Processing Chinese formulaic sequences in sentence context: a comparative study of native and non-native speakers

Formulaic sequences (FSs), or fixed, unanalyzed multi-word chunks, are crucial for fluent and natural language use. They represent a significant portion of native speakers' communication and are recommended for L2 learners to achieve native-like fluency (Erman and Warren 2000; Ellis 2008). The inclusion of FSs also serves as a valuable indicator of L2 proficiency (Perera, 2001). Research on FSs spans linguistics and psychology, exploring their length, frequency, holistic nature, and cognitive processing (Dai et al. 2018; Siyanova-Chanturia 2015; Wray 2017; Carroll and Conklin 2020). Empirical studies suggest holistic processing of FSs, leading to faster processing times compared to non-formulaic sequences. This study examines how native (NS) and non-native (NNS) speakers of Chinese differ in their processing of FSs within sentence contexts, focusing on processing advantages, contextual effects, and processing similarities/differences across proficiency levels.

Formulaic sequences are defined as word strings processed holistically, stored and retrieved as complete units, without grammatical analysis of sub-parts (Wray 2002, 2008). Their formation is driven by frequent usage and contextual factors, resulting in preconstructed, indivisible units whose meaning cannot be deduced from individual components. Research on FS acquisition and processing initially focused on first language acquisition but has since expanded to second language acquisition (Wood 2015). Quantitative methods are often used to track L2 learners' knowledge of FSs (Chen 2019; Eskildsen 2015; Serrano et al. 2015; Toomer and Elgort 2019). Studies investigate factors influencing FS processing and acquisition, including FS types, linguistic features, L1-L2 differences, instruction duration, proficiency levels, and pedagogical approaches (Ding and Reynolds 2019; Kim and Nam 2017; Nguyen and Webb 2017; Pan et al. 2018; Pulido and Dussias 2020; Wolter and Yamashita 2018; Yeldham 2020; Yi, Lu and Ma 2017). Productive data (spoken/written language) directly reflects learners' understanding and use of FSs (Chan 2019; Chen 2020; Saito 2020; Tavakoli and Uchihara 2020; Xu 2018; Yan 2020). While research supports holistic processing advantages in NSs, findings for NNSs are inconsistent (Van Lancker 2012; Wray 2002, 2013). Studies using lexical decision tasks, self-paced reading, and reading aloud tasks show faster processing for FSs in NSs (Arnon 2010; Conklin and Schmitt 2008, 2012; Tremblay and Baayen 2010; Siyanova-Chanturia et al. 2011; Underwood et al. 2004; Vilkaitė 2016). However, NNS research lacks consensus on holistic processing (Conklin and Schmitt 2008; Ellis et al. 2008; Jeong and Jiang 2019; Jiang and Nekrasova 2007; Schmitt et al. 2004; Schmitt and Underwood 2004; Underwood et al. 2004). Linguistic context significantly aids word recognition (West et al. 1983; Becker 1980; Eisenberg and Becker, 1982; Grosjean 1980; Marslen-Wilson and Tyler 1980; Connine et al. 1991; Sereno et al. 2003; McDonald and Shillcock 2001; Cervera and Rosell 2015; Baayen et al. 2011; Zheng et al. 2016). This study addresses whether context affects FS processing, potentially enhancing L2 teaching strategies.

This study employed a self-paced masking experiment to investigate the processing of Chinese FSs within sentence contexts. Two groups participated: native speakers (NSs, n=20) and non-native speakers (NNSs, n=60) of Chinese, divided into three proficiency levels (elementary, intermediate, advanced, n=20 each) based on HSK scores. All NNSs completed a character recognition test before participation. Materials consisted of 40 sequences: 20 FSs and 20 matched non-FSs, selected from a corpus based on frequency and mutual information analyses. Non-FSs were created by substituting characters in the FSs, ensuring minimal differences in stroke count. Three NS linguists and five NS college students evaluated the materials. Each FS and non-FS was embedded in a grammatically correct sentence. A readability test ensured sentence appropriateness. The experiment used E-Prime software for word-by-word masking, recording response times. Participants answered comprehension questions after each sentence (accuracy >70% required). Contextual numerical information was calculated using a formula based on the Bayesian Law, incorporating the logarithm of quantitative sentence context. The 60 sentences (20 FSs, 20 non-FSs, and 20 control) were divided into two lists. Response time data were analyzed using two-way repeated-measures ANOVA and ANCOVA, considering participant group (4 levels) and stimulus set (2 levels). One-way ANOVA and ANCOVA were also conducted. Quantitative sentence context was a covariate in ANCOVA analyses.

Analyses without considering sentence context showed a significant interaction between participant group and stimulus set (F(3, 76) = 24.9, p < .001). Response times for FSs were significantly shorter than non-FSs (p < .001). NNSs showed longer processing times than NSs, decreasing with increasing proficiency. ANCOVA analyses, including sentence context as a covariate, showed similar significant interactions (F(3, 74) = 24.49, p < .001). Again, FSs were processed faster than non-FSs (p < .001), and NSs were faster than NNSs (p < .001). One-way ANOVA revealed shorter response times for FSs versus non-FSs across all groups: elementary (F(1, 19) = 125, p < .001), intermediate (F(1, 19) = 89.0, p < .001), advanced (F(1, 19) = 80.8, p < .001), and NSs (F(1, 19) = 49.0, p < 0.001). One-way ANCOVA showed significant differences in processing times for FSs and non-FSs in elementary (F(1, 37) = 7.19, p = 0.011) and intermediate (F(1, 37) = 6.19, p = 0.017) groups, while the advanced and NS groups showed marginal significance (p = 0.042 and 0.044, respectively). Table 1 presents the mean response times (ms) for each group and stimulus type.

The findings support the holistic processing hypothesis for FSs, as both NSs and NNSs processed FSs faster than non-FSs, regardless of context. However, context significantly influenced processing times, particularly for elementary and intermediate learners. The sentence context acted as a scaffold, aiding in recognition and processing. Advanced learners and NSs relied less on context due to their higher proficiency. This contrasts with McDonald and Shillcock (2001) and Cervera and Rosell (2015) who didn't focus on proficiency levels and experimental design differences, respectively. Contextual effects were stronger for FSs than non-FSs in NNSs. The study demonstrated that while the holistic processing of FSs is prevalent across proficiency levels, contextual information significantly assists L2 learners, particularly at lower proficiency levels.

The study confirms the holistic processing of Chinese FSs by both NSs and NNSs, with processing advantages for FSs evident across all groups and contexts. Contextual effects were significant, especially for lower-proficiency learners, highlighting the importance of context in L2 teaching. The results suggest that formulaic sequences can be used as a valuable tool in second language assessment and evaluation. Future research should explore linguistic characteristics of FSs and their acquisition processes and should also consider diverse L2 learners and languages.

The study focused solely on Chinese FSs and didn't include participants from multiple L2 backgrounds, limiting generalizability. The sample sizes, while substantial, could be further increased. The specific types of FSs used may also limit the generalizability of the findings. Future research might benefit from a broader range of FS types to confirm the findings.