possessor gt
adjunct If DO relativization is ungrammatical in
a certain language, then relativization from
lower positions (such as IO or possessor) will
also be ungrammatical
11

• Do examples ( 6-7) both involve relativization
  on subjects?
• (6) orang yang dicium (oleh) dia (itu)
  ditangkap polisi.(7) orang yang dia cium
  (itu) ditangkap polisi.
• Bare verbs in SI are usually analyzed as
  passive semu.
• In colloquial Indonesian (CI) the nasal prefix
  is often omitted, as described in Chung 1978 and
  Kaswanti Purwo 2003.
• Unclear whether this is passive or active verb

12
However, in CI there are two possible derivations
for (7) (10) a. Orangi CP OPi yang IP ti
dia cium (itu) ditangkap polisi. b.
Orangi CP OPi yang IP dia cium ti (itu)
ditangkap polisi. In (10a), the relativized NP
is the subject of a passive semu sentence
while in (10b), the relativized NP is the object
of a stem sentence (Chung 1978), as illustrated
below (11) a. Orang itusubj dia
cium. (Passive semu) b. Dia cium orang
ituobj. (Stem sentence)
13

• The two derivations can be distinguished by
  syntactic tests.
• Test 1 for Passive Semu
• In Passive semu nothing can interfere between the
  agent and the verb.
• All negative, aspectual, and AUX elements must
  precede the agent
• the negative marker nggak not
• auxiliaries and modals like boleh can, bakalan
  will
• aspectual/temporal markers like (s)udah
  already, lagi in
• progress
• (12) a. Orang itu dia nggak/udah cium.
• b. Orang itu nggak/udah dia cium.

14
In stem sentences, elements like nggak/udah CAN
occur between the agent and the verb (13) a.
Dia nggak/udah cium orang itu. b. Nggak/udah
dia cium orang itu.
15

• Predictions of the Two Views
• (14) a. Orang yang nggak/udah dia cium (itu)
  ditangkap
• polisi.
• b. Orang yang dia nggak/udah cium
  (itu) ditangkap
• polisi.
• Only sentence (14a) is predicted to be possible
  by the subject-only view (since 14b) is not an
  examples of Passive Semu.
• Both sentences in (14) are predicted to be
  grammatical by the non-subject view. (14a) is
  relativization on subject (14b) is
  relativization on object position.

16
Test 2 for Passive Semu For most people, full
NPs (like gadis ini) cannot be agents in the
Passive Semu construction. The agent in Passive
Semu is required to be a personal pronoun (15)
Orang itu gadis ini cium. (Passive Semu)
17
Predictions of the Two Views The subject-only
view predicts (16) to be ungrammatical, since it
would be derived from (17) (16) Orang yang
gadis ini cium (itu) ditangkap polisi. (17)
Orangi CP OPi yang IP ti gadis ini cium itu
ditangkap polisi.
18
The non-subject view predicts (16) to be
grammatical. According to this view, (16) has the
derivation in (18), in which the DO was
relativized directly. (18) Orangi CP OPi yang
IP gadis ini cium ti (itu) ditangkap
polisi. Based on (18), the NP that is
relativized is the direct object of the verb cium
in a stem sentence like (19). (19) Gadis ini
cium orang ituobj . (Stem sentence)
19
To summarize, the subject-only and the
non-subject views make different predictions with
respect to sentences like (14) and (16), repeated
below. (14) a. Orang yang nggak/udah dia
cium itu ditangkap polisi. b. Orang
yang dia nggak/udah cium itu ditangkap
polisi. (16) Orang yang gadis ini cium (itu)
ditangkap polisi. The three sentences above are
predicted to be grammatical by the non-subject
hypothesis, while only (14a) is predicted to be
grammatical by the subject-only view. We shall
show next that data from adults and children
speaking colloquial Indonesian support the
non-subject view.
20

• Data
• a. Adults
• Sources of Data for CJI
• database of adult-to-adult speech collected by
  Yassir Tjung
• MPI database of adult-to-child speech

21

• Coding System
• RCs preceded by the complementizer yang were
  extracted and coded as having either a subject
  gap or a (potential) non-subject gap.
• RCs coded as having (potential) non-subject gaps
  were subdivided into four types
• Indeterminate
• Compatible with the word order of the Passive
  Semu construction
• Incompatible with the word order of the
  Passive Semu construction
• Adjunct relativization

22
(20) Indeterminate yang saya hubungi mungkin
bangsa koperasi-koperasi gitu. (21)
Compatible with the word order of the Passive
Semu construction Tarolah sahabat cewek yang
udah elu kenal baek
23
(22) Incompatible with the word order of the
Passive Semu construction Pasti ada sesuatu
dalam diri dia yang gua nggak punya (24)
Adjunct relativization (for ex. RC of
time_) ya(ng)... yang kita mo ke Semarang.
(Bun, this is a child utterance, could not find
and adult one, make one up? Find one from
Sneddons data? can not use your original ex with
WAKTU - since we do not understand it and YANG
preceded waktu
24
Figure 1 Subject versus Non-subject
Relativization
25
Figure 2 Non-subject Relativization
26

• Adults overwhelmingly use subject gaps and have
  few object gaps (fig 1).
• The majority of examples of (potential)
  non-subject gaps (67) are sentences which could
  be examples of either direct relativization on
  object position or relativization on a subject in
  the Passive Semu construction (fig 2).
• (25) a. Cewek yang elu mau kayak apa?
• b. Bisa aja kan cewek itu termasuk yang
  dia idamkan?
• The word order of the following examples (21)
  are also compatible with the Passive Semu
  construction.
• (26) a. Banyak yang bisa kita buat ...
• b. ... cewek yang udah elu kenal baek
  ...

27
However, the following examples (11) do not
have the characteristics of Passive Semu word
order and are arguably cases of direct
relativization from object position. (27) a.
... yang gua bisa jalanin, ya gua jalanin.
b. ... the dark side of Yuli yang banyak
orang nggak tau Adjunct relativization is also
attested, albeit there is only case in the data.
Conclusions Adults in CJI allow relativization
on non-subject positions.
28

• b. Children
• Sources of Data
• MPI database of the speech of three children
  (from the age 20 to 52)
• Coding System (same as for adults)
• RCs preceded by the complementizer yang were
  extracted and coded as having either a subject
  gap or a (potential) non-subject gap.
• RCs coded as having (potential) non-subject gaps
  are subdivided into four types
• Indeterminate
• Compatible with the word order of the Passive
  Semu construction
• Incompatible with the word order of the Passive
  Semu construction
• Adjunct relativization

29
Figure 3 Subject versus non-subject
relativization
30
Figure 4 Non-subject relativization
31

• As seen in Figure 3, children overwhelmingly use
  more subject gaps than adults do (about 99).
• In Figure 4, the majority of examples of
  (potential) non-subject gaps (80) are sentences
  which could be examples of either direct
  relativization on object position or
  relativization on a subject in the Passive Semu
  construction

32

• (28) a. Ini yang aku cari
• b. ... yang Mama beli
• Only two examples out of 30 (potential)
  non-subject gaps are clearly compatible with the
  Passive Semu construction (given our strict
  criteria).
• (29) a. Anjing yang pernah Ica liat
• b. Tante yang udah Ica kasih, ininya.

33
Recall earlier that we found 11 of cases of
clear non-subject gap relativization in adult
speech, which led us to conclude that the
non-subject-gap-relativization strategy is an
option available in the grammars of adult
speakers of colloquial Indonesian. We would like
to argue that similar cases also occur in
childrens speech even though they are much less
frequent than with adults (only two
instances) (30) a. Soalnya kan ada yang Ca
belom ... b. ... itu yang, yang Pak
Polisi lagi ...
34
Why do children have very few object
gaps? 1)Frequency Children can only relativize
on subjects, since this is the most frequent
pattern in the input. They ignore infrequent
patterns. Prediction Indonesian children will
radically differ from English children. 2)
The Subset Principle and the Accessibility
Hierarchy The AH is a guiding principle and the
various setting in the AH stand in a subset
relation. Prediction children universally
start out with the most restrictive setting
supported by the data. Subject-only is the most
restrictive setting.
35
Why do children have very few object gaps? 3)
Processing rather than competence The low
frequency of object gaps is due to a processing
effect subject relativization imposes a lighter
processing load than object relativization, since
the gap is easier to recover. (Diessel (2002),
Diessel and Tomasello (2000), Hawkins
(2000). Prediction children in all languages
will exhibit a preference for subject gaps at a
early age. This is true even if objects are
quite frequent in the adult language (English).
English and Indonesian children may not differ.
36
What do the data show A comparison of RCs in
English child language and Indonesian child
language
37
(No Transcript)
38

39

• By comparing RCs in English and Indonesian child
  language
• we conclude that
• Object gaps in both English and Indonesian are
  less frequent
• than subject gaps, but are available. This argues
  against
• the subset view and against discontinuity in the
  grammar.
• Since in both languages subject gaps are much
  more frequent, the preference is probably not
  due to the grammar but due to
• processing.

40

• Indonesian children have a much lower of
  object RCs than
• English children. This argues for an account
  which incorporates
• frequency effects (in addition to processing
  effects).
• Since relativization on object position is
  extremely rare in the Indonesian naturalistic
  child corpus, this may be an effect of how the
  data was collected. Even at age 502 (an age when
  English children exibit robust object gaps),
  Indonesian children have almost 100 subject
  gaps. Perhaps if we put children in a situation
  in which object gaps are more natural, they
  would relativize on objects with greater ease.

41

• Experimental Data
• Elicited Production Task based on Zukowskis
  (2001) RC experiment. The task for the child is
  to convey to an uninformed listener one of the
  two objects/characters whose identity is most
  felicitously expressible by an RC.
• Subjects 20 kindergarten children aged 50 -
  60
• Materials 8 base pictures (4 used to elicit
  subject gap RCs and and 4 used to elicit object
  gap RCs)
• Subject-gap pictures, 1 animate Subject-gap
  pictures, 2 animates
• a. Anak lagi tendang/tangkep c. Anak lagi tunjuk
  sapi/kuda
• bola
• b. Anak perempuan lagi nyanyi/ d. Anak perempuan
  lagi uber
• gambar kucing/kelinci

42

• Object-gap pictures, 1 animate Object-gap
  pictures, 2 animates
• e. Anak perempuan lagi dudukin/ f. Anak
  perempuan/anjing lagi
• lompatin truk maenan uber kucing
• f. Anak laki/anak perempuan lagi h. Anak
  perempuan/kucing lagi
• nonton TV lompatin orang
• Each base picture consists of a duplicated
  character or object, and the two are engaged in
  minimally different situations. For example,
  Picture a has two identical boys, each with a
  ball, but one boy is kicking it, while the other
  is catching it. The character or object intended
  to be the head of the RC is shown in bold face
  above.
• Each of the 8 base pictures are manipulated to
  create 4 stimuli
• One calling for an NP-only response to a matrix
  subject question
• One calling for an NP-only response to a matrix
  object question

43

• One calling for a full-sentence response with
  the RC modifying the matrix subject
• One calling for a full-sentence response with
  the RC modifying the matrix object
• The 8 base pictures X 4 manipulations result in
  32 trials for each child 16 with NP-only targets
  and 16 with full-sentence targets.Since the full
  sentence trials yield opportunities for 2 RCs,
  this results in 48 total opportunities for RCs
  per child.
• The pictures are collected into an Acrobat PDF
  file for presentation to children. For each of
  the 32 trials, 2 pictures will be included--the
  base picture, and immediately following it, the
  changed picture. The 32 trials are arranged in
  2 types of order, which are counterbalanced among
  subjects.

44

• Procedure
• The elicitation task is introduced to the child
  as a game, jointly administered by the child and
  the experimenter to the childs teacher.
• The child sits with the experimenter on one side
  of a table looking at a laptop computer, while
  the childs teacher sits on the other side of the
  table looking at copies of pictures.
• The child is shown the base picture and is told
  that when the space bar is pushed, something will
  happen to the picture on the screen, namely
  either a little mouse appears in the picture and
  looks at one of the characters/objects, or one of
  the characters changes colors.
• When the change happens, the child is supposed
  to tell the teacher what happens. The teacher
  then tries to picture a picture that matches the
  childs description.

45
Base Picture
46
Question tikus liat anak perempuan mana? Target
sentence anak perempuan yang lagi nyanyi
47

• Result
• 48 sentences X 20 children 960 sentences
• 24 sentences requiring RCs with subject gap and
  24 sentences requiring RCs with object gap
• Coding system
• RC with subject gap
• (31) a. orang yang lempar bola
• b. yang lagi lempar bola warna pink yang
  lagi nendang bola warna biru
• c. orang yang lagi nendang bola
• d. tikus liat orang yang lagi nendang
  bola kalo burung liat orang yang lagi
• lempar bola
• (CHI-HW)

48

• (32) a. kucing yang dikejar sama orang
• b. bebek ngeliat kucing yang dikejar sama
  orang terus tikusnya liat
• kucing yang dikejar sama anjing
• c. yang dikejar sama anjing
• d. kucingnya yang dikejar sama anjing jadi
  ungu yang dikejar sama
• orang jadi coklat
• (CHI-PL)
• RC with object gap
• (33) a. yang anak perempuan lagi dudukin
• b. mobil-mobilan yang perempuannya lagi
  loncatin
• c. bebek lagi liat mobil-mobilan yang
  perempuan lagi dudukin tikus
• liat mobil-mobilan yang perempuannya
  lagi loncatin
• d. mobil-mobilan yang anak perempuan lagi
  dudukin jadi biru yang
• ungu yang anak perempuan lagi loncatin
• (CHI-MF)

49

• (34) a. yang anjing lagi kejar (CHI-VG)
• b. tikus liat kucing yang anjing pengen
  nangkep burung liat kucing
• yang orang pengen nangkep (CHI-ML)
• RC with unclear gap
• missing verb
• (35) a. tikus liat TV yang perempuan
• b. yang perempuan TV-nya jadi merah
• (CHI-PC)
• no gap
• (36) a. yang anjing ngejar-ngejar kucing jadi
  ungu
• b. liat yang kucing lagi ngelompatin orang
• (CHI-PC)

50

• Non-RCs
• (37) Child uses simple sentences
• a. anjing lagi ngejar kucing
• b. orang lagi ngelompatin orang
• (CHI-KL)
• (38) Child uses simple NPs
• a. TV cowok
• b. TV cewek
• (CHI-KL)
• (39) Child uses VPs
• a. nunjuk sapi
• b. ngejar kelinci
• (CHI-KL)
• (40) Child uses complex predicates

51
Table 5. Target Subject Gap
52
Table 6. Target Object Gap
53

• Children do not have trouble with RCs with
  subject gap they successfully produce 448 out of
  480 opportunities (more than 93). This mirrors
  the English experimental data.
• They appear to have more trouble with RCs with
  object gap. Children easily convert objects to
  subjects (by using a di-verb), 319/480
  opportunities (about 66.5) .
• Many children produce non-RCs (about 12. 7) and
  RCs which have unclear gap (about 17) in this
  condition. Again, this mirrors the English data
  and must therefore be due to processing
  constraints.

54

• However, the subject-only view would have
  difficulty
• explaining the existence of RCs with clear object
  gap
• (18 cases, 3.8).
• NOTE
• In addition, if we count all the cases
• in which the analysis may support an object gap
  the percentage
• is much higher.gtgtgtgtgtgtgtBUNRECOUNT HERE TO
  INCLUDE ALL CASES OF PASS SEMU?
• The experimental data then show that while
  children are very sensitive to frequencies, they
  do allow object gaps in their grammar. The is
  less than the in the adult input due to
  additional processing restrictions (the same
  restrictions which are at work in English),

55
Why does Colloquial Indonesian diverge from
Standard Indonesian? Comrie ( 2003) notes
that often languages in which only the subject
can be relativized have a rich voice system that
permits non-subjects to be presented as
subjects. The subject constraint may then hold
for SI since it has a rich voice system. This
constraint, however, might be ignored in CI
since CI permits constructions with bare verbs
(stem sentences).
56
A possible scenario In CI stem sentences and
passive semu are collapsed into a single
construction. Not only are verbs bare in both
constructions, but the word order constraints and
the pronoun constraint on agents in Passive semu
have also been lost. Since Passive semu is not a
separate construction, the evidence for a subject
only constraint on relativization is lost.

57
Two grammars now compete in acquisition of
CI A. a grammar which limits gaps to
subjectsB. a grammar which allows both subject
and object gapsEach time sentences like (25)
or (26) are heard, Grammar B is supported (and
eventually will win out).
58
(25) a. Cewek yang elu mau kayak apa?
b. Bisa aja kan cewek itu termasuk yang dia
idamkan? (26) a. Banyak yang bisa kita buat
b. ... cewek yang udah elu kenal baek
This scenario, follows the computational
principles of acquisition and diachronic change
described in Yang (2003). (BUN ADD reference?)