Investigating the Effect of Chunking Ability
in Adult L2 Processing

Background: Humans tend to chunk continuous information into discrete, familiar units. For instance, a string like “199420001995” would be commonly divided into chunks such as 1994-2000-1995. The ability chunk information in real time is crucial for language processing. Recent research has provided evidence that individual differences in chunking ability are a good predictor of first language (L1) processing. However, no study so far has examined whether individual differences in chunking abilities can also predict second language (L2) processing outcomes. While this idea is congruent with a view of chunking as a domain-general mechanism, it is possible that L2 processing is better predicted by chunking ability measured in the L2.

 

Research Question: In this study, we examined whether processing in the L2 is predicted by general chunking abilities as measured by a multiword chunking task in the native language, or whether it is better predicted by the same task in the L2.

​

Design: To address this question, we designed an experiment in which L2 comprehenders read relative clauses such as Example (1) containing the thematic preposition ‘with’. Even though the interpretation of the sentence is potentially ambiguous, there is a strong bias to attach the predicate to the Noun-Phrase 2 (NP2) following ‘with’, rather than the NP1. This preference has been shown to be crosslinguistic, with processing costs (i.e., slower reading times, RTs) found when input favors the dispreferred interpretation:

 

(1) There comes the flight attendant with the pilot that is nice to everyone.

                                          NP1                                NP2

​

In the current experiment, we capitalize on the fact that gender agreement is a cue specific to Spanish that can guide how the relative clause is interpreted (sometimes requiring the predicate to be attached to the NP1). For instance, in (2), the adjectives agree in gender with the NP2 (preferred interpretation), but in (3) the adjectives agree with the NP1 (dispreferred). Participants completed a self-paced reading task with 80 target sentences (50% NP1-attachment, 50% NP2-attachment) and 80 fillers.

 

(2) Ahí viene la azafata-FEM con el piloto-MASC que es simpático-MASC con todo el mundo.

(3) Ahí viene la azafata-FEM con el piloto-MASC que es simpática-FEM con todo el mundo.

​

Results: A mixed-effects regression analysis revealed that

English chunking ability was a significant predictor of

reading times (p = .04), while Spanish (L2) chunking

ability was not (p = .82), in line with the idea that this is a 

domain-general mechanism. Furthermore, there was an

interaction of English chunking ability and type of sentence

(NP1 relative to NP2 attachment; p= .04). Reading times

showed that only readers with higher chunking ability in

the L1 aligned with patterns reported for native speakers

in previous studies, revealing a cost in processing sentences

with NP1 attachment (Fig.1). Readerswith lower

chunking ability did not show the cost found in native

speakers,suggesting that they had shallow processing of

gender-based cues

​