The Competition Model

(McWhinney 1987, Bates and McWhinney 1981)

· The relationship between element in a sentence are determined by factors or ‘cues’:

o word order

o knowledge of lexical items

§ animacy

§ subcategorization (what type of arguments a verb takes)

o morphology

§ agreement

§ case structure

· Language processing involves competition among the cues

· Languages vary in which cues the weigh more heavily

English weighs word order a primary cue

The dog chases the rabbit.

The rock crushed the man

Latin uses case marking as a primary cue

peulla acquilam portat

girl (nom) eagle (acc) carries

acquilam portat peulla

eagle (acc) carries girl (nom)

Italian uses agreement and lexical information

Ha comprato il vino Aldo

bought the wine Aldo

Japanese uses lexical information (e.g. animacy)

There is a preference for the subject to be animate

otoko-ga iwa-ni tsubusareta

man-subj rock-by was crushed

Is preferred to

iwa-ga otoko-wo tsubushita

rock-subj man-caus crushed

· What happens during L2 acquisition when new constraint rankings need to be learned?

· Learners first look for correspondences between L1 and L2 and try to transfer L1 constraint rankings

· If that strategy fails, they use the “universal” strategy of lexical knowledge and meaning.

e.g. NE learners of Japanese first apply SOV word order rigidly, but later become more flexible in the word order.

· In general speakers of L1’s that don’t use much syntactic information (word order), continue to use lexical knowledge and meaning as primary cues in an L2, even if that L2 primarily uses syntactic cues.

e.g. NJ learners of English don’t seem to rank word order highly in English.

MacWhinney, Brian. 1997. Second language acquisition and the Completion Model. In A. M. B. de Groot & J. F. Kroll (Eds.) Tutorials in Bilingualism: Psycholinguistic Perspectives (pp. 113-142). Mahwah, New Jersey: Lawrence Erlbaum.

Approaches to SLA

· Nativist-set parameters in UG

o SLA is seen as the same as FLA

o SLA is seen as different from FLA due to a critical period afterwhich UG cannot be accessed

· Empiricist-language learning is a constructive, data-driven process that relies on universals of cognitive structure

o Both SLA and FLA are seen a using the same mechanisms

o The Competition Model is of this type

The Competition Model: An Overview

1. Lexical Functionalism

Functionalism: Forms of language are determined and shaped by the communicative function to which they are placed.

Lexical functionalism: grammatical patterns are controlled by interactions between lexical items.

· Lexical basis for grammar

· Uses connectionist networks to express the syntactic and semantic properties of related lexical items

An example of Form-to-Function mapping: How do English language users determine the “agent” in a sentence?

Linguistic forms that cues “agent”

· preverbal positioning

· verb agreement morphology

· sentence initial positioning

· case marking (pronouns only)

· use of the pronoun “the”

The dog chases the cat.

She chases the cat.

If passive verbal morphology is present, it overrides the above cues

The cat was chased by the dog.

So, the relationship between the cues needs to be learned.

· Some cues are correlated with each other and can “reinforce” one another

· Some cues compete with other cues

· Determinations of grammatical status (e.g. “agent”) are probabilistic

· (Note that probabilistic determinations can lead to language change over time)

2. Connectionism

· Grammatical knowledge is seen as emergent across associations within the lexicon

· Words used for the same functions are associated with one another

· Associations betweens forms and functions are made overtime

· Associations are probabilistic in nature: They can be “strong” or “weak”

So, on this view, there are no lists of rules separate from the lexicon.

· What linguists have traditionally called rules are just emergent patterns

· “Rules” (read probabilistic patterns) do not have any existence independent from the words and phrases that gave rise to them.

Back to the agent example above

· The cues to agent have been established by the form-function pairings occurring over and over in English

3. Input Driven Learning

Form-Function mappings are learned according to the reliability of the input

· The more valid a cue is, the easier it is to learn

· So, the cues with higher validity will be acquired first and be the strongest determinates in processing

Languages differ in which cues are stronger (and thus arguably more valid)

· Experimentally tested on many different languages

· Design varies the cues under investigation

· Participants are asked to report on who they think did the action (was the agent)

Vary word order	Vary agreement
The cat is chasing the ducks.	The cat is chasing the ducks.
	The cat are chasing the ducks.
The ducks are chasing the cat.	The ducks are chasing the cat.
	The ducks is chasing the cat.
Is chasing the cat the ducks.	Is chasing the cat the ducks.
	Are chasing the cat the ducks.
Are chasing the cat the ducks.	Are chasing the cat the ducks.
	Is chasing the cat the ducks.
The ducks the cat is chasing.	The ducks the cat is chasing.
	The ducks the cat are chasing.
The cat the ducks are chasing.	The cat the ducks are chasing.
	The cat the ducks is chasing.

Languages differ in cue strength

· English SVO word order is very strong (NVN:SVO, VNN:VOS, NNV:OSV)

· Agreement in Italian in very strong. Less so for German

Figs 4.3 and 4.4 here

The first cues learned are predicted to be the most reliable

· English is word order

· Italian is Animacy, then Agreement (problem?)

· Hungarian is Case

Fig 4.5 here

The Competition Model in SLA

1. Transfer

In SLA, all aspects of the first language system (read set of connections emergent from lexical items) will transfer

· The learner begins with a parasitic lexicon and grammatical generalizations

· Overtime the two languages diverge and separate to a certain extent

o This happens as connections are established and strengthen within the L2 forms and functions

o As the L2 words and phrases develop connections distinct from L1 words and phrases, the two languages disengage.

· The L2 may require new generalizations to be made that did not exist in the L2

o e.g. learning case markings for an English speaker

2. Cues in SLA

What cues are used in L2 processing?

Are the cues developed based on the reliability of the input as in L1?

Are cues transferred from the L1? (as predicted above)

First study on German learner of English

· Strongest cues were agreement and animacy

· Word order was discounted when it disagreed with agreement or animacy

Later Studies found that cues used by native speakers of the L2 increased in use over time.

Fig 4.7 here

3. Differences in L1 and L2 learning

· Change in L2 from parasitic status (transferred) to a more separate status is dependent on using “fresh/unused” neural resources

· In learning the L1, much of the neural circuitry has been “committed”

· So, L1 and L2 learner differs in the amount of uncommitted circuitry available

Q: How different is this from critical period views?