Heritage Processing

Spanish–English bilingual heritage speakers’ processing of inanimate sentences (Casper et al., 2024)

Citation

Casper, R., Aguirre-Muñoz, Z., Spivey, M., & Bortfeld, H. (2024). Spanish–English bilingual heritage speakers’ processing of inanimate sentences. Frontiers in Language Sciences, 3, 1370569.

My thoughts

I’m a little worried about their word-order manipulation and their experimental setting. They note that some orders are ungrammatical in English. It also seems they used a sentence-reading paradigm with heritage Spanish–English bilinguals and acknowledged that many participants lacked formal education in Spanish, so the Spanish reading measures might be biased in this case? As they reported most participants were English-dominant; if many were, the Spanish reading tasks might not reflect full comprehension given lower Spanish literacy of such population, which may make sense that this is why they found non-significance for Spanish LexTALE?

Summary

This study investigated how heritage Spanish–English bilinguals (N = 30) integrate cue hierarchies to process simple sentences in both languages (N = 80, simple IA-IA sentences in two word orders, SVO and OVS) using sentence-reading and then ask the participants to do a subject-choice tasks. They also measured proficiency with English LexTALE, Spanish LexTALE-ESP, the BLP, and domain-general control with a Flanker task. Building on prior work showing that heritage and late bilinguals differ from monolinguals yet often rely on good-enough processing (e.g., first-noun/agent heuristics overriding syntax), the study probed competition between strong cues (e.g., word order) and weaker cues (e.g., animacy of the nouns) from Spanish and English. They used Linear mixed-effects regression, Generalized linear mixed-effects models to build Subject choice selection models, RT models and Subject choice RT models for both languages. Results showed that, for Spanish sentences, participants read OVS faster; in English, sentence RTs did not differ by word order. Higher English proficiency (LexTALE) is a significant predictor in their English model but not in their Spanish model, which predicted faster reading in English and faster subject-choice times in both languages, whereas Spanish LexTALE-ESP was not a significant predictor in both Spanish and English models. They also found that greater English dominance (BLP) slowed Spanish reading. The authors interpret these patterns as consistent with semantic-driven, good-enough processing under cue competition.

Key Concepts

The unified competition model

Bilinguals show a variety of patterns in sentence interpretation strategies depending on language proficiency

Good-enough processing

When the first-noun agent heuristic, a semantic representation of the sentence, will be processed instead of a syntactic one
heuristics and semantic processing
syntactic processing

Background

Describe lexical-based content and sentence internal content, since they will be the most relevant due to the use of simple sentences
for sentence internal interpretation cues = used to determine the thematic roles of a sentence, which are the relationships between the verb and its nouns
Interpretation cues can have a variety of expressions:
- morphological case marking
- Animacy of the nouns
- intonation
- verbal inflections
- syntax or word order
  Theory
The competition Model (CM) and Unified Competition Model
- sentence interpretation is achieved by weighing the various cues that are present in the sentence against the language’s cue hierarchy
  - English monolinguals: rely most on word order followed by agreement and animacy
  - Spanish monolingual: rely most on differential object marking (DOM), which is more in the stimuli section, followed by verb agreement, clitic agreement, then word order
  - Bilinguals: 2 competing cue sets which based on language profienciecy
  - 4 different types of cue hierarchies
    1. amalgamation (merged cues from all the known languages) -> past research suggest this appraoch
    2. differentiation (each language uses its own distinct cues)
    3. forward transfer (L1 cues used in L2)
    4. backward transfer (L2 used in L1)
bilinguals aquired the agreement cue earlier than English monolinguals but later than Spanish monolingulas Overall: Spanish-English bilinguals do not process or produced sentences in the same way as monolinguals do.
Individual differences
Özsoy et al. (2023) has monolingual Turkish and Turkish-German heritage speakers listen to sentences in Turkish while looking at visual world paradigm display with reference objects, and found heritage speakers do use case cues predictively, but more monolinguals use predictive processing than bilinguals -> might be individual differences and learner background

Language Proficiency

bilinguals’ language experiences, especially literacy in both languages were signficant predcitors (prediction-by-production accounts), suggests that reading and writing training in either of the speaker’s languages can help with cue prediction (Karaca et al., 2024).
Higher proficiency later bilinguals were more accurate and faster in response to the OVS word order due to more exposure and experience in L2 (Copeland, 2022).
especially for late bilinguals, L1/L2 similarities predcited sucessgul prediction of cues.

Overall

Based on the previous literature, it is evident that bilingual sentence processing involves a complex interplay between different cues and strategies, language proficiency, cue hierarchies, and the integration of lexico-semantic information. The studies conducted on Spanish–English bilinguals, heritage speakers, and late bilinguals highlight the variation in cue utilization and sentence interpretation patterns compared to monolinguals.

Research Aim

describe lexical-based content and sentence internal content, since they will be the most relevant due to the use of simple sentences, and looks at hoe heritage sprakers have integrated two of the most reliable cues from English and Spanish
Building upon these findings, the present study aims to investigate the cues employed by bilingual Spanish–English speakers when agreement and word-order cues are in competition. Additionally, the study aims to explore the influence of proficiency on cue interpretation RTs during sentence reading
we analyzed RT for sentence reading and choice selection and looked at the difference in the binary subject choice in both English and Spanish

Research questions and Hypothesis

What cues do bilingual Spanish–English speakers use when agreement and word-order cues are in competition with each other? Do cues vary by language input (English or Spanish sentences)?
- In English and Spanish, subject–verb–object (SVO) sentences will elicit faster RTs.
- OVS sentences in English will elicit slower RTs, but participants will still overwhelmingly choose the traditional SVO word-order subject.
  1. How does proficiency affect cue interpretation RTs during sentence reading?
Participants of higher proficiency will choose the verb-agreeing subject (second noun) significantly more than those of lower proficiency on the OVS Spanish sentences
Participants that score higher on measures related to proficiency (the Lexical Test for Advanced Learners of English [LexTALE], the Lexical Test for Advanced Learners of English–Spanish [LexTALE-ESP], the Bilingual Language Profile [BLP], and Flanker) will elicit faster RTs to the Spanish OVS sentences than those with lower proficiency.

Method

2 independent variables

word order and verb agreement (word orders, some sentences is ungrammatical, SVO and OVS sentences in both Spanish and English, OVS is not a viable word order in English)
English only marks the accusative via word order and pronouns
Spanish marks the accusative on animate nouns via the accusative/personal a (DOM), which is used with animate references
Heritage speakers tend to omit DOM more but due to individual factors which is due to language exposure and birth order
Participants
- 32 Spanish-English heritage speakers
Materials
- 80 inanimate sentences in English (40) and Spanish (40) with subject-verb-object and object-verb-subject word orders.
- The task itself involved determining the subject of each sentence.
- proficiency and dominance
  - proficiency: LexTALE and LexTALE-ESP (measures receptive vocabulary? or regonization of form, also positively correlated with proficiency)
  - dominance:
    - BLP: measure of language dominance that is related to language proficiency and usage
cognitive skills - Flanker task: domain general cognitive skills, such as working memory, attention, and the like - Declines in cognitive skills associated with aging also lead to declines in language processing and prediction

Analyses

Linear mixed-effects regression
Generalized linear mixed-effects models
RTs were clustered and centered by language and word order, 4 clusters were averaged and then the cluster mean was substracted from the vlaue
The LexTALE and LexTALE-ESP were scored via the averaged percentage, and then were z-scored to have comparable scales in data analysis
The BLP was scored according to traditional scoring methods (Birdsong et al., 2012) and then standardized.
The Flanker task was scored by taking the RTs of the correct congruent trials from the correct incongruent trials for an interference RT.
2 GLMM or logistic mixed models for each language
- to predict subject choice with word order, LexTALE score, LexTALE-ESP score, BLP score, and Flanker Interference RT with the following R code:
- subject choice ∼ word order + LexTALE score + LexTALE-ESP score + BLP score + Flanker Interference RT. The models included ParticipantID as a random effect (code: ∼1 | ParticipantID)2.
  Results
  
  Subject choice selection models
Spanish
- The effect of word order OVS is statistically significant and positive
- The effect of LexTALE score is statistically non-significant and positive
- The effect of LexTALE-ESP score is statistically nonsignificant and positive
The effect of BLP score is statistically non-significant and negative
LexTALE, LexTALE-ESP, and BLP scores are not significant predictors in choosing the subject of the sentence.
English
- The effect of word order OVS is statistically significant and positive
- The effect of LexTALE score is statistically nonsignificant and negative
- The effect of LexTALE-ESP score is statistically non-significant and positive
- The effect of BLP score is statistically nonsignificant and negative
- The effect of Flanker Interference RT is statistically non-significant and negative
- Like with Spanish choice selection, LexTALE, LexTALE-ESP, BLP scores, and Flanker Interference RT are not significant predictors.

RT models = answer our second research question about whether proficiency affected RTs

4 linear mixed models, one for each language by both sentence or choice reading

(RT ∼ word order + LexTALE score + LexTALE-ESP score + BLP score + Flanker Interference RT). The sentence RT models included word order as random effects (code: ∼1 + word order

ParticipantID)3

Spanish
The effect of word order OVS is statistically significant and negative (predicts faster RTs in reading Spanish sentences)
The effect of LexTALE score is nearly statistically significant and negative
The effect of LexTALE-ESP score is statistically non-significant and positive
The effect of BLP score is statistically significant and positive (Participants who are more English-dominant are slower at reading Spanish sentences)
English
The effect of word order OVS is statistically non-significant and negative (Unlike with the Spanish condition, the OVS word order is not significantly different from the SVO word order)
The effect of LexTALE score is statistically significant and negative
The effect of LexTALEESP score is statistically non-significant and positive
- for the English sentences, English vocabulary size or proficiency predicts faster RTs, and Spanish vocabulary size or proficiency is not a significant predictor
- The effect of BLP score is non-significant and positive
- Finally, the effect of Flanker Interference RT is statistically non-significant and positive

Subject choice RT models

Spanish
- word order OVS is statistically significant and negative, predicting faster RTs in choosing a subject
- The effect of LexTALE score is statistically significant and negative
- The effect of LexTALE-ESP score is statistically non-significant and negative
- when choosing a subject in Spanish, increased LexTALE score predicts faster RTs, but LexTALE-ESP scores do not
- The effect of BLP score is nearly statistically significant and positive
- The effect of Flanker Interference RT is statistically non-significant and positive
English
- The effect of word order OVS is statistically significant and predicted faster RTs
- The effect of LexTALE score is statistically significant and negative (Greater English vocabulary size leads to faster RTs in English sentences)
- The effect of LexTALE-ESP score is statistically non-significant and positive (LexTALE scores predict faster RTs in choosing a subject in English sentences, but LexTALE-ESP scores do not)
- The effect of BLP score is statistically significant and positive

Overall results

For Spanish sentences, participants read the OVS word order faster
increase in English proficiency leads to a decrease in reading time in English and near-significant decrease in reading time in Spansih, but this pattern was not expected in the Spanish condition
Many participants did not have formal education in their heritage language and might have been forced to rely on their English proficiency or literacy.
English proficiency (LexTALE) was a significant predictor for sentence reading time and choice selection time, but did not predict word choice
Inanimate-Inanimate sentence pose challenges for cue utilization and thematic role assignment due to semantic limitation
Prioritize semantic plausibility over syntactic representations in sentence processing and suggest a good-enough processing model
Additionally, our results demonstrate that when reading Spanish sentences and choosing subjects, the noncanonical word order predicted faster RTs, due to reliance on lexico-semantics as supported by vocabulary size in English, rather than due to the syntactic frame
Overall, these findings lend support to the good-enough processing model proposed