Background
Sentence constraint in monolingual lexical processing
Research indicates that sentence constraint significantly influences visual word recognition High constraining sentences enhance lexical decision times for expected words, while unexpected words experience delays (Schwanenflugel & Shoben, 1985) According to the Interactive Activation Model (McClelland & Rumelhart, 1981), high constraint sentences activate more semantic nodes, facilitating compatible word activation and inhibiting incompatible ones Conversely, low constraint sentences activate fewer nodes, allowing for a broader range of lexical candidates and less inhibition of unrelated words Additionally, context can bias the interpretation of subordinate meanings (Kambe, Rayner, & Duffy, 2001) Importantly, eye movement studies reveal that reading is affected by sentence constraint, with target words in low constraining sentences receiving more fixation and longer viewing times, while words in high constraining sentences are often skipped (Ehrlich & Rayner, 1981; Rayner, 1998; Rayner & Well, 1996).
Selectivity in bilingual lexical processing
Research on selectiveness in lexical access has highlighted the processing of cognates, revealing that cognates are processed more efficiently than non-cognate control words Studies across various languages confirm that cognates are recognized faster in a second language (L2) during tasks such as lexical decision, priming, reading, and language production This cognate facilitation effect supports the idea of non-selective, parallel activation in bilingual lexical access, suggesting that lexical competition occurs between languages simultaneously The faster processing of cognates is attributed to shared mental representations across languages, facilitated by spreading activation among lexical items.
The cognate facilitation effect is influenced by various factors, including the language in which a task is performed and the proficiency of the individual While this effect has been observed in first language (L1) tasks, its strength and presence can vary depending on the language used This study specifically examines cognate processing in a second language (L2) Additionally, proficiency plays a significant role; research by Schwartz and Kroll (2006) indicates that more proficient readers experience a reduced cognate facilitation effect in a rapid serial visual presentation (RSVP) naming task.
Research indicates that word processing is enhanced when a target word appears in a high constraining context Additionally, cognates are processed more quickly than their matched controls Recent studies have explored the combined effects of sentence constraint and cognate status on word processing Notably, the cognate advantage remains evident even in low-constraint sentences, as demonstrated by various studies (Dijkstra et al., 2015; Libben & Titone, 2009; Schwartz & Kroll, 2006; van Assche et al., 2011; van Assche et al., 2009; van Hell & de Groot, 2008) For example, Duyck et al (2007) found that identical cognates, like "bar," are processed faster in low constraining contexts compared to non-identical cognates, such as "apple" – "appel," which showed no significant difference from control words.
Research on cognate effects in high constraining sentences has yielded inconsistent results, with some studies indicating either an inhibition or a complete elimination of the cognate effect (Schwartz & Kroll, 2006; van Hell & de Groot).
Schwartz and Kroll (2006) investigated L2 word recognition among Spanish-English bilinguals using an RSVP paradigm, revealing that cognates facilitated recognition in low-constraint sentences but not in high-constraint ones However, subsequent research indicated that the cognate advantage remained unaffected even in high-constraint contexts (Dijkstra et al., 2015; van Assche et al., 2011) Notably, Dijkstra et al (2015) found that Dutch-English bilinguals experienced cognate facilitation in a lexical decision task when the target was preceded by an L2 semantic context, regardless of sentence constraint.
Recent studies indicate that a high constraining sentence context may diminish the facilitatory effect of cognates, yet the findings are inconsistent, complicating conclusions about sentence constraint effects A quantitative meta-analysis by Lauro and Schwartz (2017) revealed that cognate facilitation was significant in both high and low constraint sentences, but the effect was notably smaller in high constraint contexts They identified two variables that could explain the varying results across the studies they reviewed.
Research indicates that task type significantly influences effect sizes, particularly for tasks necessitating overt responses, such as the lexical decision task (LDT) Studies in second language (L2) contexts demonstrate substantial effect sizes for cognate facilitation across both high and low constraint conditions, while first language (L1) studies reveal effects only in low constraint sentences The authors propose that selectivity in bilingualism is not a binary process; instead, it is a dynamic phenomenon that varies over time and across different processing stages, from word identification to the integration of word meanings.
This study investigates the hypothesis of gradient selection in bilingual lexical access by utilizing eye-tracking during reading to analyze the cognate facilitation effect in L2 This method allows for a detailed examination of the time course of word processing, enabling us to explore potential differences between the initial and later stages of lexical access Our findings indicate cognate facilitation in high constraints for one eye-tracking measure, reinforcing the concept of dynamic selectivity in bilingual lexical access.
In the next section, we present the experimental design of our experiment, along with the results
This study aims to explore the cognate effect and the influence of sentence constraint on lexical access in bilingual reading Utilizing a 2 × 2 design, Norwegian-English bilingual participants read cognate words and non-cognate controls within high- and low-constraint sentences The sentences were presented in English (L2), with cognates being orthographically identical to their Norwegian (L1) counterparts.
This analysis investigates the impact of word type, sentence constraint, and word frequency on early and late eye tracking measures Building on previous research regarding cognate facilitation, we anticipate shorter reading and processing times for cognate words compared to matched controls Additionally, we expect sentence constraint to influence the cognate facilitation effect, with three potential outcomes: a high-constraining sentence may (1) eliminate, (2) reduce, or (3) have no effect on cognate facilitation compared to a low-constraining sentence Furthermore, we predict differences between early and late measures, with cognate effects appearing in early measures but not in late measures To enhance comparability with prior studies, our experimental design and data analysis draw on the work of van Assche et al (2011).
This study enhances existing literature by validating previous findings on the cognate effect and sentence constraint's impact on lexical access in a different language Additionally, the use of eye-tracking technology provides valuable insights into the timing of language processing, making it ideal for examining lexical access and its subsequent processes Furthermore, since reading does not necessitate overt responses, this research may shed light on the dynamic nature of selectivity in bilingual lexical access.
2017), examining the question of if, when, and why non-selective access occurs.
Experiment
Method
30 Norwegian-English successive bilinguals (15 female, 15 male) participated in the experiment At the time of the experiment, the mean age of the participants was 27.5 years (SD
Participants in the study, all native Norwegian speakers, were recruited through advertisements in Trondheim, Norway, with the requirement of having grown up in Norway and using Norwegian as their first language A language history questionnaire confirmed that none were fluent in any language other than Norwegian or English All participants began learning English in primary school and lived in a monolingual Norwegian community, reporting a mean self-rated knowledge of L2 English at 5.78 (SD = 1) on a seven-point Likert scale They all utilized Bokmål as their written standard, which aligned with the written standard of the cognate words Each participant received a gift voucher worth 100 NOK (approximately £9) for their involvement in the study.
The age of English acquisition varied among participants, ranging from 22 to 37 years According to national curriculums from 1987 and 1997, those who began primary education in 1997 started learning English as a second language in 1st grade (ages 5-6), while earlier participants began in 4th grade (ages 10-11) Consequently, 16 participants commenced their English education approximately three years earlier than others This aspect will be revisited in the data analysis.
We created 264 sentences using 132 identical Norwegian-English cognate words and 132 control words, resulting in four groups of sentences with two constraint conditions (high vs low) for each word type The sentences were designed to be as similar as possible within each word type, differing by no more than two words, ensuring that high and low constraint sentences with the same target words were nearly identical Cognate and control sentences were matched in terms of word count, the word preceding the target word, and syntactic structure to facilitate data analysis and minimize variations in gaze patterns.
Overview of the language background of the participants
Self-rated written understanding (reading) b 6.13 (0.93)
Self-rated written production (writing) b 5.5 (1.04)
Self-rated oral understanding (listening) b 6.13 (0.82)
Self-rated oral production (speaking) b
LexTALE score c 84.07 (13.27) a Primary + higher education b Seven point Likert-scale (1 = low, 7 = high) c Score in % (min = 50%, max = 100%)
Examples of the sentences in each condition Target words in bold
High The airplane was flown by a pilot who was experienced
The doctor was accompanied by a nurse who worked there
Low The tour was guided by a pilot who was experienced
The seller was accompanied by a nurse who worked there
Cognates and control words were carefully matched based on word length, frequency, and word class, ensuring they were not phonologically similar Both cognate and control words had a mean length of 5.75, with standard deviations of 1.24 and 1.27, respectively The frequency difference between cognates and non-cognates was minimal, not exceeding 0.4 on the Zipf frequency scale, which ranges from 1 (low frequency) to 7 (high frequency) These frequencies were derived from the SUBTLEX-UK word frequency database.
In a study examining Zipf frequency, the mean English frequency for cognates was found to be 4.33 (SD = 0.38), closely matching the control words' mean frequency of 4.32 (SD = 0.44) In Norwegian, the mean frequency for cognates was 4.04 (SD = 0.46), revealing a difference of 0.25 in Zipf frequencies between the two languages for cognate words.
A study involving 30 participants from the same population conducted an online sentence completion test to assess Cloze probabilities, which measure the likelihood of specific words appearing in sentences The test comprised 264 sentences, split into two groups of 15 participants each to prevent repetition between high and low constraint pairs Each participant completed 132 sentences, providing the first word that came to mind Following the tests, 148 sentences were excluded: 43 high constraint sentences failed to exceed a 60% probability, and 9 low constraint sentences did not fall below a 40% probability Consequently, 118 sentences remained for the main experiment, consisting of 59 cognate and non-cognate pairs in both high and low constraining sentences (refer to Table 3 for mean Cloze probabilities).
Mean Cloze probabilities for the sentence constraints per condition in % (SDs in parentheses)
In order to obtain an objective measure of the participants’ English proficiency, they were asked to complete LexTALE, a ‘lexical test for advanced learners of English’(Lemhửfer
The average score on the English proficiency test was 84.07% (SD = 13.26) Participants self-rated their English qualifications on a Likert scale from 1 to 7, assessing their reading and speaking skills A linear regression analysis was conducted to determine how age, education, and L2 language use (daily, weekly, or monthly) predict L2 knowledge Significant results were found for the frequency of English use (p < 001), indicating that increased daily use of English correlates with higher LexTALE test scores Although not statistically significant, there was a numerical increase in LexTALE results with age (p = 566), suggesting that age of acquisition does not confound the study, as older participants with less formal English education were not less proficient in their L2 English.
We utilized an SR EyeLink 1000 eyetracker to capture the eye movements of participants throughout the experiment Binocular eye movements were recorded with gaze locations noted every millisecond The stimulus sentences, presented in black text (14 pt Times New Roman) on a white background, were limited to 80 characters and formatted in a single line Each target word was preceded by at least four words and followed by at least one word.
The study was conducted at the Language Acquisition and Language Processing Lab at the Norwegian University of Science and Technology in Trondheim, Norway, focusing on how Norwegian speakers read English text Participants received instructions in English and were asked to read sentences naturally, similar to reading a newspaper, pressing a button upon completion of each sentence The experiment continued until all sentences were read, including a calibration phase using a standard 9-point grid, and a language test, taking approximately 30 minutes to complete.
To prevent repetition priming, each participant was exposed to only one sentence from the word type condition, ensuring that no nearly identical sentences or target words were repeated Participants encountered one of four randomly assigned combinations: high constraining control with low constraining cognate, or high constraining cognate.
L control, or H cognate and H control, or L cognate and L control The sentences were presented in random order and each participant read a total of 59 sentences.
Data analysis
In our analysis, we incorporated both early and late eye-movement measures, as outlined by van Assche et al (2011) and Clifton et al (2007) The interest area for the target word was defined to include the space preceding it, excluding any area that follows For early eye tracking measures, we examined gaze duration, first fixation duration, and the number of skipped targets Gaze duration represents the total time spent fixating on the target word during the first encounter, while first fixation duration refers to the length of the initial fixation within the target area Additionally, we assessed the percentage of first pass regressions from the interest area, indicating how often readers returned to previous parts of the sentence For later eye tracking measures, we analyzed go-past time and total reading time, with go-past time being the cumulative duration of all fixations on the target word before moving on to the subsequent word.
According to Assche et al (2011), total time is calculated by summing the duration of all fixations on the target area (SR Research Ltd., 2002-2008) Late measures are believed to reflect semantic integration, while early measures indicate word identification processes and lexical access (van Assche et al., 2011) Additionally, processing times can provide valuable insights into these cognitive processes.
“spillover” to the upcoming word (Rayner & Duffy, 1986; Rayner, Sereno, Morris, Schmauder,
In our study, we examined the duration of the first fixation on the word following the target, referred to as the spillover region This region was defined as a fixed-width area of interest, encompassing two words if the subsequent word was short, while including only one word if it exceeded four characters.
In our data analysis, we employed linear mixed-effects models using the Lme4 package in R, coding skipped target words (15.7% of total) as missing values for various eye-tracking metrics We incorporated random intercepts for participants and items to account for random effects, and included a correlation between intercept and sentence constraint for each target word, except in cases where models failed to converge Although the interaction between sentence constraint and word type was not significant for eye-tracking measures, we retained it in the models due to its relevance in our experimental design Our models utilized sentence constraint, word type, and frequency as predictors To further explore the impact of sentence constraint, we conducted separate analyses on eye-tracking measures for high constraining sentences, with word type as a predictor Additionally, we applied a logistic model for the skipped words variable, given its binomial nature We present the results of first-pass measures followed by later stage eye movement measures, which are discussed in the General Discussion.
Results
First-pass eye-tracking measures
The study found that the percentage of skipped targets was higher in high constraining sentences compared to low constraining sentences Specifically, cognates were skipped more frequently than controls in low constraint sentences, while the skipping rates for both target word types were similar in high constraint sentences Statistical analysis indicated that readers were more likely to skip high constraint words than low constraint words (z = -2.567, p = 010) However, there was no significant effect of word type (z = -1.504, p = 133) or interaction between word type and sentence constraint (z = 1.584, p = 113) Additionally, the frequency of words did not significantly impact skipping behavior (z = 0.865, p = 387), and no effects of word type were observed in the high constraint data regarding target skipping.
Percentages of skipped target words
Sentence constraint Total Cognate Control
The study found that sentence constraint significantly affected gaze duration, with target words in low constraint sentences exhibiting longer gaze durations than those in high constraint sentences Additionally, cognates demonstrated shorter gaze durations compared to control words However, the interaction between word type and sentence constraint, as well as the effect of frequency, were not significant Further analysis of high constraint sentences revealed that cognates still had significantly shorter gaze durations than control words.
Fig 2 Mean first fixation durations (in milliseconds) by word type for high constraints (left) and low constraints (right) Error bars represent the standard error of the mean
Fig 1 Mean gaze durations (in milliseconds) by word type for high constraints (left) and low constraints (right) Error bars represent the standard error of the mean
First fixation duration Frequency had a significant effect on the first fixation durations
The analysis revealed that more frequent words are associated with shorter fixation times, indicated by a significant result of [ |t| = 2.207] However, the effects of word type [ |t| = 1.019] and sentence constraint [ |t| = 0.845] were not significant, nor was the interaction between these variables [ |t| = 0.065] As illustrated in Fig 2, cognates in high constraining sentences exhibited numerically shorter first fixation durations compared to controls, although this difference was not statistically significant Additionally, word type did not show significance for first fixation durations in the high constraint analysis, with a result of [ |t| = 1.778].
In our analysis of the spillover region, we investigated the duration of the first fixation following the target word The results revealed no significant differences based on word type (|t| = 0.524), sentence constraint (|t| = 0.464), their interaction (|t| = 0.508), or frequency (|t| = 0.192) Notably, control words exhibited numerically shorter first fixations in the spillover region compared to cognates, as illustrated in the accompanying figure.
3 The durations for spillover regions was not affected by word type in the high constraining sentences either [ |t| = 0.432]
Fig 3 Mean durations for the spillover region (in milliseconds) by word type for high constraints (left) and low constraints (right) Error bars represent the standard error of the mean
In the first-pass run, there were no significant differences among the four conditions regarding regressions from the target area to an earlier part of the sentence, as shown in Table 5, where the percentages across all conditions were comparable Consequently, we decided against conducting further analyses on this measure.
Percentages of regressions made out from the region of the target words
The analysis of go past time revealed no significant effects for word type (\$|t| = 0.551\$), sentence constraint (\$|t| = 0.931\$), the interaction between word type and sentence constraint (\$|t| = 0.580\$), or frequency (\$|t| = 0.319\$) This is further supported by the plots in Fig 4, which show that the means of go past times do not vary between word types within each constraint condition Additionally, in the high constraint analysis, the effect of word type remained insignificant (\$|t| = 0.098\$).
Total time For total time, the analysis revealed no significant results for word type [ |t|
The interaction between word type and sentence constraint was not significant, with values of |t| = 0.590 for sentence constraint and |t| = 0.285 for frequency Although the effect of sentence constraint was not statistically significant (|t| = 1.945), total time values were numerically larger for low constraining sentences and non-cognate target words This trend is illustrated in Fig 5, which shows that mean total times for targets in high constraining sentences exceed those in low constraint sentences Additionally, the analysis of high constraint sentences indicated that the effect of word type was also not significant (|t| = 0.566).
Fig 4 Mean go past times (in milliseconds) by word type for high constraints (left) and low constraints
(right) Error bars represent the standard error of the mean
Fig 5 Mean total reading time (in milliseconds) by word type for high constraints (left) and low constraints (right) Error bars represent the standard error of the mean
This study aimed to investigate the impact of sentence context on non-selective language activation by analyzing the cognate facilitation effect The findings reveal that word type significantly influences gaze durations, with cognates exhibiting shorter durations compared to control words, supporting previous research on cognate facilitation Although not statistically significant, the analysis indicated that first fixation durations, total time, and go past time were numerically shorter for cognates, suggesting that cognates may be processed more quickly than non-cognates.
The study found that sentence constraint significantly influenced gaze durations and skipping rates, with target words in low constraining contexts receiving longer gaze durations and being skipped less frequently than those in high constraining contexts This aligns with previous research indicating that high constraining sentence contexts facilitate target processing However, the lack of significant interaction between sentence constraint and word type for gaze durations and other eye tracking measures complicates conclusions about their combined effects on reading times and fixations.
The analysis of high constraining sentences indicates that word type significantly affects gaze durations, aligning with Libben and Titone (2009), who observed a similar effect in early reading measures Our findings suggest that high constraining sentences do not inhibit the cognate facilitation effect; in fact, gaze duration metrics show lower means for cognates compared to controls in these sentences, implying a potential boost in the cognate facilitation effect This contrasts with Lauro and Schwartz (2017), who reported a smaller effect size in high constraining contexts A vital next step is to replicate this study with English monolinguals.
Our study replicates findings from van Assche et al (2011), showing shorter gaze durations on cognates, reduced gaze durations in high sentence constraints, and increased target skipping in high constraining contexts However, unlike their study, we did not find significant effects for sentence constraint and word type in other first pass measures Additionally, while van Assche et al reported that cognates were skipped more frequently than controls, our late eye tracking measures revealed no significant effects for any predictors This discrepancy may stem from the different cognitive processes reflected in early versus late measures, with early measures indicating initial stages of sentence processing, such as lexical access Therefore, the absence of results in later stage measures does not contradict the hypothesis of non-selective activation in the bilingual lexicon during lexical access.
The type of word does not significantly influence the percentage of skipped targets, highlighting the impact of sentence constraint While sentence constraint plays a crucial role in determining whether a target word is skipped, it does not vary between different word types This indicates that the semantic context of a sentence can override the cognate facilitation effect regarding the likelihood of skipping targets Specifically, a highly constraining sentence context increases the probability of skipping a target, irrespective of its linguistic status This aligns with the findings of Rayner and Well (1996), who suggested that fixation times are indicative of word comprehension processes, while skipping rates may be influenced by lower-level visual features, such as the length of the subsequent word, which could explain the absence of a word type effect in this measure.
The significance of sentence constraint and word type was only observed in some early pass measures, likely due to the exclusion of phonological similarity in our stimulus selection and rating Previous research indicates that phonological overlap can inhibit processing for bilinguals, while semantic and orthographic overlap can facilitate it Our study focused on orthographically identical cognates, which should have resulted in facilitation, as supported by Dijkstra et al (1999) However, we only observed this effect in one eye tracking measure, suggesting a follow-up study is necessary to assess phonological similarity ratings of the stimulus words Additionally, the presence of participants from five major dialect groups in Norway may have introduced significant variation in phonological representations, potentially impacting our findings.
Our study revealed that word frequency significantly influenced first fixation durations, aligning with previous research indicating that more frequent words are associated with shorter fixation times (Inhoff & Rayner, 1986; Rayner & Duffy, 1986) However, it was unexpected that this effect did not extend to other early eye tracking measures, particularly gaze duration, which is typically more responsive to word frequency (Inhoff & Rayner, 1986) This lack of effect may be attributed to the relatively similar frequencies of our target words, which ranged from a minimum Zipf frequency of 3.4 to a maximum of 5.2, with a mean of 4.33 (SD = 0.33) Consequently, the insufficient variation in word frequencies may have hindered our ability to draw definitive conclusions regarding the modulation of the cognate effect or sentence constraint effect by frequency.
Our study indicates that bilingual lexical access operates in a non-selective manner, aligning with earlier research that shows sentence constraints can reduce selectivity We will examine our findings alongside previous studies within the framework of the BIA+ model proposed by Dijkstra and van Heuven (2002).
Fig 6 The BIA+ model (Dijkstra & van Heuven, 2002)