What Makes Second Language Learning So Difficult?
Natasha Tokowicz
Difficulties in Adult Second Language Learning
• A lot of information to learn
• Embarrassment at speaking languageDifferences between the new language and
your native language(e.g., The Competition Model, MacWhinney & Bates, 1989)
So, Why Learn Another Language?
So, Why Learn Another Language?
QuickTime™ and a decompressor
are needed to see this picture.
So, Why Learn Another Language?
• Communicate with people of different language backgrounds– Especially an important message!
• Identify with another group
• Learn about another culture
• Learn more about your own language
What factors make adult second language learning difficult?
• Exp. 1: Meaning differences across languages– Multiple translations-more than one way to say
something
• Exp. 2: Syntactic differences across languages– Grammatical constructions that differ
• Experiment in progress– ERP techniques
Current Knowledge Both Helps and Hurts
• Adult second language learners:– Full set of concepts– Full set of labels for these concepts– Full grammatical system– Full system for contrasting sounds
• Sometimes these will transfer appropriately– E.g., same or similar labels (cognates): e.g., color-color
Mismatches between languages create problems
Connecting Meanings to Labels
• Initially, concepts are strongly connected to L1 words
• Eventually concepts must get connected to L2 words for comprehension or production
Two Labels for the Same Meaning?
• Most models assume the concepts activated by the two languages are the same– Exception: Distributed Feature Model
• Word concreteness– cat
– health
• Cognate status of translation pair– color-color
– house-casa
Conceptual Salience
Highconceptual
salience
Lowconceptual
salience
Distributed Feature ModelDe Groot (1992)
L1lexical(word)level
conceptual(meaning)
level
lexical (word)level
conceptual(meaning)
level
Concrete WordsCognate Translations L2
L1: First LanguageL2: Second Language
L1 Abstract WordsNoncognate Translations
L2
Why Else Meanings May Differ
• Different lexical concepts– “sibling” in Dutch = broers en zussen (brothers and sisters)
• Culturally-specific concepts– “gezellig” in Dutch = ???
• Culturally-distinct meanings– “sombrero”, “iglesia”
Broadness of application of terms in the two languages--semantic boundaries
Prepositions (Ijaz, 1986)
Semantic boundaries differ across languages– German learners of English under-emphasize
contact and over-emphasize movement for “on”• close translation equivalent of “on” in German is
“auf”, which can denote a motional meaning like the English word “up”
• Result is multiple translations, which are problematic for the L2 learner
Sources of Multiple Translations
• Imprecise correspondence across languages
• Synonymy– sofá sofa or couch
• Ambiguity within a language– glass vidrio or vaso in Spanish– to be ser or estar in Spanish
Ambiguity at Its Worst: “Trunk”
“achterbak” or “kofferbak”“romp”
“stam” or “boomstronk”
“slurf”
Experiment 1
• What happens when meanings differ? • Different consequences depending on type
of multiple translations– synonyms– multiple meanings
• Translation Task
+
cat
“gato”
Method of Experiment 1
• Participants: adult L2 learners– 24 Dutch-English Speakers
• Translated words aloud in both directions– L1 to L2
– L2 to L1
• Recorded responses– Accuracy
– Reaction time for correct responses
Design
• Manipulated – Number and type of translations
• Number of meanings
• Number of forms for meaning selected
– “Conceptual salience” (CS)• A composite measure of concreteness and context
availability
Stimuli
• Number of translations norms– Number of translations– Number of meanings translated
• E.g., if said “slurf”, translated snout meaning
• Semantic similarity ratings– How similar in meaning are these words?
• All pairs were considered translation equivalents
• Form similarity ratings– How similar are these words in spelling/sound?
• All pairs were considered translation equivalents
Predictions for Experiment 1
• High conceptual salience words faster than low conceptual salience words
• Multiple forms slower than one form– Need to select one to say (e.g., vaso, vidrio)
• Multiple meanings slower than one meaning– Need to select one to translate (e.g., trunk)
Data Analysis of Experiment 1
• Hierarchical regression analysis
• Covaried effects of word length and frequency
• Directionality taken into account– Data collapsed across direction
Results of Experiment 1
Low High
800
900
1000
1100
1200
One Meaning, One Form for Meaning Selected
Two Meanings, One Form for Meaning Selected
One Meaning, Two Forms for Meaning Selected
Conceptual Salience
Estimated Translation Latency (ms)
• Unambiguous words show CS effect
• Cost for multiple forms similar for high and low CS words
• Cost for multiple meanings only for high CS words
Accuracy Data: Experiment 1
• Unambiguous words show CS effect
• High CS words with multiple meanings lower in accuracy
Low High0
10
20
30
40
50
60
70
80
90
100One Meaning, One Form for Meaning Selected
Two Meanings, One Form for Meaning Selected
One Meaning, Two Forms for Meaning Selected
Conceptual Salience
Estimated Accuracy (%)
Tokowicz & KrollModel of Language Production
• Adapted from Poulisse & Bongaerts (1994)
• Stages of language production
• Sources of competition
Tokowicz & Kroll Model
MeaningLevel
WordLevel
strawberry “aardbei”
door strawberry raspberry deur aardbei framboos
Tokowicz & Kroll Model
MeaningLevel
WordLevel
strawberry “aardbei”
door strawberry raspberry deur aardbei framboos
Tokowicz & Kroll Model
MeaningLevel
WordLevel
strawberry “aardbei”
door strawberry raspberry deur aardbei framboos
Tokowicz & Kroll Model
MeaningLevel
WordLevel
strawberry “aardbei”
door strawberry raspberry deur aardbei framboos
Tokowicz & Kroll Model
MeaningLevel
WordLevel
strawberry “aardbei”
door strawberry raspberry deur aardbei framboos
Tokowicz & Kroll Model
MeaningLevel
WordLevel
strawberry “aardbei”
door strawberry raspberry deur aardbei framboos
Tokowicz & Kroll ModelMultiple Forms
MeaningLevel
WordLevel
throat “strot”
door throat mouth deur strot keel
Tokowicz & Kroll ModelMultiple Forms
MeaningLevel
WordLevel
throat “strot”
door throat mouth deur strot keel
Tokowicz & Kroll ModelMultiple Forms
MeaningLevel
WordLevel
throat “strot”
door throat mouth deur strot keel
Tokowicz & Kroll ModelMultiple Forms
MeaningLevel
WordLevel
throat “strot”
door throat mouth deur strot keel
Tokowicz & Kroll ModelMultiple Forms
MeaningLevel
WordLevel
throat “strot”
door throat mouth deur strot keel
competition
Tokowicz & Kroll ModelMultiple Forms
MeaningLevel
WordLevel
throat “strot”
door throat mouth deur strot keel
Tokowicz & Kroll ModelMultiple Meanings
MeaningLevel
WordLevel
calf “kuit”
door calf thigh calf cow deur dij kuit koe
Tokowicz & Kroll ModelMultiple Meanings
MeaningLevel
WordLevel
calf “kuit”
door calf thigh calf cow deur dij kuit koe
Tokowicz & Kroll ModelMultiple Meanings
MeaningLevel
WordLevel
calf “kuit”
door calf thigh calf cow deur dij kuit koe
Tokowicz & Kroll ModelMultiple Meanings
MeaningLevel
WordLevel
calf “kuit”
door calf thigh calf cow deur dij kuit koe
competition
Tokowicz & Kroll ModelMultiple Meanings
MeaningLevel
WordLevel
calf “kuit”
door calf thigh calf cow deur dij kuit koe
Tokowicz & Kroll ModelMultiple Meanings
MeaningLevel
WordLevel
calf “kuit”
door calf thigh calf cow deur dij kuit koe
Interim Summary
• Multiple forms create competition
• Multiple meanings create competition if simultaneously active
• Conceptual salience interacts with type of multiple translations
Extensions
• Add sentence or paragraph context– Reduce or eliminate conceptual salience effect– Reduce meaning effect– Not reduce form effect
• Developmental change– Less proficient learners will show less
competition
Competition at Other Levels of Language Production
• Start out like L1 and shift to L2 – Like prepositions (e.g., McDonald, 1987)
• Does the L1 really interfere with L2 grammatical processing?
Experiment 2:Grammaticality Judgment
• What happens when the two languages are in direct conflict?
• 14 Native English speakers learning Spanish as a second language
• Read grammatical and ungrammatical sentences in Spanish
• Push button responses regarding grammaticality
Grammatically Acceptable Stimuli
• Correct in Spanish (not English)– Se lavó el coche.
*Was washed the car.
• Correct in both languages– El coche fue lavado.
The car was washed.
Grammatically Unacceptable Stimuli
• Correct in English (not Spanish)– *El coche estuvo lavado.
The car was washed.
• Correct in neither language– *El coche lavado.
*The car washed.
Predictions-Experiment 2
• If L1 grammar interferes with L2 grammar:– “yes” responses
Should see difference between “both” and “Spanish” conditions
– “no” responsesShould see difference between “neither” and
“English” conditions
Should be difficult to reject English sentences
Reaction Time Data-Experiment 2
• RT corrected for length of sentences
• More difficult to reject English sentences
• No difference between Spanish and Both
Accuracy Data-Experiment 2
• Accuracy corrected for length
• More accurate at rejecting English sentences
• False alarms in Neither condition
Conclusions
• Stimuli in “both” condition less familiar?, or
• Accuracy in L1 leads learners to question acceptability in “Both” condition?
• Good at English only, but slow
Strategic effects?
Event-Related Brain Potentials
• Temporally sensitive measure of on-line processing
• Derived from the electroencephalographic (EEG) record
• Reflect synchronous depolarization of populations of neurons
ERP Setup
• Electrodes are placed painlessly on the surface of the scalp
• These electrodes record brain activity– Background activity
– Stimulus-locked activity
ERP Components
• Grammatical violations elicit a “P600” response– A positive-going deflection in voltage that
occurs between approximately 500 and 700 ms post stimulus
Legend:
Osterhout & Nicol (1999)
ERP Components
• Semantic violations elicit a “N400” response– A negative-going deflection in voltage that
occurs between approximately 300 and 500 ms post stimulus
Legend:
Osterhout & Nicol (1999)
Grammaticality Judgment-ERP
• Native English speakers learning Spanish as a second language
• Read grammatical and ungrammatical sentences in English and Spanish– Separate blocks for English and Spanish
• Push button responses regarding grammaticality
The
old
blender
doesn’t
crushing
ice
cubes.
?
Predictions
• Ungrammatical sentences will elicit a P600 response compared to grammatical sentences– Significantly more positive amplitude between
500 and 700 ms post-stimulus
Acceptability in English
N400
P600
Acceptable Unacceptable
The old blender doesn’t crushing ice cubes.
Acceptability in Spanish
Acceptable Unacceptable
P600?
N400
Él trabajando cada día.
Summary
• Why is second language learning so difficult?– Differences between the two languages
• Multiple levels– Multiple forms
– Multiple meanings
– Different grammatical constructions
Future Directions
• On-line examination of semantic differences – E.g., The pizza was too hot to eat/*drink.
• Are the effects similar for different language pairs that have more/fewer differences?
• Are there competition effects in phonology?• Would context reduce the competition
observed in Experiment 1?
Acknowledgments
• Brian MacWhinney
• Judith Kroll
• NIH
• NSF
• Sigma Xi
• Penn State RGSO