To To what what extent extent does does the the difficulty difficulty of of processing
processing relative clauses relative clauses parallel parallel typological
typological complexity complexity ? ?
Claire Delle Luche Claire Delle Luche Frédérique Gayraud Frédérique Gayraud
Bruno Martinie Bruno Martinie Fanny Meunier
Fanny Meunier--HoenHoen
Laboratoire Dynamique du Langage, Université Lyon 2 Lumière, Lyo
Laboratoire Dynamique du Langage, Université Lyon 2 Lumière, Lyon, Francen, France
Complexity and language Complexity and language
Quantity = Complexity?
Quantity = Complexity?
Difficulty = Complexity?
Difficulty = Complexity?
Quantity
Quantity of information isof information is not the predominantnot the predominant factor of complexityfactor of complexity (1) Please(1) Please pickpick up four up four tomatoestomatoes, a pound of , a pound of apricots, prune apricots, prune juicejuice, ,
shallots
shallots, six , six applesapples and a bag of carrotsand a bag of carrots on the on the wayway home.home.
(2) The man
(2) The man thatthat the womanthe woman thatthat the childthe child huggedhugged kissedkissed laughed.laughed.
–– (2) is more difficult to process than (1), the syntactic (2) is more difficult to process than (1), the syntactic
Introduction (1) Introduction (1)
Relativization in general Relativization in general
–– RCsRCs are sentential modifiers of nouns.are sentential modifiers of nouns.
–– The operation of relativization can be marked by different The operation of relativization can be marked by different ways, but in any case, the relativized constituent is moved ways, but in any case, the relativized constituent is moved
from its canonical position.
from its canonical position.
–– Main idiosyncratic properties in the world languages:Main idiosyncratic properties in the world languages:
The relativizer can be compulsory (French), optional The relativizer can be compulsory (French), optional
(English, Arabic) or never available (Japanese).
(English, Arabic) or never available (Japanese).
Possible morphological variation of the mark according to Possible morphological variation of the mark according to
the relativized position (French
the relativized position (French vs.vs. Chinese).Chinese).
Possible marking by special verbal morphology (Turkish).
Possible marking by special verbal morphology (Turkish).
Introduction (2) Introduction (2)
Relativization in standard French Relativization in standard French
Example
Example : center : center - - embedded embedded O RC O RC
–– The objectThe object NP NP isis movedmoved fromfrom itsits canonical postverbal canonical postverbal position
position
–– PresencePresence of a relativizerof a relativizer
–– The morphologyThe morphology of the relativizer (QUE) of the relativizer (QUE) indicatesindicates the the
(3) L’homme fume. Le chien chasse l’homme.
(3) L’homme fume. Le chien chasse l’homme.
(4) L’homme que le chien chasse fume.
(4) L’homme que le chien chasse fume.
(3) L’homme fume. Le chien chasse l’homme.
(3) L’homme fume. Le chien chasse l’homme.
(4) L’
(4) L’hommehomme que le chien chasse fume.que le chien chasse fume.
(3) L’homme fume. Le chien chasse l’homme.
(3) L’homme fume. Le chien chasse l’homme.
(4) L’
(4) L’hommehomme queque le chien chasse fume.le chien chasse fume.
Introduction (3) Introduction (3)
Difference
Difference between between O and S RCS : O O and S RCS : O RCs RCs (5) (5) more
more difficult difficult to to process process than than S RCS (6) S RCS (6)
(Caplan, & (Caplan, &Waters, 1999; Ford, 1983;
Waters, 1999; Ford, 1983; FrazierFrazier, 1985; Gibson, 1998, 2000; , 1985; Gibson, 1998, 2000; GouveaGouvea, 2000; King, & , 2000; King, &
Just, 1991;
Just, 1991; TraxlerTraxler, Morris, & , Morris, & SeelySeely, 2002; , 2002; YngveYngve, 1960), 1960)
(5) (5) The reporter The reporter thatthat the the senatorsenator attackedattacked admittedadmitted the the errorerror.. (6) The reporter (6) The reporter thatthat attackedattacked the senatorthe senator admittedadmitted the errorthe error..
Several
Several theoretical theoretical attempts attempts to to account account for the S for the S O O difference difference
Canonical word
Canonical word orderorder Functions
Functions parallelismparallelism
Canonical word
Canonical word order order
French canonical word order = SVO French canonical word order = SVO
S RC: SV(O) S RC: SV(O)
(7) Le
(7) Le sénateursénateur qui attaquequi attaque le reporter admetle reporter admet l’erreur.l’erreur.
(The senator that attacked the reporter admitted the error.) (The senator that attacked the reporter admitted the error.)
O RC: OSV O RC: OSV
(8) Le
(8) Le sénateursénateur queque le reporter attaquele reporter attaque admetadmet l’erreur.l’erreur.
(The senator that the reporter attacked admitted the error.) (The senator that the reporter attacked admitted the error.)
Function
Function parallelism parallelism
In a language acquisition perspective, Sheldon In a language acquisition perspective, Sheldon
(1974) claims that
(1974) claims that (9) (9) is easier to process than is easier to process than (10) (10) and and (11), (11), because only in because only in (9), (9), the relativized the relativized NP NP The singer The singer has the same function both in the has the same function both in the
matrix and in the relative clause:
matrix and in the relative clause:
(9) She(9) She despisesdespises the singer thatthe singer that youyou admire.admire.
(10)
(10) The singer The singer thatthat youyou admire admire isis on stage on stage tonighttonight.. (11)
(11) SheShe despisesdespises the singer thatthe singer that isis on stage.on stage.
Depth
Depth of of embedding embedding
Hawkins (1994) : in a language such as French, a S RC Hawkins (1994) : in a language such as French, a S RC
is predicted to be less complex to process than a O RC.
is predicted to be less complex to process than a O RC.
The reason is the following : in O
The reason is the following : in O RCs RCs , the size of the , the size of the portion of the tree involved in the
portion of the tree involved in the coindexation coindexation between between
t tand the head noun is larger.
and the head noun is larger.
Working
Working memory memory accounts accounts of RC of RC processing
processing
Working
Working memory memory accounts accounts of RC of RC processing processing : :
–– King, & Just (1991)King, & Just (1991)
–– Caplan, & Waters (1999)Caplan, & Waters (1999)
–– Gibson (1998, 2000): the DLT, a Gibson (1998, 2000): the DLT, a computationalcomputational approach
approach
Extended
Extended predictions predictions
(12) OG RC:
(12) OG RC: Le sénateur dont le reporter attaque le parti admet l’erreur.Le sénateur dont le reporter attaque le parti admet l’erreur.
The The DLTheory DLTheory
Two Two processing processing costs costs : : memory memory and and integration integration costs
costs
Predictions Predictions
(8) O RC:
(8) O RC: Le Le sénateursénateur queque le reporter le reporter attaqueattaque admetadmet l’erreurl’erreur.. (7) S RC:
(7) S RC: Le Le sénateursénateur qui qui attaqueattaque le reporter le reporter admetadmet l’erreurl’erreur.. 1+21+2 33
33 11
11
A A typological typological approach approach
The The Accessibility Accessibility Hierarchy Hierarchy
(Keenan(Keenan, & , & ComrieComrie, 1987), 1987): :
S > O > IO > OBL > G >
S > O > IO > OBL > G > OCompOComp
Principles Principles : :
–– «« A languageA language must must bebe able to relativize [S position].able to relativize [S position].
–– [Strategies[Strategies applyapply] on a ] on a contiguouscontiguous segment of the AH.segment of the AH.
–– StrategiesStrategies can can thatthat applyapply atat one point of the AH may one point of the AH may cease
cease to to applyapply atat anyany lowerlower levels.levels. » » (Keenan(Keenan, & , & ComrieComrie, 1987, p. 6), 1987, p. 6)
Examples Examples : :
–– S S onlyonly: Malagasy: Malagasy –– SS-G: French-G: French
The AH and
The AH and comprehension comprehension
Keenan
Keenan , & Hawkins (1987): , & Hawkins (1987):
repetitionrepetition procedureprocedureMaterials
Materials : :
36 sentences (9 36 sentences (9 differentdifferent positions in the AH)positions in the AH)(13) S:
(13) S: I know I know thatthat the girl the girl whowho gotgot the the answeranswer right right isis cleverclever.. (14) O:
(14) O: He He rememberedremembered thatthat the sweetsthe sweets whichwhich Dave gave Sally wereDave gave Sally were a treata treat.. (15) OG:
(15) OG: The The factfact thatthat the the sailorsailor whosewhose shipship Jim Jim tooktook hadhad one one legleg isis important.important.
(16)
(16) OCompOComp: He : He rememberedremembered thatthat the girl whothe girl who Jane isJane is olderolder thanthan couldcould read.read.
Results
Results : :
the AH the AH isis supportedsupported by by adultadult and and children’schildren’s datadataPredictions Predictions
Canonical word
Canonical word order order Functions
Functions parallelism parallelism
Depth
Depth of of embedding embedding Working
Working memory memory cost cost
S > O, IO, OBL, G S > O, IO, OBL, G
Our Our experiment experiment
Method Method
–– Participants: 30 native speakers of French (21.26, s.d. Participants: 30 native speakers of French (21.26, s.d.
2.49) 2.49)
–– MaterialsMaterials: 36 : 36 experimentalexperimental sentences, 3 conditions (S, O, sentences, 3 conditions (S, O, OG)OG)
(17) S:
(17) S: La danseuse qui regarde le chanteur appelle le studio La danseuse qui regarde le chanteur appelle le studio (The dancer(The dancer thatthat isis lookinglooking atatthe singer the singer isis callingcallingthe studio)the studio)
(18) O:
(18) O: La danseuse que le chanteur regarde appelle le studio La danseuse que le chanteur regarde appelle le studio (The dancer(The dancer thatthat the singer isthe singer is lookinglooking atat isis callingcallingthe studio)the studio)
(19) OG:
(19) OG: La danseuse dont le chanteur regarde les jambes appelle le La danseuse dont le chanteur regarde les jambes appelle le studio
studio
Results Results
Participants
Participants with with more more than than 35% of 35% of extreme extreme reading times (<100ms or >1500ms)
reading times (<100ms or >1500ms) were were discarded
discarded , , thus thus leaving leaving 27 for 27 for statistical statistical analysis analysis Comprehension
Comprehension errors errors : :
Significant
Significant effecteffect of RC typeof RC type
Mean error rate
0 5 10 15 20 25 30
Error rate (%)
Reading times (1) Reading times (1)
Regions
Regions of of interest interest : :
–– Object NP of Object NP of the S and OG the S and OG RCsRCs
–– Embedded Embedded verbverb
–– Main clause Main clause verbverb
Mean reading times for each region
400 450 500 550
RT (ms)
S RC O RC OG RC
Mean reading times for each region
400 450 500 550
RT (ms)
S RC O RC OG RC
Reading times (2) Reading times (2)
Object NP of the S Object NP of the S
and OG
and OG RCs RCs
–– No differenceNo difference
M ean RT for the embedded object NPs
0 200 400 600
RT (ms)
Reading times (3) Reading times (3)
RC RC verb verb
–– EffectEffect of RC typeof RC type
M ean RT on the RC verb
400 600
RT (ms)
*
*
M ean RT on the RC verb
400 600
RT (ms)
**
**
Reading times (4) Reading times (4)
Main
Main verb verb
–– No No significantsignificant difference
difference
M ean RT on the main verb
200 400 600
RT (ms)
Discussion (1) Discussion (1)
S S RCs RCs are are faster faster to to read read and and generate generate less less errors
errors than than the the other other types (O, OG) types (O, OG) Different
Different from from previous previous findings findings Different
Different from from what what is is predicted predicted by the by the depth depth of of embedding
embedding and DL and DL theories theories Compatible
Compatible with with canonical word canonical word order order and and
Discussion (2) Discussion (2)
On On theories theories of of language language processing processing : :
– – Memory Memory cost cost is is not not infirmed infirmed by canonical by canonical word
word order order nor nor functions functions parallelism parallelism theories theories – – Necessity Necessity to to disentangle disentangle canonical word canonical word order order
from from functions functions parallelism parallelism
Further experiment Further experiment
L’homme
L’homme que que la la femme regarde Parallel
Parallel - -
L’homme
L’homme qui qui regarde
regarde la femme la femme promène
promène le le chien. chien . Parallel
Parallel + +
Canonical Canonical - - Canonical +
Canonical +
L’homme
L’homme regarde regarde la femme qui
La femme
La femme regarde regarde le le chien chien que que
l’homme
l’homme promène promène . .
Discussion (3) Discussion (3)
Complexity
Complexity = = Difficulty Difficulty ? ? Structural
Structural complexity complexity (depth ( depth of embedding of embedding , , number number of of syntactic syntactic nodes nodes , long distance , long distance dependencies dependencies ) ) does does not automatically not automatically trigger processing trigger processing difficulty difficulty Nevertheless
Nevertheless it it does does not not mean mean that that complex complex sentences are
sentences are easy easy to to process process
The lion
The lion thatthat the the gorillagorilla thatthat the the tigertiger bit bit chasedchased dieddied
Difficulty
Difficulty is is not equivalent not equivalent to to complexity complexity
Why? Why?
YES YES Processing
Processing
YES YES
S RC < O RC S RC < O RC