Iconicity does not always facilitate complex sentence comprehension: Behavioral and pupillometric evidence from Mandarin–English bilingual adults

Shijie Zhang; Alissa Ferry; Silke Brandt; Emily Warren; Anna Theakston

doi:10.1017/S014271642610085X

Iconicity does not always facilitate complex sentence comprehension: Behavioral and pupillometric evidence from Mandarin–English bilingual adults

Published online by Cambridge University Press: 28 July 2026

Emily Warren and

Shijie Zhang: Affiliation:
Division of Psychology, Communication and Human Neuroscience, The University of Manchester, UK
Alissa Ferry: Affiliation:
Division of Psychology, Communication and Human Neuroscience, The University of Manchester, UK
Silke Brandt: Affiliation:
Linguistics and English Language, Lancaster University, UK
Emily Warren: Affiliation:
Division of Psychology, Communication and Human Neuroscience, The University of Manchester, UK
Anna Theakston*: Affiliation:
Division of Psychology, Communication and Human Neuroscience, The University of Manchester, UK
*: Corresponding author: Anna Theakston; Email: anna.theakston@manchester.ac.uk

Article contents

Abstract
Introduction
Experiment 1
Experiment 2
Discussion
Conclusions
Supplementary material
Replication package
Funding statement
Competing interests
Ethical standards
Footnotes
References

Rights & Permissions

Abstract

English adverbial sentences allow flexible clause ordering: main-adverbial and adverbial-main, and their comprehension by English monolingual speakers is influenced by iconicity. Iconic sentences, whose clause order reflects the chronological sequence of real-world events, are more easily comprehended (“We ate dinner before we went to see a movie”; “After we ate dinner, we went to see a movie”) than non-iconic sentences where the order of the clauses does not match the order of events in the real world (“Before we went to see a movie, we ate dinner”; “We went to see a movie after we ate dinner”). However, Mandarin speakers strongly prefer the adverbial-main order in spontaneous Mandarin speech. This paper examines whether iconicity represents a generally preferred strategy for interpreting English adverbial sentences or whether a speaker’s first language affects processing. In two studies, we tested the comprehension of English before- and after-sentences by Mandarin-English bilingual and English monolingual adults, using online pupillometry with offline accuracy in a sentence-picture matching task. We found that bilinguals, like monolinguals, comprehended after-sentences in the iconic adverbial-main order (also the preferred order in Mandarin) more accurately and with less effort than in the reverse order. However, unlike monolinguals, bilinguals did not comprehend before-sentences better in either the iconic main-adverbial order or the preferred adverbial-main order for Mandarin. Our results suggest that iconicity is not universal; instead, language-specific usage patterns, such as preferred clause order, compete with iconicity to influence processing of adverbial sentences.

Keywords

Complex syntax crosslinguistic influence iconicity language comprehension pupillometry

Information

Type: Original Article
Information: Applied Psycholinguistics , Volume 47 , 2026 , e44

DOI: https://doi.org/10.1017/S014271642610085X [Opens in a new window]
Creative Commons: This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use, distribution and reproduction, provided the original article is properly cited.
Copyright: © The Author(s), 2026. Published by Cambridge University Press

Introduction

While in the real world, time unfolds sequentially, with one event following another, language allows us to communicate events in different orders through temporal adverbial sentences. In English, temporal adverbial sentences can occur in two orders: main-adverbial (e.g., “We ate dinner before we went to see a movie”) and adverbial-main (e.g., “Before we went to see a movie, we ate dinner”). However, these two orders are not processed equally. For English monolingual children and adults, converging evidence suggests that the clause order that reflects the chronological sequence of real-world events is more easily processed (e.g., “We ate dinner before we went to see a movie”; “After we ate dinner, we went to see a movie”; De Ruiter et al., Reference De Ruiter, Theakston, Brandt and Lieven2018; Politzer-Ahles et al., Reference Politzer-Ahles, Xiang and Almeida2017). In addition to this iconic semantic mapping, the cognitive load required to process each clause order has been proposed to differ, such that main-adverbial orders require less cognitive load and are therefore easier to process than adverbial-main orders overall (e.g., Diessel, Reference Diessel2005). The extent to which iconicity and clause order determine ease of sentence processing is also affected by the wider discourse context and associated pragmatic considerations (e.g., De Ruiter et al., Reference De Ruiter, Lieven, Brandt and Theakston2020, Reference De Ruiter, Lemen, Lieven, Brandt and Theakston2021), though these factors are beyond the scope of the current paper.

Despite the relatively well-understood (complex) sentence processing mechanisms employed by English monolinguals, it remains largely unexplored whether these processing mechanisms universally impact individuals from different language backgrounds, such as Mandarin-English bilinguals. In Mandarin Chinese, there is a strong preference for the temporal adverbial clause to precede the main clause (i.e., adverbial-main order), as illustrated in examples (1) and (2). Corpus studies on Mandarin show that 86.5%–97.0% of temporal adverbial sentences, such as (dang/zai)…de shihou “when,” yiqian/zhiqian “before,” and yihou/zhihou “after,” appear in adverbial-main order in spontaneous speech (Wang, Reference Wang2002; Yeh, Reference Yeh, GIkeya and Kawamori2000). When only considering yiqian/zhiqian “before” and yihou/zhihou “after,” this increases to 100.0%, although the corpus size may be small (Yeh, Reference Yeh, GIkeya and Kawamori2000). By contrast, in English, 41.5% of temporal adverbial sentences appear in adverbial-main order in spontaneous speech (Diessel, Reference Diessel2005), but when only considering before- and after-sentences, the mean percentage in adverbial-main order is only 32.8%–39.8% (De Ruiter et al., Reference De Ruiter, Lemen, Lieven, Brandt and Theakston2021; Diessel, Reference Diessel2005). Furthermore, English shows a preference for an iconic clause order, with a mean of 71.3%–73.8% of before- and after-sentences exhibiting iconic order (De Ruiter et al., Reference De Ruiter, Lemen, Lieven, Brandt and Theakston2021; Diessel, Reference Diessel2005). This difference in clause order preference between Mandarin and English may lead Mandarin-English bilinguals to use different strategies when constructing mental representations of events described in English, compared to English monolinguals.

(1)	[Women	chi-wan	wanfan	zhihou]_ADVERBIAL,	[qu	kan-le	dianying]_MAIN
	we	eat-finish	dinner	after	go	see-ASP	movie
	“After we ate dinner, (we) went to see a movie.”

(2)	[Women	qu	kan	dianying	zhiqian]_ADVERBIAL,	[chi-le	wanfan]_MAIN
	we	go	see	movie	before	eat-ASP	dinner
	“Before we went to see a movie, (we) ate dinner.”

The present studies investigate the comprehension of English temporal before- and after-sentences by Mandarin-English bilingual and English monolingual adults. We explore whether the processing strategies employed by English monolinguals uniformly affect language processing or whether language-specific usage patterns in Mandarin influence Mandarin-English bilinguals’ sentence processing, showing crosslinguistic influence. In the following sections, we review general theoretical accounts of (complex) sentence processing and the empirical evidence from English and then discuss how these accounts apply to Mandarin and the potential crosslinguistic influence from Mandarin to English.

Theoretical accounts

Semantic account

The semantic principle of iconicity has been proposed to explain sentence processing (Clark, Reference Clark1971; Croft, Reference Croft2003; Diessel, Reference Diessel2008; Givón, Reference Givón and Haiman1985). The basic idea is that “the structure of language reflects in some way the structure of experience” (Croft, Reference Croft2003, p. 102). The more closely the structure of language mirrors our real-world experience, the easier it is for us to communicate (Givón, Reference Givón and Haiman1985). In the case of adverbial sentences, those in which the clause order reflects the chronological sequence of real-world events are expected to be more easily processed than those that do not. For example, in examples (4) and (5) in Table 1, the event “she hoovers the house,” which is mentioned first in the sentence, also occurs first in the real world—these are iconic sentences. In contrast, in examples (3) and (6), the order of events in the sentence is reversed relative to the real-world order—these are non-iconic. Accordingly, iconic sentences are expected to be processed more easily than non-iconic ones.

Table 1.

The predictions from theoretical accounts on the comprehension of temporal adverbial sentences

A table comparing sentence structures and their processing accounts.

Table 1. Long description

A table with four rows and five columns. The columns are labeled Connective, Clause order, Sentence, Semantic account, Processing-based account, and Crosslinguistic influence and usage patterns. The rows are labeled with different sentence structures involving the connectives Before and After. Each row provides examples of sentences with different clause orders and indicates whether they align with semantic, processing-based, and crosslinguistic accounts. Row 1: Connective Before, Clause order Adverbial-main, Sentence Before she paints the old fence, she hoovers the house, Crosslinguistic influence and usage patterns checkmark. Row 2: Connective Before, Clause order Main-adverbial, Sentence She hoovers the house before she paints the old fence, Semantic account checkmark, Processing-based account checkmark. Row 3: Connective After, Clause order Adverbial-main, Sentence After she hoovers the house, she paints the old fence, Semantic account checkmark, Crosslinguistic influence and usage patterns checkmark. Row 4: Connective After, Clause order Main-adverbial, Sentence She paints the old fence after she hoovers the house, Processing-based account checkmark.

Note: Tick mark refers to the clause order that the account would predict to be easiest to comprehend.

The results from child experimental studies support the effect of iconicity on English monolingual children’s comprehension of adverbial sentences (e.g., Blything & Cain, Reference Blything and Cain2016; Blything et al., Reference Blything, Davies and Cain2015; De Ruiter et al., Reference De Ruiter, Theakston, Brandt and Lieven2018, Reference De Ruiter, Lieven, Brandt and Theakston2020; Wagner & Holt, Reference Wagner and Holt2025; Zhang et al., Reference Zhang, Brandt, Warren and Theakstonsubmitted). For example, De Ruiter and colleagues (2018) examined English monolingual children’s comprehension of temporal and causal/conditional adverbial sentences (before, after, because, if) using a sentence-picture matching task and found that five-year-olds comprehended all types of sentences more accurately when the sentences were iconic rather than non-iconic.

However, the strength of the iconicity effect may vary across different connectives. In De Ruiter et al. (Reference De Ruiter, Theakston, Brandt and Lieven2018), the iconicity effect was stronger for after-sentences than for before-sentences, with before-sentences being comprehended more accurately overall. A similar pattern was observed in Zhang et al. Reference Zhang, Brandt, Warren and Theakston(submitted) for English monolinguals aged four to five and seven to nine. Likewise, Wagner and Holt (Reference Wagner and Holt2025) found that English monolinguals aged seven to twelve exhibited the iconicity effect for after-sentences, but not for before-sentences.Footnote ¹ ^, Footnote ² These studies suggest that the varied strength of the iconicity effect for different connectives could be related to their semantic differences. Before-sentences have a more consistent form-meaning mapping when used as a temporal connective, supporting more accurate comprehension overall. In contrast, “after” often appears in other constructions such as “look after” and “after all” (De Ruiter et al., Reference De Ruiter, Lemen, Lieven, Brandt and Theakston2021; Wagner & Holt, Reference Wagner and Holt2025), which may weaken its lexical representation in temporal adverbial sentence comprehension (see Ambridge et al. (Reference Ambridge, Kidd, Rowland and Theakston2015) and Theakston (Reference Theakston, Rowland, Theakston, Ambridge and Twomey2020) for an overview of usage/frequency-based arguments). As a result, when processing after-sentences, children rely more heavily on the non-optimal semantic strategy of iconicity to map the order of events in the sentence directly onto the sequence of real-world events (Zhang et al., Reference Zhang, Brandt, Warren and Theakstonsubmitted).

Regarding English adults’ comprehension of adverbial sentences, empirical evidence also supports the effect of iconicity, but differences between connectives are not consistently reported. For example, Politzer-Ahles et al. (Reference Politzer-Ahles, Xiang and Almeida2017) examined native English adults’ online comprehension of temporal before- and after-sentences. In this study, adults read a sentence word by word while their electroencephalographic (EEG) activity was recorded to measure an anterior event-related potential (ERP) component, where greater negativity is associated with increased processing costs (e.g., Vos et al., Reference Vos, Gunter, Kolk and Mulder2001). The results showed that, in adverbial-main orders, before-sentences (non-iconic) elicited more negative ERPs than after-sentences (iconic), whereas in main-adverbial orders, after-sentences (non-iconic) elicited more negative ERPs than before-sentences (iconic), indicating greater processing costs related to non-iconic sentences (see also Münte et al. (Reference Münte, Schiltz and Kutas1998) using ERPs, and Ye et al. (Reference Ye, Kutas, George, Sereno, Ling and Münte2012) using functional magnetic resonance imaging (fMRI)). However, unlike in child comprehension, Politzer-Ahles et al. (Reference Politzer-Ahles, Xiang and Almeida2017) did not find an overall difference between before- and after-sentences in adults’ processing. This could be related to task design and the test stimuli used (see Zhang et al. (Reference Zhang, Brandt and Theakston2025) for discussion), or it could be that the differences in the lexical representations of “before” and “after” are less pronounced in adults due to their richer lexical and semantic knowledge.

Processing-based account

The ordering of the main and adverbial clause alone has also been proposed to affect the processing of adverbial sentences (Diessel, Reference Diessel2005; see also Hawkins, Reference Hawkins1990, Reference Hawkins and Jacobs1992, Reference Hawkins1994). Main-adverbial orders have been suggested to be easier to process than adverbial-main orders. Compare examples (4) and (6) with examples (3) and (5) in Table 1. This is because, when processing main-adverbial orders, the listener can process the main clause and the adverbial clause one after the other, building the link between them only when the adverbial clause is encountered. In contrast, when processing adverbial-main orders, on hearing the adverbial the listener immediately recognizes that the sentence contains at least two clauses and must keep the adverbial clause in memory until the main clause is encountered, which increases processing load.

However, previous child and adult comprehension studies did not find that one clause order was generally comprehended better than the other (e.g., De Ruiter et al., Reference De Ruiter, Theakston, Brandt and Lieven2018; Politzer-Ahles et al., Reference Politzer-Ahles, Xiang and Almeida2017). Instead, as mentioned above, the effect of iconicity (i.e., an interaction between clause order and connective type) was predominantly reported: before-sentences were better comprehended in the iconic main-adverbial order, and after-sentences were better comprehended in the iconic adverbial-main order (see Table 1 for examples). Only Zhang et al. (Reference Zhang, Junge, Lieven, Brandt and Theakston2023), who examined the production of when-sentences (e.g., “When Pig is swimming, Little Panda is drinking very fast”) by three- to five-year-old children and adults, offer some support for this processing-based hypothesis. They found that three-year-olds tended to reverse sentences in adverbial-main order into main-adverbial order in their repetitions, but this tendency decreased with age. This led the authors to suggest that younger children, due to their lower processing capacities, tend to prefer main-adverbial orders. However, the simultaneous clauses with “when” in Zhang et al. (Reference Zhang, Junge, Lieven, Brandt and Theakston2023) differ from before- and after-sentences in that they do not encode a clear order of events, thereby limiting the study’s ability to investigate the effect of iconicity, and the test sentences were preceded by a discourse context which has also been found to affect sentence processing (e.g., De Ruiter et al., Reference De Ruiter, Lieven, Brandt and Theakston2020).

Clause order in Mandarin adverbial sentences and potential crosslinguistic influence

As in sentence comprehension, English monolingual adults tend to be influenced by iconicity when ordering main and temporal adverbial clauses in spontaneous English speech (71.3%–73.8% of adult before- and after-sentences are produced in iconic order; De Ruiter et al., Reference De Ruiter, Lemen, Lieven, Brandt and Theakston2021; Diessel, Reference Diessel2005). In contrast, Mandarin Chinese speakers show a strong preference for the temporal adverbial clause to precede the main clause (i.e., adverbial-main order) in spontaneous Mandarin speech (86.5%–97.0% of temporal adverbial sentences, increasing to 100% for before-/after-sentences only; Wang, Reference Wang2002; Yeh, Reference Yeh, GIkeya and Kawamori2000).

The preference for the adverbial-main order in Mandarin can be linked to its information structure. Mandarin is widely recognized as a topic-comment language where the topic typically occupies the sentence-initial position (e.g., Chao, Reference Chao1968; Li & Thompson, Reference Li and Thompson1981). Despite some differences in the precise definition of “topic,” it is commonly assumed to encode given or shared information (e.g., Halliday & McDonald, Reference Halliday, McDonald, Caffarel, Martin and Matthiessen2004; Li & Thompson, Reference Li and Thompson1981). This aligns with the functional role of initial temporal adverbial clauses, which serve to provide pragmatically presupposed information and establish the thematic ground for subsequent clauses (Diessel, Reference Diessel2001, Reference Diessel2005).

Furthermore, this preference for the adverbial-main order in Mandarin does not necessarily contradict processing-based explanations (e.g., Diessel, Reference Diessel2005). While processing-based accounts suggest that adverbial-main orders in English impose higher processing load (relative to main-adverbial orders), Mandarin avoids this through its clause-final connective structure (see Diessel (Reference Diessel2001) for more details on typological differences in adverbial sentences across languages). Specifically, English connectives occur in the initial position of adverbial clauses, immediately signaling the sentence type and requiring the listener to anticipate a subsequent main clause. However, Mandarin connectives occur at the end of adverbial clauses. As shown in examples (1) and (2), the listener does not necessarily identify the adverbial clause until the connective is encountered at the end; thus, the link between the two clauses is established only at that point. Due to this structural difference, the processing load for adverbial-main orders in Mandarin does not mirror that of English, also providing a processing-based justification for its prevalence in spontaneous speech.

However, Mandarin’s strong preference for the adverbial-main order limits the ability to fully test the iconicity effect. Chen et al. (Reference Chen, Lu and & Xu2022) provided ERP evidence for iconicity in Mandarin, showing that Mandarin adults exhibited greater processing costs for before-sentences than for after-sentences. This is because, although both types of test sentences followed the adverbial-main order, in before-sentences this order is non-iconic, whereas in after-sentences it is iconic. Yet, these findings offer only a partial view. If the iconicity effect is truly universal, one would expect an advantage of iconic after-sentences over non-iconic after-sentences, and an advantage of iconic before-sentences over non-iconic before-sentences. Such an investigation, however, is not feasible in Mandarin due to its relatively rigid word order in temporal adverbial sentences.

Examining how Mandarin-English bilinguals comprehend English adverbial sentences, which allow for both adverbial-main and main-adverbial orders, provides an ideal test of whether the iconicity effect observed in English monolinguals is universal, or modulated by Mandarin-specific usage patterns, leading to crosslinguistic influence (e.g., Hulk & Müller, Reference Hulk and Müller2000; Müller & Hulk, Reference Müller and Hulk2001). If Mandarin usage patterns influence Mandarin-English bilinguals’ processing of English, we would expect the adverbial-main order that dominates in Mandarin to facilitate their processing of the same order in English, whereas the dispreferred main-adverbial order may pose a greater difficulty. Put another way, based on the crosslinguistic influence and usage patterns, Mandarin-English bilingual adults should comprehend adverbial-main orders better than main-adverbial orders in English.

Note that for after-sentences, the predictions of the crosslinguistic influence and usage patterns and of iconicity converge: the dominant adverbial-main order in Mandarin is also the iconic order for after-sentences. Therefore, according to both hypotheses, Mandarin-English bilinguals should comprehend after-sentences in the iconic adverbial-main order better than in the non-iconic main-adverbial order, similar to what has been found in English monolinguals (Münte et al., Reference Münte, Schiltz and Kutas1998; Politzer-Ahles et al., Reference Politzer-Ahles, Xiang and Almeida2017; Ye et al., Reference Ye, Kutas, George, Sereno, Ling and Münte2012). However, the crosslinguistic influence and usage patterns conflict with iconicity for before-sentences: according to the former, the adverbial-main order should be better comprehended, whereas according to the latter, the iconic main-adverbial order should be better comprehended (see Table 1 for examples).

To our knowledge, no studies have examined Mandarin-English bilinguals’ processing of English temporal adverbial sentences. However, several corpus-based studies have investigated the ordering distribution of English temporal adverbial sentences in the writing of Chinese learners of English. For example, Li and Long (Reference Li and Long2022) examined English temporal adverbial sentences (e.g., when, before, after, until) in the writing of Chinese learners of English, comparing it to native English and native Chinese written texts. They found that Chinese learners of English tended to use adverbial-main orders. However, as English language proficiency increased, the proportion of adverbial-main orders decreased—from 78.0% in beginners to 56.5% in advanced learners. The proportion of adverbial-main orders in beginners was similar to the proportion found in native Chinese texts (96.5%), while the proportion in advanced learners was closer to the proportion in native English essays (36.0%). Therefore, Li and Long’s (Reference Li and Long2022) results indicate a clear crosslinguistic influence from Chinese to English, with this influence decreasing as English language proficiency increases. However, they did not analyze different types of temporal adverbial sentences separately, so it remains unclear whether there were differences between after- and before-sentences (and whether Chinese learners of English would also show an effect of iconicity).

Fang (Reference Fang2009) also found a general preference for adverbial-main orders in the writing of Chinese learners of English. However, when examining after- and before-sentences separately, they found that only after-sentences occurred more frequently in the adverbial-main order (iconic, 65.98%). Before-sentences more often appeared in the main-adverbial order (iconic, 64.37%). This suggests an effect of iconicity in Chinese learners’ English writing, despite crosslinguistic influence. However, they did not account for the Chinese learners’ English proficiency.

To summarize, previous studies have reported crosslinguistic influence on Chinese learners’ use of English temporal adverbial sentences, but the results remain inconclusive. Moreover, these studies focused exclusively on written data, where individuals have more time to plan their sentences than in spoken language. In two studies, we examined Mandarin-English bilinguals’ comprehension of spoken English temporal adverbial sentences both offline and online and compared this to that of English monolinguals. Our goal was to disentangle whether Mandarin-English bilinguals employ the same processing strategies (e.g., iconicity) as English monolinguals or whether their processing is affected by language-specific usage patterns in Mandarin.

The present studies

The present studies investigated Mandarin-English bilingual and English monolingual adults’ comprehension of sentences containing two types of connectives (“after” and “before”) in two clause orders (main-adverbial and adverbial-main), using an offline sentence-picture matching task (Experiment 1) and an online version of the task employing pupillometry (Experiment 2). Additionally, we investigated whether Mandarin-English bilingual adults’ comprehension of temporal adverbial sentences was affected by their English language proficiency. Based on theoretical predictions and previous empirical evidence, the following hypotheses were formulated in relation to behavior accuracy, reaction times (RTs), and changes in pupil size:

(i) Semantic account of iconicity: Iconic clause orders are comprehended better (i.e., with higher accuracy, faster RTs, and smaller changes in pupil size) than non-iconic clause orders (e.g., Clark, Reference Clark1971; Croft, Reference Croft2003; Diessel, Reference Diessel2008; Givón, Reference Givón and Haiman1985).
(ii) Semantic consistency: Before-sentences are comprehended better than after-sentences (e.g., Ambridge et al, Reference Ambridge, Kidd, Rowland and Theakston2015; Theakston, Reference Theakston, Rowland, Theakston, Ambridge and Twomey2020).
(iii) Processing-based account: Main-adverbial clause orders are comprehended better than adverbial-main clause orders (e.g., Diessel, Reference Diessel2005).
(iv) Crosslinguistic influence and usage patterns: Adverbial-main clause orders are comprehended better than main-adverbial clause orders by Mandarin-English bilingual adults (e.g., Hulk & Müller, Reference Hulk and Müller2000).
(v) Individual differences: The higher the English proficiency of Mandarin-English bilingual adults, the less crosslinguistic influence from Mandarin on their English sentence comprehension (Li & Long, Reference Li and Long2022).

Experiment 1

Participants

A total of 46 Mandarin-English bilingual and 40 English monolingual adults were included in the final sample.Footnote ³ The bilinguals (age range = 18–38 years, mean age = 22.74 years, SD = 3.73 years, 39 females) were recruited through social media and the University of Manchester’s credit system. They all spoke both Mandarin and English, with Mandarin as their native language, and were living in the UK at the time of testing. They were either pursuing or had obtained university degrees: undergraduate (N = 20), master’s (N = 21), and PhD (N = 5). Most bilinguals (N = 38) had taken the International English Language Testing System (IELTS), and we used their IELTS scores as a proxy for their English language proficiency in this experiment. The remaining bilinguals (N = 8) either did not have any standardized English test scores or had scores from other English tests (e.g., the Chinese College English test), which could not be directly compared.

The English monolingual adults (age range = 18–30 years, mean age = 19.35 years, SD = 2.01 years, 35 females) were all recruited through the University of Manchester’s credit system, and all were undergraduates. Neither the monolingual nor the bilingual participants had any known speech and language difficulties, nor any vision or hearing impairments. In addition to the 46 bilinguals and 40 monolinguals in the final sample, two additional bilinguals were tested, but their data had to be excluded due to experimenter error.

Offline sentence-picture matching task

The participants’ comprehension of adverbial sentences was tested using an offline sentence-picture matching task. In the task, as shown in Table 2, the participant first listened to an audio-recorded adverbial sentence that ended with a beep sound while viewing a blank screen. Immediately after the beep, two picture sequences were displayed at the top and bottom of the screen, respectively, and the participant was asked to select the picture sequence that matched the sentence they had heard. Both picture sequences depicted the same pair of events described in the sentence but in different orders: one matched the order of events in the sentence, and the other presented them in reverse. Within each picture sequence, the pictures followed a left-to-right reading order: the left picture represented the event that happened first, and the right picture represented the event that happened second, a convention shared by native English and Mandarin speakers. RTs were measured from the offset of the beep sound to the participant’s click on a picture sequence.

Table 2.

Structure of the experimental trials

Diagram of an experimental trial structure for a sentence-picture matching task.

Table 2. Long description

Panel A: A blank screen with text indicating the start of the picture sequence. Panel B: Two picture sequences displayed at the top and bottom of the screen. Each sequence contains two pictures showing different orders of events. The left picture in each sequence represents the first event, and the right picture represents the second event. The participant is asked to select the picture sequence that matches the sentence they heard.

The task was presented using the experiment builder Gorilla.sc. The experimenter provided participants with the Gorilla link to access the task and observed them as they completed it, either in person in a lab at the University of Manchester or online via Zoom. The task took between 15 and 20 minutes to complete.

Design

The offline sentence-picture matching task had three factors: one between-subjects factor (LanguageGroup), and two within-subject factors (ConnectiveType, ClauseOrder), each with the following levels:

• LanguageGroup: bilingual, monolingual
• ConnectiveType: before, after
• ClauseOrder: main-adv (main-adverbial order), adv-main (adverbial-main order)

Materials

A total of 24 test sentences were constructed, with six test sentences per condition (four conditions: two ConnectiveType × two ClauseOrder). Each sentence consisted of a main and an adverbial clause, describing two events performed by a single actor (either a boy or a girl). To create these sentences, we used 12 event pairs (e.g., the “paint the old fence” and “hoover the house” pair), with each pair being connected once with “after” and once with “before.” Among the after-sentences (N = 12), half followed a main-adv order, while the other half followed an adv-main order. The same distribution applied to the before-sentences (N = 12). A full list of test sentences can be found in Table S1 in the Supplemental Material.

The pair of events in each test sentence was arbitrarily related; the events could happen in either order, with only the connective “before/after” indicating the correct order. The subject in each test sentence was pronominal (“he” in half of the sentences and “she” in the other half), and the objects of the transitive verbs were always inanimate. Each event was described using a different verb in the present tense, and each event description was five syllables long.

Four different experimental lists were constructed. Each list consisted of two sessions, with the test sentence containing the same event pair appearing separately in each session (see examples (7) and (8)). The order of the test sentences within each session was pseudo-randomized, allowing a maximum of two consecutive test sentences in the same condition. The position of the correct picture sequence in each session was counterbalanced: in half of the trials, the correct picture sequence was at the top and in the other half of the trials at the bottom. List 2 was the same as List 1, with the difference that all main-adv sentences were turned into adv-main sentences and vice versa (see examples (9) and (10)). Lists 3 and 4 were created by swapping session 1 and session 2 of Lists 1 and 2. Participants were randomly assigned to one of the four experimental lists.

(7)	List 1 Session 1: After she paints the old fence, she hoovers the house.
(8)	List 1 Session 2: Before she paints the old fence, she hoovers the house.
(9)	List 2 Session 1: She hoovers the house after she paints the old fence.
(10)	List 2 Session 2: She hoovers the house before she paints the old fence.

In addition to the 24 test sentences, four warm-up trials were provided before the test to familiarize participants with the task. The first two warm-up trials were simple sentences (e.g., “Seal is catching a ball”) with only two pictures (e.g., a seal catching a ball vs. a seal opening a door). The next two warm-up trials resembled the test trials (i.e., two picture sequences, e.g., water his plant + switch the light on vs. switch the light + water his plant), except the sentences followed a “First…, then…” structure (e.g., “First he switches the light on, then he waters his plant”). In addition, seven filler trials were interspersed, one after every three test trials. They were also simple sentences, like the first two warm-up trials. All sentences were audio-recorded by a young female native speaker of British English, and all audio and visual stimuli were borrowed from De Ruiter et al. (Reference De Ruiter, Theakston, Brandt and Lieven2018).

Data analysis

We analyzed participants’ behavioral accuracy and RTs statistically, one after another, followed by simple correlation analyses between bilingual participants’ accuracy and RT performance in sentence comprehension and their IELTS scores.

Behavioral accuracy

In total, we collected 1104 responses from bilinguals, of which 955 responses were correct and 149 were incorrect, achieving an accuracy of 86.50%. We collected 960 responses from monolinguals, of which 898 responses were correct and 62 were incorrect, achieving an accuracy of 93.54%. Figure 1 shows the proportion of correct and incorrect responses for both language groups across the four conditions (two types of adverbial sentences and two clause orders). It reveals an apparent three-way interaction between ClauseOrder and Type across the two language groups, which we analyzed statistically to examine its significance.

Figure 1.

Proportion of correct responses for bilinguals’ and monolinguals’ comprehension of adverbial sentences (Experiment 1).

Note: The asterisk above the comparison of two orders for one connective indicates a significant difference.

Bar graph comparing accuracy of bilingual and monolingual comprehension of adverbial sentences.

Statistical analysis was carried out using generalized linear mixed-effects models (GLMMs; Baayen et al., Reference Baayen, Davidson and Bates2008) with the lme4 package (Bates et al., Reference Bates, Mächler, Bolker and Walker2015) in R, version 4.5.2. GLMMs are well-suited for analyzing repeated measures data in psycholinguistics, because they account for both the fixed effects of experimental conditions and the random variation across participants and test items (Baayen et al., Reference Baayen, Davidson and Bates2008). The fixed effects included ClauseOrder (sum-coded: main-adv = 0.5, adv-main = −0.5), Type (sum-coded: before = 0.5, after = −0.5), and LanguageGroup (sum-coded: bilingual = 0.5, monolingual = −0.5). Random intercepts were included for participants and items to account for individual and item-level variability. Model building followed a stepwise forward selection procedure. First, the contribution of each fixed effect was tested by comparing it individually to the null model containing only random intercepts using the ANOVA function. The fixed effects that significantly improved model fit were retained for the next stage. Next, two-way interactions (ClauseOrder × Type, ClauseOrder × LanguageGroup, Type × LanguageGroup) were added one at a time to the model containing the significant fixed effects. The two-way interactions that significantly improved model fit (along with the included fixed effects) were retained. Finally, the model containing the significant two-way interactions was compared to the model containing the three-way interaction (ClauseOrder × Type × LanguageGroup). Model comparison indicated that the three-way interaction significantly improved model fit, χ² (3) = 10.44, p = 0.015, and was therefore retained in the final model (see Table 3). By-participant random slopes were not included, as they did not improve the fit of the final model. We interpreted the three-way interaction using post hoc comparisons using the emmeans package (Lenth et al., Reference Lenth, Bolker, Buerkner, Giné-Vázquez, Herve, Jung, Love, Miguez, Riebl and Singmann2023) with Bonferroni correction.

Table 3.

Final model for bilinguals’ and monolinguals’ comprehension of adverbial sentences in terms of behavioral accuracy (Experiment 1)

A table summarizing the results of a generalized linear mixed-effects model analysis on behavioral accuracy in bilinguals and monolinguals.

Table 3. Long description

The table presents the results of a generalized linear mixed-effects model analysis. It includes eight rows and five columns. The columns are labeled as follows: Beta, SE(Beta), z, and p. The rows are labeled as follows: Intercept, LanguageGroup [bilingual], ClauseOrder [main-adv], Type [before], ClauseOrder [main-adv]: Type [before], LanguageGroup [bilingual]: ClauseOrder [main-adv], LanguageGroup [bilingual]: Type [before], and LanguageGroup [bilingual]: ClauseOrder [main-adv]: Type [before]. The values in the table are as follows: Row 1: Beta, 3.03; SE(Beta), 0.28; z, 10.62; p, <0.001. Row 2: Beta, -1.21; SE(Beta), 0.32; z, -3.77; p, <0.001. Row 3: Beta, -0.40; SE(Beta), 0.19; z, -2.13; p, 0.033. Row 4: Beta, -0.32; SE(Beta), 0.18; z, -1.73; p, 0.084. Row 5: Beta, 1.73; SE(Beta), 0.37; z, 4.65; p, <0.001. Row 6: Beta, 0.11; SE(Beta), 0.37; z, 0.30; p, 0.767. Row 7: Beta, -0.05; SE(Beta), 0.37; z, -0.13; p, 0.900. Row 8: Beta, -2.36; SE(Beta), 0.74; z, -3.18; p, 0.001.

Note: Omitted level: LanguageGroup = monolingual; ClauseOrder = adv-main; Type = after. Number of observations = 2064. Significant effects and interactions are highlighted in bold.

As shown in Figure 1 and supported by post hoc analyses in Table S2 (Supplemental Material), monolinguals comprehended before-sentences more accurately when they were in the iconic main-adv order than in the non-iconic adv-main order, and they comprehended after-sentences more accurately when they were in the iconic adv-main order than in the non-iconic main-adv order. Further support for this pattern is seen in main-adv orders where before-sentences were comprehended more accurately than after-sentences and vice versa in adv-main orders. These results suggest that monolinguals were significantly more accurate at comprehending temporal adverbial sentences in iconic clause order. However, they did not show a consistent advantage for main-adv orders or for before- over after-sentences.

Bilinguals also comprehended after-sentences more accurately when they were in the iconic adv-main order (which is also the preferred order in Mandarin). However, they comprehended before-sentences equally, regardless of whether they were in main-adv or adv-main orders thus showing no consistent preference for either order. Furthermore, in main-adv orders before-sentences were comprehended similarly to after-sentences, while in adv-main orders after-sentences were comprehended more accurately than before-sentences. This indicates that, unlike monolinguals, bilinguals showed the effect of iconicity and/or clause order only in after-sentences, but not in before-sentences.

Behavioral reaction times

For the analyses of RTs, only the correct responses were included. After inspecting the correct responses, we removed the responses with RTs that exceeded three standard deviations above or below the mean RT. In total, 941 out of 955 correct responses from bilinguals (excluded 14 outliers) and 883 out of 898 responses from monolinguals (excluded 15 outliers) were included in the statistical analysis. The mean RT was 2746.75 ms (SD = 1742.42 ms) for bilinguals and 1505.70 ms (SD = 761.12 ms) for monolinguals. Figure 2 shows that monolinguals comprehended adverbial sentences more quickly than bilinguals in general. Furthermore, there seems to be an interaction between ClauseOrder and Type for bilinguals, but not for monolinguals, which we analyzed statistically to examine its significance.

Figure 2.

RTs for bilinguals’ and monolinguals’ comprehension of adverbial sentences (Experiment 1).

Bar graph comparing reaction times for bilinguals and monolinguals in understanding adverbial sentences.

We analyzed the RTs using the same strategy as for accuracy. Model comparison indicated that the interaction between LanguageGroup and Type significantly improved model fit, χ² (1) = 5.17, p = 0.023, and was therefore retained in the final model (see Table 4). As shown in Figure 2 and supported by post hoc analyses in Table S3, monolinguals comprehended both types of adverbial sentences significantly more quickly than bilinguals. Moreover, monolinguals comprehended before-sentences as quickly as after-sentences, whereas bilinguals comprehended before-sentences significantly more slowly than after-sentences. Figure 2 further shows that bilinguals’ slower comprehension of before-sentences was likely driven by the fact that before-sentences in the iconic main-adv order were comprehended the slowest compared to other conditions. However, the three-way interaction between ClauseOrder, Type, and LanguageGroup did not improve model fit, χ² (4) = 7.23, p = 0.125, so was not retained in the final model.

Table 4.

Final model for bilinguals’ and monolinguals’ comprehension of adverbial sentences in terms of behavioral reaction times (Experiment 1)

A table comparing the effects of language group and sentence type on reaction times in a comprehension experiment.

Table 4. Long description

A table with five rows and four columns comparing the effects of language group and sentence type on reaction times in a comprehension experiment. The columns are labeled Beta, SE(Beta), t, and p. The rows are labeled Intercept, LanguageGroup [bilingual], Type [before], and LanguageGroup [bilingual]: Type [before]. Row 1: Intercept, Beta 2204.55, SE(Beta) 137.99, t 15.98, p less than 0.001. Row 2: LanguageGroup [bilingual], Beta 1318.14, SE(Beta) 183.88, t 7.17, p less than 0.001. Row 3: Type [before], Beta 136.25, SE(Beta) 49.96, t 2.73, p 0.006. Row 4: LanguageGroup [bilingual]: Type [before], Beta 227.32, SE(Beta) 99.92, t 2.28, p 0.023.

Note. Omitted level: LanguageGroup = monolingual; Type = after. Number of observations = 1824. Significant effects and interaction are highlighted in bold.

To conclude, in Experiment 1 using the offline sentence-picture matching task, monolinguals comprehended iconic before- and after-sentences more accurately than non-iconic ones but showed no differences in their comprehension of these sentences in terms of RTs. Bilinguals also showed the effect of iconicity and/or clause order in after-sentences in terms of accuracy, but not in before-sentences. Instead, before-sentences were comprehended more slowly than after-sentences, largely driven by sentences in the iconic main-adv order (which is the dispreferred order in Mandarin). Neither group showed a consistent advantage for either clause order, nor an advantage for before- over after-sentences.

Bilinguals’ individual differences

We then ran simple correlations to examine whether the bilingual participants’ performance in the offline sentence-picture matching task (i.e., mean accuracy and mean RTs) correlated with their IELTS scores across the board and separately for specific sentence types (as we observed differences in accuracy and RTs according to sentence type). The eight bilinguals who did not have IELTS scores were excluded from the individual differences analyses, leaving 38 bilinguals.

The results showed that bilinguals’ IELTS scores marginally significantly correlated with their overall mean accuracy, r(36) = 0.31, p = 0.054, and significantly correlated with their overall mean RTs, r(36) = −0.59, p < 0.001. Furthermore, as shown in Table 5, bilinguals’ IELTS scores significantly correlated with the mean RTs of all sentence types and with the accuracy of after-sentences in the main-adv order. These results suggest that bilinguals with better English proficiency were more accurate and quicker in comprehending both before- and after-sentences overall compared to those with lower proficiency, and this correlation was especially pronounced in the accuracy of after-sentences in the main-adv order (which is the non-iconic and dispreferred order in Mandarin).

Table 5.

Correlations between bilinguals’ IELTS scores and their mean accuracy and reaction times in the offline sentence-picture matching task (Experiment 1)

A table showing correlations between bilinguals’ IELTS scores and their mean accuracy and reaction times.

Table 5. Long description

A table with four rows and three columns. The columns are labeled Sentences, Mean accuracy, and Mean reaction times. The rows are labeled main-after, after-main, main-before, and before-main. The values in the table are as follows: Row 1: main-after, 0.46**, -0.44**. Row 2: after-main, 0.02, -0.77***. Row 3: main-before, 0.23, -0.40*. Row 4: before-main, 0.07, -0.44**. The table shows the correlations between bilinguals’ IELTS scores and their mean accuracy and reaction times for different sentence types.

*p < 0.05. **p < 0.01. ***p < 0.001.

However, these individual difference results should be interpreted with caution, as we relied on bilinguals’ self-reported IELTS scores as a proxy for their English language proficiency. The timing of when bilinguals took the IELTS test likely varied and was not formally recorded, so their scores may not accurately reflect their current English proficiency. Moreover, the offline behavioral measure used in Experiment 1 may have lacked sensitivity to detect fine-grained differences in sentence comprehension, where both language groups performed with high accuracy. To address these concerns, we conducted Experiment 2—an online version of the sentence-picture matching task—to examine potential differences in the cognitive load required during the comprehension of different types of adverbial sentences. Additionally, we measured bilinguals’ English proficiency at the time of testing to ensure a more accurate reflection of their current English proficiency.

Experiment 2

To measure cognitive load during the comprehension of adverbial sentences, we tracked participants’ pupil size changes. Pupil size measurement (pupillometry) has been shown to be a reliable psychophysiological indicator of processing effort during syntactic comprehension (e.g., Ayasse et al., Reference Ayasse, Hodson and Wingfield2021; Chapman & Hallowell, Reference Chapman and Hallowell2021; Fernandez et al., Reference Fernandez, Höhle, Brock and Nickels2018; Scherger et al., Reference Scherger, Urbanczik, Ludwigs and Kizilirmak2021). The more difficult a sentence is to comprehend, the greater the cognitive resources required, leading to larger increases in pupil size. Moreover, pupillometry has the advantage of allowing for more natural auditory stimulus presentation, resembling real-life listening scenarios, compared to methods like self-paced reading. It also provides a non-invasive and easy-to-set-up alternative to techniques such as EEG and fMRI.

Based on the results from Experiment 1, we pre-registered Experiment 2 in the Open Science Framework (OSF) repository, which specified research hypotheses, study design, sampling, and analysis plan (https://osf.io/m7u6v).