Input-related determinants of L2 construction learning Afra Alishahi Ad Backus Tilburg University, the Netherlands ICCG-8, Osnabrück
Study overview L2 construction learning Computational model How do quantitative properties of language input affect L2 learning?
Determinants of L2 learning What can affect L2 learning results?
Determinants of L2 learning What can affect L2 learning results?
Determinants of L2 learning What can affect L2 learning results? E.g., age of onset, memory, amount of exposure and motivation [Foster, Bolibaugh, & Kotula, 2014]
Input-related determinants of L2 learning Focus on input properties [Goldschneider & DeKeyser, 2001] Explaining the Natural Order of L2 Morpheme Acquisition in English: A Meta-analysis of Multiple Determinants - frequency - phonological salience - semantic complexity - etc.
Input-related determinants of L2 learning Item Variables Frequency Salience Complexity... Acquisition score morph1 morph2... [Goldschneider & DeKeyser, 2001]
Input-related determinants of L2 construction learning Further focus on construction learning [Ellis, O'Donnell, & Römer, 2014]: Determinants of learning argument structure constructions in L1: (1) verb frequency (F) (2) contingency between verb and construction (ΔP) (3) semantic prototypicality [Ellis & Ferreira-Junior, 2009]: Similar work for L2, but: - 3 constructions - 3 variables tend to correlate with each other
Input-related determinants of L2 construction learning RQ: Do the same factors (F and ΔP) determine L2 construction learning? Compared to [Ellis & Ferreira-Junior, 2009]: - Larger data - Highly controlled setting (computational model) - Show individual contribution of each factor
1. Frequency Frequency of verbs in a certain argument structure construction: E.g., prepositional dative (transfer) construction: He it to someone. give 1000 show 150 send 50 lend 10...
2. Contingency Contingency, or reliability of verb construction mapping. Prepositional dative (transfer) construction give 1000 Prepositional dative (transfer) construction post 10
2. Contingency Contingency, or reliability of verb construction mapping. Prepositional dative (transfer) construction Other constructions give 1000 5000 other verbs 3000 Prepositional dative (transfer) construction Other constructions post 10 3 other verbs 3990
2. Contingency Contingency, or reliability of verb construction mapping. ΔP (construction verb) = a/(a+b) c/(c+d) Prepositional dative (transfer) construction Other constructions give a c other verbs b d
The model Based on a probabilistic model of early argument structure learning [Alishahi & Stevenson, 2008] Simulates the process of learning constructions in two languages
Learning argument structure constructions
Learning argument structure constructions
Learning argument structure constructions
Learning argument structure constructions The bear gives you the ball!
Learning argument structure constructions The bear gives you the ball
Learning argument structure constructions The bear gives you the ball Daddy's coming home!
Learning argument structure constructions The bear gives you the ball Daddy's coming home
Learning argument structure constructions The bear gives you the ball Daddy's coming home Grandma sent you some cookies. John passed you the ball! Mr. Rich donated us a thousand dollars.
Learning argument structure constructions Grandma sent you some cookies Mr. Rich donated us a thousand dollars The bear gives you the ball John passed you the ball Daddy's coming home
Learning argument structure constructions Grandma sent you some cookies Mr. Rich donated us a thousand dollars The bear gives you the ball Daddy's coming home John passed you the ball Predicate meaning changing object possessor Number of arguments 3 Word order X verb Y Z Argument meanings {human}; {human}; {object} Argument roles Giver; Recipient; Theme
Learning argument structure constructions Ditransitive transfer construction Daddy's coming home
Learning argument structure constructions: L2 Ditransitive transfer construction Resultative construction...
Learning argument structure constructions: L2 Ditransitive transfer construction Resultative construction... Meine Schwester lieh mir Geld. (My sister lent me some money.)
Learning argument structure constructions: L2 Ditransitive transfer construction Resultative construction... Meine Schwester lieh mir Geld. (My sister lent me some money.)
Learning argument structure constructions: L2 Ditransitive transfer construction Resultative construction... Das Geld gab ich meiner Mutter. (I gave the money to my mother.)
Learning argument structure constructions: L2 Ditransitive transfer construction Resultative construction... Das Geld gab ich meiner Mutter Das Geld gab ich meiner Mutter. (I gave the money to my mother.)
Learning argument structure constructions: L2 L2 L2 L1 mixed L1
Evaluating language knowledge Predicate give Predicate meaning changing object possessor Number of arguments 3 Word order X verb Y Z Argument meanings {human}; {human}; {object} Argument roles Giver; Recipient; Theme
Evaluating language knowledge Predicate XXXXXXXX give Predicate meaning changing object possessor Number of arguments 3 Word order X verb Y Z Argument meanings {human}; {human}; {object} Argument roles Giver; Recipient; Theme
Evaluating language knowledge Giver verb Recipient Theme The bear gives you the ball!
Evaluating language knowledge Elicited production task Giver verb Recipient Theme The bear you the ball!
Design Items Variables Construction Verb Frequency Contingency Probability of use Constr1 Constr2 Verb1 Verb2 Verb1 Verb2...
Data English PropBank + Penn Treebank + SemLink + FrameNet + WordNet ~ 3,800 frame instances SALSA + TIGER corpus + dict.cc + WordNet ~ 3,400 frame instances German
Data English PropBank + Penn Treebank + SemLink + FrameNet + WordNet ~ 3,800 frame instances SALSA + TIGER corpus + dict.cc + WordNet ~ 3,400 frame instances German Both datasets can be used as either L1 or L2. The actual input in each simulation is sampled from the data.
Learning scenario L1 input Mixed L1 + L2 input
Learning scenario L1 input Mixed L1 + L2 input Test
Results: raw data Construction 'Process start' ARG1 VERB Items Variables Score Verb Frequency F Contingency ΔP Probability begin 14 0.56 0.37 start 3 0.37 0.01 erupt 1 0.04 0.09
Results: raw data Construction 'Process start' ARG1 VERB Items Variables Score Verb Frequency F Contingency ΔP Probability begin 14 0.56 0.37 start 3 0.37 0.01 erupt 1 0.04 0.09 - mixed effect model - fixed effects of F and ΔP - random effect of construction
Results: statistics L2 Predictor Coefficient β P-value English F 1.19 <.001 *** ΔP 0.17 <.001 *** German F 1.26 <.001 *** ΔP 0.74 <.001 ***
Summary 1. Verb input frequency F determines its production frequency. 2. But not only! ΔP is important too.
Summary 1. Verb input frequency F determines its production frequency. 2. But not only! ΔP is important too. 3. Supports and extends the existing results for L2 learning. [Ellis & Ferreira-Junior, 2009] both F and ΔP are important, despite their collinearity holds for large data in a controlled setting
Summary 1. Verb input frequency F determines its production frequency. 2. But not only! ΔP is important too. 3. Supports and extends the existing results for L2 learning. [Ellis & Ferreira-Junior, 2009] both F and ΔP are important, despite their collinearity holds for large data in a controlled setting 4. Partly in line with the existing results for L1 learning. [Ellis, O'Donnell, & Römer, 2014] + both F and ΔP are important for L2 (unlike for L1) F is a better predictor than ΔP
Representing language knowledge 1. Distribution Input properties distribution of verbs within a certain construction Open task [Ellis, O'Donnell, & Römer, 2014]
Representing language knowledge 1. Distribution 2. Proficiency score Input properties Input properties distribution of verbs within a certain construction proficiency score for verbs within a certain construction Open task Closed task [Ellis, O'Donnell, & Römer, 2014] [Goldschneider & DeKeyser, 2001]
Evaluating language knowledge 1. Distribution (elicited production) Giver verb Recipient Theme The bear you the ball! 1. Verb production probability
Evaluating language knowledge 1. Distribution (elicited production) 2. Proficiency score (comprehension) Giver verb Recipient Theme The bear you the ball! Giver verb Recipient Theme The bear gives you the ball!? 1. Verb production probability 2. Verb comprehension score
Results: raw data Construction 'Process start' ARG1 VERB Items Variables Score Verb Frequency F Contingency ΔP Proficiency begin 14 0.56 1 start 3 0.37 0.68 erupt 1 0.04 0.19
Results: proficiency score L2 Predictor Coefficient β P-value English F 0.31 <.001 *** ΔP -0.02 >.05 German F 0.15 <.001 *** ΔP 0.00 >.05
Results: proficiency score L2 Predictor Coefficient β P-value English F 0.31 <.001 *** ΔP -0.02 >.05 German F 0.15 <.001 *** ΔP 0.00 >.05 Only frequency is important. A. Distribution vs. proficiency? B. Production vs. comprehension? C. Specific to our statistical learner? D. Specific to data?
Future work Add other components, such as semantics (centrality of verb meaning). Test the proficiency vs. distribution issue in human subject research.
References Matusevych, Y., Alishahi, A., & Backus, A. (2014). Isolating second language learning factors in a computational study of bilingual construction acquisition. In Proceedings of CogSci-2014. Ellis, N. C., & Ferreira-Junior, F. (2009). Constructions and their acquisition: Islands and the distinctiveness of their occupancy. Annual Review of Cognitive Linguistics, 7(1), 187 220. Ellis, N. C., O Donnell, M. B., & Römer, U. (2014). The Processing of Verb-Argument Constructions is Sensitive to Form, Function, Frequency, Contingency, and Prototypicality. Cognitive Linguistics, 25(1), 55 98. Goldschneider, J. M., & DeKeyser, R. M. (2001). Explaining the Natural Order of L2 Morpheme Acquisition in English: A Meta analysis of Multiple Determinants. Language Learning, 51(1), 1 50.
SLA modeling: opinions We need models of acquisition that relate such measures to longitudinal patterns of child language and second language acquisition (Nick Ellis) Future research on implicit learning must implement computer simulations of language learning (Jan Hulstijn) Modeling SLA is hardest problem in linguistics. It is only hoped that computational models may contribute (Rens Bod)
An example frame I ate a tuna sandwich.