Linking Processing to Knowledge through Short-Term Memory for Input

GoalsTo be passed on to the developing system, intake needs to survive in short-term memory (STM).Are distortions of L2 input (by L1 and general processing preferences) lasting?Is short-term memory different for units extracted from input according to L1 or L2 rules by learners in the initial state?DesignSequence: exposure (varied by group) followed by test of recognition memory for units extracted from input sentences (Fig.2)

Correlation between RT, Accuracy and Rating Score

(p < .05,r = 0.35)

Katya Solovyeva & Robert DeKeyser (kss@umd.edu)

ResultsDiscrimination scores (old versus new items), D’, computed for items following L1 syntax (blond athlete) and L2 syntax (journalist blond): z(hits) – z(false alarms). Overall discrimination higher in E group [F(1,73) = 9.69, p < .001, η2

part. = .12], driven by better recognition of old items [F(1, 72) = 17.82, p < .0001, η2

part. = .20].

BackgroundA transition account of SLA relies on connecting online processing of input to resulting linguistic representations[1]. However, these two endpoints (and interim steps) have been addressed by separate theoretical accounts.Input: Shallow Structure Hypothesis[2]

Input/Intake à Output: Processing Instruction[3]

Output: Processability Theory[4], Universal Grammar accounts[5]From VanPatten (2004, p. 26)

An integration of theories is complicated by:• Circularity, mutual influences between processing and prior knowledge.

Input processing is causally related to the developing linguistic system, and is itself influenced by knowledge of L1 and L2 competence.

Initial input processing is uniquely positioned to disentangle the two.• Lack of common vocabulary of constructs.The causal sequence (Fig. 1) logically relies on memory, which can serve as the common denominator for theoretical accounts.

Figure 2. Sequence of procedures.

Artificial language: • vocabulary from cognates between English

(L1) and Dutch, French• structure with opposite parameter values for

word order (Fig. 3)

Conclusions• Without meaning cues (Control group),

learners tended to retain sequences extracted by L1 word order rules.

• When provided with cues to meaning, learners (Experimental group) overcame the L1 bias and were equally sensitive to L2 units.

• Contrary to expectations, L2 units were not significantly disadvantaged, compared to units following L1 syntax.

This may underlie the relative ease of acquisition of word order. Structures known to cause more persistent L1 transfer may exhibit an L2-L1 imbalance in STM.There was considerable variation among participants in the direction of the preference, so effect size is very small and power is low.Future studies need to pursue the full causal chain from processing to linguistic knowledge (e.g., grammaticality intuitions).

References1. Gregg,K.(1996).Thelogicalanddevelopmental

problemsofsecondlanguageacquisition.InW.C.Ritchie&T.K.Bhatia(Eds.),Handbookofsecondlanguageacquisition.London:AcademicPress.

2. Clahsen,H.,&Felser,C.(2006).Grammaticalprocessinginlanguagelearners.AppliedPsycholinguistics,27,3-42.

3. Pienemann,M.(2015).AnoutlineofProcessabilityTheoryanditsrelationshiptootherapproachestoSLA. LanguageLearning, 65(1),123-151.

4. VanPatten,B.(2004). InputprocessinginSLA.InB.VanPatten(Ed.),Processinginstruction:Theory,research,andcommentary.Mahwah,NJ:LawrenceErlbaumAssociates,Publishers.

5. White,L.(2003).SecondlanguageacquisitionandUniversalGrammar.Cambridge,UK:CambridgeUniversityPress.

Groups (different in exposure only): • Experimental (E), 2-3 pictures before each

input sentence; task—decide if new picture matches the sentence

• Control (C), no pictures (asterisks for same duration as for E group); task—easy or difficult to follow

• Identical recognition memory test

Cross-linguistic garden path

Which are remembered better?

Figure 3. Word order in L1 (English) and target language.

Materials• 18 sentences following pattern of interest, S-

Adj-O-V; 20 fillers ~ 20-25 minutes• Split 9-9 for testing, rotated among subjects• Testing: old or new? 78 phrases, words

• Discrimination of L1 units (blond athlete) did not significantly differ between groups

F(1, 72) = 1.71, p = .19 η2part. = .02

• Discrimination of L2 units (journalist blond) was significantly higher in E group

F(1, 71) = 6.59, p = .01 η2part. = .08

Figure 4. Discrimination of L1, L2 units (d’ scores) in Experimental, Control groups.

Figure 5. Distribution of d’ scores for L1, L2 units in Experimental, Control groups.

Mixed-effects model with Language (L1, L2) as within factor; Group (E, C) as between factor; overall memory accuracy as covariate: Language n.s. F(1, 69) = 0.50, p = .48Group F(1, 71) = 9.88, p < .01Overall accuracy F(1, 71) = 42.8, p < .0001Accuracy*Group F(1, 71) = 5.42, p = .02

00.10.20.30.40.50.60.70.80.91

L2units L1units

meanD'sc

ore

Experimental

Control

Journalist blond athlete interviewed.

… blond athlete…

Journalist blond…

Exposure Trial Structure• Visual stimuli: 1166 ms. per picture (or *)

E: pic.1, pic. 2., blank screen. C: *, *, blank screen• Audio of sentence, push à to advance• Distractor task or new sequence

Participants• 48 in E group, 26 in C group• L1 English, no childhood bilinguals, ages 18–

24, mean 19.5• No concurrent foreign language courses