The canonical word order myth: Investigating a processingtypological puzzle in the Cantonese double object construction Antonio Cheung (University of Hawai’i at Manoa) & Stephen Matthews (The University of Hong Kong) Contact: antonio1 @ hawaii.edu
2. Cantonese: the testing ground
1. Canonical advantage or head proximity?
Potential center-embedding in the canonical order a language processing problem inefficient to process
Rare combination (Dryer, 2005) a typological puzzle
Double object construction
Double object construction (DOC) Structure: [S V DO IO]
S
S
Alternative construction NP
Cantonese DOC is canonical:
The BA construction Structure: [S BA DO VASP IO]
BA construction
frequent for expressions of giving structurally basic, and pragmatically neutral, BUT Heavy DOs center-embedding
N
佢
NP
VP
NPDO
V
畀咗
NPIO
XP
N
N
我 買 嗰 本 得意 嘅
書
你
[ S/he ] give [ I buy that CL funny
VP
BA
N
NPDO
將
佢
XP
N
我 買 嗰 本 得意 嘅
書
[ S/he ] BA [ I buy that CL funny
NPIO
V
Structural frequency not high Use more restrictive in Cantonese (Matthews & Yip, 1994) than in Mandarin (Li & Thomson, 1981) Only for highly-transitive VPs Only definite direct objects can occur after BA morpheme
N
你
畀咗
book ] give [ you ] VP processing domain “[S/he] gives [you] [the funny book that [I bought]]”
book ] [ you ] VP processing domain “[S/he] gives [you] [the funny book that [I bought]]” PRT
PRT
3. Experiments : Effects of center-embedding in language comprehension and production Expt 1: Elicited production : DOC
Method: Self-paced reading with elicited imitation 1. Self-paced word-by-word reading (comprehension) 2. “REPEAT” probe at end of sentence 3. Elicited imitation of the sentence (production) Expt 2 results (comprehension) Participants consistently read BA faster than the DOC (average RT/word, F1(1,17)=4.933, p=.040, p2 n.s.) DO complexity also significant (by participants) (F1(2,16)=5.246, p=.024, p2 n.s.).
Expt 2 results (production) Imitation considered correct when same structure repeated. DOC is recalled less accurately than the BA construction (F1(1,17)=11.86, p=.003, F2(1,23)=0.419, p=.524) DO complexity also significant (p’s<0.001) Expt 2 : Elicited imitation (accuracy of structure)
Expt 2 : Self-paced reading (average time per region) 1.0
450
0.9 440
0.8 0.7
430 Accuracy
Method: Untimed elicited production 1. Recorded sentence played at key press 2. Key press to indicate participant is ready 3. Question “What does [the subject] do?” presented 4. Free response (full sentence) to question Expt 1 results (production) Attested responses: DOC (repetition) (1) BA construction (2) Serial verb construction (SVC) [S V DO bei2 IO] Verb doubling [S give DO give IO] Topicalization [DO, S give IO] Heavy NP shift (HNPS) [S V IO [DO]] Shifting from the DOC increases as DO gets more complex (F1(2,40)=55.97, p<.001, F2(2,33)=17.01, p<.001).
Compare between both language comprehension and production of the DOC and the BA construction Participants: 18 Cantonese native speakers Items: 12 sets of full sentences (with bei2 “give”) [S V [N] IO] [S BA [N] V IO] 3 DO lengths: N / Adj-N / RC-N [S V [Adj-N] IO] [S BA [Adj-N] V IO] 2 constructions (DOC or BA) [S V [[RC] N] IO] [S BA [[RC] N] V IO]
RT (ms)
No parsed Cantonese corpora Informal corpus search: DOC frequent, RCs rare Elicit prod. of DOC with heavy DOs to quantify effects of center-embedding Participants: 21 Cantonese native speakers Items: 12 sets of full DOC sentences (with bei2 “give”) 3 DO lengths: N / Adj-N / RC-N [S V [N] IO] [S V [Adj-N] IO] [S V [[RC] N] IO]
Expt 2: Dual-task (comprehension + production) : DOC vs. BA
420 410 400
0.1 0.0
200
DOC: canonical word order most frequent response (imitation) very little shifting when DO is simple
150 Frequency
0.4
0.2
390 380 DOC BA
Bare N
Adj-N
RC-N
407.55 387.34
398.44 389.97
438.25 422.59
DOC BA
520
100
Alternative word orders: less frequent frequency of shifting increases with DO complexity
50
500
480
DOC
V NP V NP
SVC
BA
Others
238 206 169
9 28 38
4 6 13
0 1 15
1 11 17
Selected references Hawkins, J. (2004). Efficiency and Complexity in Grammars. Oxford: OUP. Matthews, S. & L. Yeung (2001). Processing motivations for topicalization in Cantonese. In Horie & Sato, eds., Cognitive-Functional Linguistics in an East Asian Context. Tokyo:Kurosio.
Acknowledgements The authors would like to thank Dr William O’Grady, Dr Amy Schafer, Dr Kamil Deen, Dr Conrad Perry and anonymous reviewers for their helpful comments. All remaining errors are our own. This research was partially supported by grants from the Hong Kong Research Grants Council (proj refnos: HKU 725804H, HKU 748207H) & by a Postgraduate Studentship from the University of Hong Kong.
RT (ms)
460 0
440
420
400
380
360
Bare N
Adj-N
RC-N
0.81 0.96
0.72 0.94
0.54 0.86
4. Implications and conclusion
Expt 2 : Self-paced reading
Bare N Adj-N RC-N
0.5
0.3
Expt 1 : Elicited production
250
0.6
[ Cl_Subj
Subj ]
[ give / BA
[[ N_RC
V_RC ]
Det
Cl_DO
[ Adj ]
ge
DO]
DOC
378.9
407.2
400.8
426.4
486.9
427.6
385.9
451.6
437.5
444.7
BA
385.3
401.9
396.5
396.4
468.1
395.2
394.3
437.4
402.1
420.3
(∅ / give)
[ Cl_IO 411.1
391.8
388.1
Canonical structures not necessarily easier to process. Cantonese DOC harder to process than BA Increased DO complexity center-embedding slower comprehension (Expt 2: slower RT) & production difficulty (Exp 1: lower freq of DOC in elicited production; Expt 2: lower accuracy in elicited imitation) “Canonical advantage”: NOT supported Frequency is not an explanation: frequent ≠ easier to process Direction of causality: easy to process frequent ? Non-canonical BA processed faster than the DOC Not all alternative structures incur a processing penalty (cf. topicalization in Matthews & Yeung, 2001) “Penalty” comes from unpredictability, not for the BA-cxn: BA reliably predicts DOsyntactic relations assigned early (cf. Maximize On-line Processing, Hawkins, 2004). Word orders with potential center-embedding are rare due to the demands on processing which they incur (Hawkins, 2004). Cantonese canonical word order retained because speakers can use alternative constructions to avoid serious center-embedding in the canonical word order. Speakers can frequently use the DOC because the working memory can tolerate center-embedding when the DO is relatively simple.