Context Facilitates the Decoding of Lexically Ambiguous Words for Adult Literacy Learners

Abstract

An estimated one-fifth of adults in the United States possess low literacy skills, which includes minimal proficiency in reading and difficulty processing contextual information. One way to study reading behavior of adults with low literacy is through eye movement studies; however, these investigations have been generally limited. Thus, the present study collected eye movement data (e.g., gaze duration, total time, regressions) from adult literacy learners while they read sentences to investigate online reading behavior. We manipulated the lexical ambiguity of the target words, context strength, and context location in the sentences. The role of vocabulary depth, which refers to the deeper understanding of a word in one’s vocabulary, was also examined. Results show that adult literacy learners spent more total time reading ambiguous words compared to control words and vocabulary depth was significantly correlated with processing of lexically ambiguous words. Participants with higher depth scores were more sensitive to the complexity of ambiguous words and more effective at utilizing context compared to those with lower depth scores, which is reflected by more total time reading ambiguous words when more informative context was available and more regressions made to the target word by participants with higher depth scores. Overall, there is evidence to demonstrate the benefits of context use in lexical processing, as well as adult learners’ sensitivity to changes in lexical ambiguity.

Introduction

According to the Organization for Economic Cooperation and Development (OECD), 19% of American adults possessed minimal to basic literacy skills in 2016, with an improvement of less than 2% as of 2019 (OECD, 2016, 2019). The OECD characterizes these individuals as being at or below Level 1 of literacy proficiency, meaning they may only be able to read short texts, recognize a certain number of words, and process contextual information that is easily comprehensible (Grotlüschen et al., 2016). Low levels of literacy can hinder employment opportunities, socioeconomic mobility, and successful communication in everyday life. In fact, 28% of these adults fall in the lowest income brackets for their respective countries (Grotlüschen et al., 2016). Additionally, more than half of adults at Level 1 or below (52.2%) were raised by parents with low literacy; hence, the cycle becomes recursive, stunting the academic achievement and socioeconomic mobility of children raised by these adults (Grotlüschen et al., 2016; Rindermann & Ceci, 2018). In contrast to the size of the issue at hand, there is limited research focusing on these individuals. For this reason, research that specifically targets adults with low literacy is necessary to create better programs with improved instructional methods that will help these individuals develop their vocabulary knowledge and literacy skills.

The existing literature makes it clear that the adult learner population varies extensively (Strucker & Davidson, 2003; Talwar et al., 2020; Tighe & Schatschneider, 2016). In fact, the current study included an ethnically diverse sample of adults recruited from local education sites, where age ranged from 16 to 63 and reading grade level varied from 1.3 to 12.8 according to the Woodcock-Johnson Picture Vocabulary Test (PVT) (WJ-III BR; Woodcock et al., 2001). Furthermore, Strucker and Davidson (2003) identified ten distinct skill clusters within a sample of 676 adult learners. These clusters included both low-level readers who exhibited deficiencies in phonemic awareness and basic word recognition skills, as well as intermediate readers with a better grasp on phonics and fluency, but who lacked the vocabulary and background knowledge to comprehend longer texts. Additionally, Talwar et al. (2020) identified four distinct classes within their sample of 542 adult learners. These consisted of adults who were weak in all measured competencies (e.g., metalinguistic awareness, word reading, comprehension, background knowledge, and inferencing), individuals who were weak in lower-level skills but relatively strong in higher-level skills and were thus labelled as weak decoders, those who exhibited the reverse pattern and were labelled as weak language comprehenders, and those who were relatively skilled in all areas. Finally, Barnes et al. (2017) examined eye movements while adult literacy learners read passages aloud and found that their participants made longer fixations and more regressions than skilled adults, suggesting a higher level of difficulty decoding and comprehending text. Findings such as these illuminate the need for more research focusing on this incredibly diverse population, as well as the use of methodologies that provide insight into what is happening while reading and comprehending are actively taking place so as to best support these readers’ successful decoding and comprehension.

Past research has shown a strong positive correlation between knowledge of vocabulary and reading comprehension; two necessary components of proficient literacy skills (Braze et al., 2007; Oakhill et al., 2019; Ouellette, 2006; Sabatini et al., 2010). Within vocabulary knowledge, vocabulary breadth and vocabulary depth each make important contributions to overall literacy (Binder et al., 2017; Proctor et al., 2011; Qian, 1999; Tran et al., 2020). Vocabulary breadth refers to the number of words a reader knows, while vocabulary depth relates to how much is understood about each word, including additional meanings and roles within sentence structures (Binder et al., 2017; Tran et al., 2020). In the past, research has focused primarily on vocabulary breadth; however, more recently, emphasis has been placed on studying vocabulary depth since findings suggest that depth acts as a stronger predictor of reading comprehension than vocabulary breadth (Binder et al., 2017; Binder et al., 2020). Many words possess multiple forms that may vary across contexts, tapping into readers’ vocabulary depth abilities and requiring them to identify which form is appropriate in a given situation (Rodd, 2020). A multi-faceted understanding of the words in one’s lexicon, and how their usage may change across contexts, is necessary to develop a robust vocabulary and comprehend more advanced materials.

One way of studying vocabulary depth involves lexically ambiguous words. Lexically ambiguous words are words with at least two different meanings that are either balanced or biased. Balanced ambiguous words have two equally frequent meanings, as well as other potential subordinate meanings (Kambe et al., 2001; Sereno et al., 2006). For example, the balanced ambiguous word charm has two meanings that are close in frequency: those associated with bracelet and wit. In contrast, biased ambiguous words have one dominant, most frequent meaning and one or more subordinate, less frequent meanings (Kambe et al., 2001; Sereno et al., 2006). According to Twilley et al.’s (1994) norms, an example of a biased ambiguous word would be ball, which has a dominant association with bat and a subordinate association with dance.

The literature offers several models to describe how lexical processing of ambiguous words occurs. One such model is the re-ordered access model (Duffy et al., 1988). According to this model, relative meaning frequency and biasing context influence how ambiguous words are processed (Duffy et al., 1988; Kambe et al., 2001; Wiley et al., 2018). Dominant meanings are recalled more quickly, as they are more likely to be correct given their higher frequency. Subordinate meanings, in contrast, are less likely to fit in a given context and are brought to the forefront of attention more slowly, thus preventing unnecessary distractions during comprehension (Rodd, 2020). Prior research has found that the surrounding context is one of the most important and useful keys for disambiguation (Rodd, 2020). When context reinforces word recognition, texts become more accessible to readers and provide the opportunity to decode unfamiliar words. Qian (2005) found that high depth readers made more use of contextual cues than low depth readers, while low depth readers mainly focused on the unknown target word. This suggests that higher depth individuals were better able to locate and recognize a contextual cue as being helpful in identifying the target word’s meaning, whereas lower depth individuals fixated longer on target words, had greater difficulty retrieving the meaning of target words, and were less successful at identifying informative context (Qian, 1999). Since higher vocabulary depth proficiency allows readers to take advantage of contextual cues within a sentence, an individual’s vocabulary depth may influence how a word and its context are located, accessed, and interpreted (Qian, 1999).

The location and strength of context can further influence how an ambiguous word is recognized and processed. In previous eye movement studies, when disambiguating context precedes an ambiguous word and biases the dominant meaning, readers show no significant differences between processing of a balanced ambiguous, biased ambiguous, or unambiguous word (Binder & Morris 1995; Kambe et al., 2001; Rayner et al., 2006). Research indicates this is due to the preceding disambiguating context facilitating the reader’s retrieval of the dominant meaning, making it unlikely that the reader will access the inappropriate and distracting subordinate meaning of the target word (Binder & Rayner, 1998). If the disambiguating context biases the subordinate meaning, however, readers spend longer processing a biased ambiguous word than a balanced or unambiguous word (Binder & Morris, 1995; Kambe et al., 2001). Research refers to this as the subordinate-bias effect (Kambe et al., 2001; Sereno et al., 2006). One possible explanation for this phenomenon is an increase in the speed at which a reader activates and retrieves the subordinate meaning of the word; thus, the inflated processing time reflects competition between the alternative meanings (Davidson & Weismer, 2017; Kambe et al., 2001; Sheridan et al., 2009). During lexical processing reinforced by context, the order of activation of dominant and subordinate interpretations becomes dependent upon the contextual support embedded within the sentence (Binder & Morris, 1995). Without the availability of contextual cues and disambiguating information, readers would be unable to retrieve the appropriate meaning of the word. With the support of context, readers can better understand the various meanings of a lexically ambiguous word and assimilate those individual meanings to expand their depth of vocabulary knowledge.

Context can also play a role in disambiguation when it follows the target word. When neutral context precedes an ambiguous word and disambiguating context follows, readers will spend the same initial processing time (e.g., first fixation duration or gaze duration) on a biased ambiguous word compared to a control word. Research indicates that the reader will have accessed the dominant meaning for the biased ambiguous word (Binder & Morris, 1995; Kambe et al., 2001). Eye movement measures of total reading time, however, show that the reader refixates on the biased word longer when the subsequent disambiguating context biases the subordinate meaning (Binder & Morris 1995; Kambe et al., 2001; Sereno et al., 2006). Kambe et al. (2001) suggest that this is because the reader initially accessed the dominant meaning of the biased ambiguous word prior to reading the disambiguating context. Once read, the reader most likely returns to the biased word to re-interpret and access the subordinate meaning, thus increasing the overall processing time of the biased word compared to the unambiguous control.

Reading behavior has been assessed in many ways. Some researchers rely on self-reports from participants (Cataldo & Oakhill, 2000), while others utilize technology to measure how participants process what they read (Ardoin et al., 2019; Vidal-Abarca et al., 2010). One such method of measuring reading behavior is through the use of eye-tracking technology. Eye-tracking measures eye movement behaviors, which allows for the monitoring of reading behavior in real time, without relying on potentially inaccurate or biased reporting from readers themselves (Hadwin et al., 2001; Kirk & Ashcraft, 2001; Rayner, 1998). In addition, researchers are able to extract several measures from the eye movement record, which allows for us to see how processing unfolds over time. We can examine gaze duration, which is the sum of all consecutive fixations made on a word before a reader moves past the word. Gaze duration is thought to reflect initial word recognition processes (Rayner, 1998). We can also examine measures that are thought to reflect more integrative processing such as total time on the word or region and number of times the reader regresses back to a word or region (Rayner, 1998). Total time is the sum of all fixations made on a word, and this would include any re-reading that happens. In the current study, we monitored participants’ eye movements as they read sentences that contained lexically ambiguous words and biasing context. This allowed us to examine how readers processed the target words initially, and if they reread the target word under certain context conditions, which allowed us to fill a critical gap in the literature examining adult literacy learners.

The Current Study

The purpose of this study was to understand how context influences adult learners’ processing of lexically ambiguous words and how this relates to their vocabulary breadth and depth. Participants read stimuli sentences on a computer screen while their eye movements were monitored. The sentences contained either a biased ambiguous word or an unambiguous control word supported by context, and we manipulated how informative contextual cues were, as well as varying the location of the context. Participants then completed a battery of tests to measure vocabulary breadth and depth. We hypothesized that (a) adult literacy learners would be sensitive to the strength and location of contextual cues; (b) more informative context that biased the subordinate meaning of the ambiguous target word would lead to shorter total reading time on the target when it preceded the target, but longer total reading times when it followed the target, due to the need to regress to the target and reinterpret the word. Additionally, we hypothesized there would be no difference in reading times (gaze duration, total time, or regressive eye movements) between the context conditions for the unambiguous control words; and (c) readers with higher vocabulary depth would make more use of supportive context and have shorter overall total reading times on the ambiguous target words.

Method

Participants

Participants included 53 adult literacy learners from educational sites in the northeastern region of the US. Participants ranged in age from 16 years to 63 years old (M = 29.21, SD = 12.18) and participants below the age of 18 years obtained parental consent before participating. While there is a wide age range, 85% of the participants were 40 years of age or younger. All participants were recruited from ABE or HiSET Prep classes at their educational sites. Since many ABE programs are structured around space and resource limitations, the number of classes provided by each site can vary. Our larger sites were able to split their student population into three levels, representing low, intermediate, and high skills, while our smaller sites were often split into two levels; those of basic education and HiSET Prep. Within the smaller sites, students who would be placed in the low and intermediate classes at our larger sites are often combined into a basic education class while students with higher skills participate in a HiSET Prep class to focus on preparing for the test. In our sample, 11% of participants were enrolled in a low-level class, 17% in an intermediate-level class, 19% in a higher-level class, 17% in a basic education class, 24% in a HiSET Prep class, and 11% did not know their class level. Additionally, according to the Woodcock-Johnson Picture Vocabulary Test (PVT) (WJ-III BR; Woodcock et al., 2001), the average reading grade level of our participants was 5.9, with scores ranging from 1.3 to 12.8. Finally, on average, participants read 98 words per minute, putting our readers at approximately a 2^nd grade reading level according to the DIBELS standardized testing benchmarks (DIBELS, 2020).

Fifty-one percent of participants racially identified as White, 9% as Black, African, or African American, 4% as Native Hawaiian or Pacific Islander, 2% as American Indian or Alaska Native, 2% as Asian, 4% as Multiracial, and the remaining participants identified as “Other,” which included responses such as “Hispanic,” “Latino,” “Puerto Rican,” or “Spanish.” Sixty percent of participants identified as female and 5 participants reported a first language of Spanish, but all participants were able to read in English.

Materials

Vocabulary Breadth and Depth Measures

One vocabulary breadth task and four vocabulary depth tasks were administered. The Picture Vocabulary Test (PVT), originally from the Woodcock-Johnson III Broad Reading test (WJ-III BR; Woodcock et al., 2001), was used to measure vocabulary breadth. Experimenters showed participants images that increased in difficulty and asked them to orally identify them. The task was terminated once the participant made 6 consecutive errors. Tran et al. (2020) report a reliability of .92 for this task for this population.

The four vocabulary depth tasks were the Suffix Choice, Word Families, Derivational Morphemes, and Target Word tasks. The Suffix Choice task (Mahony, 1994; Singson et al., 2000; Tyler & Nagy, 1989, 1990) presented participants with 14 incomplete sentences and 4 nonword answer options for each. The answer options were morphological derivations of the same nonword. For example, the sentence “The girl dances _____” was accompanied by the following choices: spridderish, spriddered, spridderly, and spridding, with spridderly as the correct answer. Participants followed along on a paper version of the test while the experimenter read the questions and answer options aloud. The participant then circled their answer choice on the paper. The task was terminated after 6 total errors. Tran et al. (2020) report a reliability of .89 for this task in this population.

During the Word Families task (Binder et al., 2017), the experimenter read a root word aloud and participants named as many derivatives of that root word as possible within 30 seconds. Participants orally provided derivatives for 10 root words. For example, for the root word act, possible derivatives could be actor, actress, acting, and acted. Tran et al. (2020) report a reliability .89 for this task in this population.

The Derivational Morpheme task (Carlisle, 2000) required participants to morphologically alter the structure of a target word to form a new word that best completed the sentence provided. For example, the experimenter read “warm. He chose the jacket for its _____,” aloud and the participant’s oral response in turn was expected to be warmth. The experimenter ended the task once the participant made 6 total errors. Tran et al. (2020) report a reliability of .97 for this task in this population.

Finally, the Target Word task (Richard, 2011) required participants to identify a single word that would complete each sentence in a cluster of 6 sentences. There were 30 items and participants had 15 minutes to complete the task. For example, participants would read the following item on paper and write their response on the blank line:

• Target Word: __________

• a. The car [ ]ed at the light. b. She [ ]ed and walked away. c. It is your [ ] to clean up after we eat. d. I [ ]ed it to face the window. e. My father [ ]ed the key and opened the door. f. There is a big [ ] in the road, so be careful.

The correct word for this item would be turn. Tran et al. (2020) report a reliability of .87 for this task for this population.

Stimuli

Target Words.

Forty-eight biased lexically ambiguous words were adapted from Kambe et al.’s (2001) study and Twilley et al.’s (1994) norming task to serve as targets. The 48 biased ambiguous words were carefully chosen based on their subordinate meaning and part of speech. In addition, these ambiguous words were matched to 48 unambiguous control words. The control words were matched based on word form frequency (Francis et al., 1982), length, and relevancy to the context of the ambiguous words. For the 48 ambiguous target words, the average of the bias for the subordinate meaning was 5.7% (range: 0.0%−18.0%) calculated from Twilley et al.’s (1994) norms of relative meaning frequency. The average word-form frequency was 52.8 per million for the biased ambiguous words and 48.6 per million for the control words (Francis et al., 1982). After conducting a paired samples t-test, there was no significant difference between the word-form frequency of the biased ambiguous words compared to the unambiguous control words, t(47) = 1.872, p = 0.07. See Table S1 for frequencies and bias norms for the target and control words.

Stimulus Sentences.

Participants read 48 experimental sentences. Sentences were adapted from Sereno et al.’s (2006) stimuli, as well as created by the researchers, and adhered to the 2 (biased ambiguous word vs. control word) × 2 (more informative context vs. less informative context) × 2 (context placed before target word vs. context placed after target word) design. For each item, the sentence contained either a biased ambiguous word or an unambiguous control word that served as the target word. Apart from changes to the target word, strength of context was manipulated to assess whether the target word was processed differently when supported by more or less informative context. More informative context included 3 to 4 contextual cues for the subordinate meaning of the target word. In the less informative context, the contextual cues were limited to 1 or 2 words. Furthermore, location of the context was manipulated, and the context either preceded or followed the target word, resulting in 8 conditions for each stimulus item. Participants viewed one version of each of the 48 passages in a counterbalanced manner, totaling 6 items for each of the 8 cells. See Table 1 for an example of a stimulus item and its conditions. During eye-tracking, participants were randomly presented with comprehension questions after some of the stimuli items to ensure focus throughout the task, but these questions did not require interpretation of the target word and were only intended to keep the participant engaged.

Table 1.

An Example of the 8 Conditions of a Stimulus

	Before Target Word (TW), Less Informative	After TW, Less Informative	Before TW, More Informative	After TW, More Informative
Biased Ambiguous Word	At the governor’s mansion during a lunch time meeting, Rob sat in front of the cabinet in the room.	Rob sat in front of the cabinet at the governor’s mansion during a lunch time meeting in the room.	For a debriefing and planning session at the governor’s mansion during a lunch time meeting, Rob sat in front of the cabinet in the room.	Rob sat in front of the cabinet for a debriefing and planning session at the governor’s mansion during a lunch time meeting in the room.
Matching Control Word	At the governor’s mansion during a lunch time meeting, Rob sat in front of the analyst in the room.	Rob sat in front of the analyst at the governor’s mansion during a lunch time meeting in the room.	For a debriefing and planning session at the governor’s mansion during a lunch time meeting, Rob sat in front of the analyst in the room.	Rob sat in front of the analyst for a debriefing and planning session at the governor’s lunch time meeting in the room.

Note. Target word is shown in bold.

Apparatus

Stimuli were presented on a 19-in. DELL 1907FP computer monitor while an SR Research EyeLink 1000 (SR Research Ltd.) eye-tracking camera recorded participants’ eye movements. The eye-tracker has a sampling rate of 1000 Hz, a resolution of 0.01 degrees of visual angle, and a range of 32 degrees horizontally and 25 degrees vertically. The camera recorded the movements of the right eye, but participants’ viewing was binocular.

Procedure

Data collection began with participants completing informed consent and demographic paperwork. Participants then sat in front of a computer monitor and placed their chin in a chin rest. Participants began the eye-tracking session by reading 2 practice sentences and answering comprehension questions about them. Participants then read 84 sentences silently to themselves. Of the items read, 48 were experimental stimuli and 36 were filler items that did not involve lexically ambiguous words. They answered comprehension questions as sentences were presented to ensure that they were maintaining focus. Participants indicated their answer choices and progressed to the next stimulus item by pressing the appropriate button on the connected controller. After the eye-tracking session was complete, the experimenter administered the battery of vocabulary tasks to the participant. Each participant received $10 for participating and all data collection was completed at the participants’ educational sites. The entire data collection session took approximately 1 hour, with the eye-tracking portion and the depth and breadth portion each taking about 30 minutes, depending upon the speed of the participant.

Results

Table S2 contains the descriptive statistics for all the measures. Several eye movement measures served as the dependent variables in the analyses. These measures were made on the target words (ambiguous targets and control targets) as well as time spent on the context regions. For the context regions, we identified the context regions, summed across them, and then divided by the number of characters to obtain average reading times on those regions. Linear mixed-effects (LME) regression models were run for each of the dependent measures. The appeal of using this type of regression model to analyze the sizable dataset at hand (i.e., 2,500+ data points) was, in part, due to the ability to incorporate by-subject and by-item influences as random effects into the same analysis (Baayen et al., 2008; Bates et al., 2018). The intercepts for subjects and items were included as random effects, while the fixed effects in all of the models included the main effects, 2-way interactions, 3-way interactions, and the 4-way interaction included in Table 2. Target Word, Contextual Strength, and Context Location were categorical variables, while Vocabulary Depth was a continuous variable in the original models. However, when Vocabulary Depth interacted with the other variables, we created three Vocabulary Depth categories (Low, Medium, and High) to establish the nature of how Vocabulary Depth interacted with the other variables. We should note that, as indicated in the Method section, several measures were used to assess Vocabulary Depth. Thus, we created a composite Depth score by running a factor analysis and using the regression method to create depth scores for each individual. As mentioned above, when the depth variable interacted with any of the other variables, we investigated the interactions by separating the participants into three depth groups, with approximately the same number of participants in each group. The range and means for each group was as follows: low, range −1.91 to −.52, M = −1.12; medium, range −.48 to .31, M = −.10; high, range .33 to 2.99, M = 1.12. We also note that the vocabulary breadth measure was not correlated with any of the dependent measures and was never a significant predictor in any of the models, so breadth was not included in the final analyses. However, for completeness, the means for the breadth measure for each of the depth categories was as follows: low, M = 32.12; medium, M = 31.68; high, M = 35.07.

Table 2.

Descriptive Statistics for Vocabulary, Acquisition, and Eye-tracking Measures

	Mean	SD
PVT Score	33.03	3.463
PVT Grade Level	5.93	3.120
Suffix Choice	8.62	3.895
Word Families	19.82	6.527
Derivational Morpheme	16.53	9.494
Target Word (VDT)	8.75	5.615
Vocabulary Acquisition	27.09	5.925
Target Gaze Duration	354.29	193.872
Target Total Time	562.82	378.547
Target Regression In Count	0.38	0.675
Target Regression In Count Sum	0.62	1.033
Target Regression Out Count	0.25	0.492
Target Regression Out Count Sum	0.70	0.897
Average Context Character Duration	85.22	62.820

Before analyses, outliers were dealt with using the outlier labeling rule, and this was specifically utilized for the target word data (Hoaglin & Iglewicz, 1987). This rule involved calculating a boundary value from the difference between the 75^th and 25^th percentile fixation values. This difference was then multiplied by g, a value of 2.2 (Hoaglin & Iglewicz, 1987). Then, the product was subtracted by the 25^th percentile value and added to the 75^th percentile value. By doing so, the boundaries were made, and any values that fell below the 25^th percentile or above the 75^th percentile values were considered outliers and replaced using the winsorizing method (Hoaglin & Iglewicz, 1987). This affected less than 2% of the data.

Target Word Analyses.

For the target word, we analyzed gaze duration, total time on the target word, regressions to the target word, and regressions from the target word to an earlier section of the sentence. See Table 3 for the fixed effects for all dependent measures. Gaze duration is the sum of all consecutive fixations on a word before leaving the word. This measure is thought to reflect initial stages of word processing (Rayner, 1998). Participants spent more time on the ambiguous word (M = 357 ms) compared to its control (M = 339 ms), as reflected in a significant effect of target word. There was also a significant effect of depth (β = −46.97, p = .001), with higher depth scores associated with shorter gaze durations, as well as a significant interaction between Target and Contextual Strength. When the context was less informative, there was no difference between the ambiguous target (M = 345 ms) and the control word (M = 346 ms; t = 0.40, p = .69). However, when the context was more informative, participants had longer gaze durations on the ambiguous word (M = 364 ms) compared to the control word (M = 331 ms; t = 3.33, p < .001). None of the other effects were significant.

Table 3.

Potential Fixed Effects in Models: Main Effects and Interactions Between Variables

Main Effects	2-way Interactions	3-way Interactions	4-way Interaction
Target word: Ambiguous or Control	Target*Contextual Strength	Target, Contextual Strength, and Vocabulary Depth	Target, Contextual Strength, Context Location, and Vocabulary Depth
Contextual Strength: Less or More Informative	Target*Context Location	Target, Context Location, and Vocabulary Depth
Context Location: Before or After Target	Contextual Strength*Context Location	Contextual Strength, Context Location, and Vocabulary Depth
Vocabulary Depth		Target, Contextual Strength, and Context Location

Total time is the sum of all fixations on a target word, which includes any rereading the participants did of the target word. Participants spent more total time on the target when it was an ambiguous word (M = 578 ms) compared to when it was a control word (M = 529 ms). This was supported by a significant effect of Target word. Participants spent more total time reading the target when it was found in a less informative context (M = 572 ms) compared to a more informative context (M = 536 ms), as indicated by a significant effect of Contextual Strength. Participants also spent more total time on the target when the context followed the target word (M = 575 ms) relative to when the context preceded the target (M = 533 ms). There was a significant 3-way interaction among Target word, Contextual Strength, and Context Location. Readers spent more time on the ambiguous word compared to the control word when the context preceded the target word, but was less informative (t = 16.67, p < .001), and when the context came after the target word and was more informative (t = 4.04, p < .001). There was no difference between the ambiguous word and control word when the context followed the target and was less informative (t = 0.25, p = .80), or came before the target and was more informative (t = 1.75, p = .08). See Figure 1.

Figure 1.

Total Time on Target Words in Relation to Contextual Strength and Location (Means and standard error bars)

Note: ***p < 0.001

Finally, the 4-way interaction among Target word, Contextual Strength, Context Location, and Vocabulary Depth was also significant. To investigate the nature of this interaction, we first created Vocabulary Depth groups (Low, Medium, and High) and then separated the data file by Contextual Strength and Context Location. We then examined differences between the ambiguous target word and the unambiguous control word. See Table 4 for interaction contrasts. Individuals who had higher vocabulary depth scores had longer reading times on the ambiguous targets relative to the unambiguous controls for 3 of the 4 contrasts. See Figure 2. The only difference that was not significant was when the context was less informative and came after the target words. Thus, it appears that high depth individuals appreciated the complexity of ambiguous words. There were no differences between ambiguous and unambiguous target words for the medium depth individuals. For the lower depth individuals, they spent more time on the ambiguous target word than the unambiguous control word when the context was more informative and came after the target words. No other differences were significant.

Table 4.

Statistics for Fixed Components of Target Word and Context Region Analyses

	Target Word								Context Region
	Gaze Duration		Total Time		Regressions In		Regressions Out		Total Time		Regressions In		Regressions Out
	F	p	F	p	F	p	F	p	F	p	F	p	F	p
Target	6.259	.012	13.202	.000	1.127	.288	.572	.450	.592	.442	.003	.953	.154	.695
Context	.430	.512	7.033	.008	1.446	.229	3.829	.050	26.313	.000	54.125	.000	105.921	.000
Context Location	.075	.784	9.557	.002	.083	.773	2.159	.142	9.228	.002	62.780	.000	27.633	.000
Depth Factor	12.443	.001	.274	.602	15.816	.000	1.522	.223	1.111	.294	.740	.393	6.540	.013
Target*Context	7.198	.007	2.042	.153	1.191	.275	.019	.889	.000	.994	5.036	.025	.440	.507
Target* Context Location	3.888	.049	.001	.976	.053	.817	3.469	.063	.224	.636	.439	.508	.064	.800
Context* Context Location	.464	.496	3.952	.047	6.204	.013	.874	.350	.086	.769	.654	.419	.003	.955
Target*Context* Depth Factor	.015	.904	2.917	.088	1.659	.198	1.268	.260	1.035	.309	.021	.885	5.236	.022
Target* Context Location* Depth Factor	.696	.404	.023	.879	4.032	.045	.166	.684	1.395	.238	.152	.696	.208	.649
Context* Context Location* Depth Factor	.072	.788	.002	.967	1.391	.238	.211	.646	.010	.919	1.135	.287	1.239	.266
Target*Context* Context Location	3.388	.066	4.931	.026	.864	.353	2.598	.107	.322	.570	.118	.731	1.425	.233
Target*Context* Context Location* Depth Factor	2.029	.154	8.204	.004	8.746	.003	1.367	.243	.446	.504	.283	.595	4.070	.044

Figure 2.

Total Time on Target Words in Relation to Contextual Strength and Location for Each Depth Group (Means and standard error bars)

Note: *p < 0.05, **p < 0.01, ***p < 0.001

We analyzed the number of regressions made to the target word. See Figure S1 for an example of regressive eye movement patterns. Vocabulary Depth was related to the number of regressions that participants made. The higher the depth score, the more regressions they made (β = .16, p = .002). There was also an interaction between Contextual Strength and Context Location. When the context came before the target word, participants regressed back to the target word more often when the context was less informative (M = .43) than more informative (M = .33; t = 2.70, p = .007). When the context came after the target word, there was no difference in the number of regressions between more (M = .39) and less informative contexts (M = .37; t = 0.55, p = .59). The 4-way interaction among Target, Contextual Strength, Context Location, and Vocabulary Depth was also significant. To investigate the nature of this interaction, we separated the data file by Vocabulary Depth group, Contextual Strength, and Context Location. We then examined differences between the ambiguous target word and the unambiguous control word. See Table 4 for interaction contrasts. Individuals who had higher vocabulary depth scores made more regressions back to the ambiguous targets relative to the unambiguous controls when the context was less informative and came after the target word. See Figure 3. There were no other mean differences across any of the other contrasts.

Figure 3.

Regressions to the Target Word in Relation to Contextual Strength and Location for Each Depth Group (Means and standard error bars)

Note: **p < 0.01

We also examined the number of regressions made from the target word to other regions of the sentence. See Figures S2 and S3 for examples of regressive eye movement patterns. There were no significant effects associated with this measure.

Context Words.

We identified all of the context regions in the passages, and for total time, we summed the total reading times associated with those words, and then we divided by the number of characters to equate the regions across items. Participants spent more time on context regions when they were less informative (M = 89.74 ms/char) compared to more informative (M = 79.37 ms/char). This was supported by a significant effect of Contextual Strength. In addition, participants spent more time on the context regions when they came before the target (M = 87.63 ms/char) compared to when they followed the target (M = 81.43 ms/char), as indicated by a significant effect of Context Location. No other effects were significant.

When we examined regressions to the context regions, we obtained effects of both Contextual Strength and Context Location. There were more regressions to the context regions when they were more informative (M = .75) compared to less informative (M = .47). In addition, there were more regressions to the context regions when they came before the targets (M = .76) compared to when the context followed the targets (M = .46). There was also an interaction between Target and Contextual Strength. Readers made more regressions to the context in the ambiguous word condition (M = .51) relative to the control word (M = .43; t = 1.90, p = .05) when the context was more informative. However, there was no difference between target word conditions for the less informative context condition (ambiguous, M = .71; control M = .79; t = 1.37, p = .17). No other effects were significant. Finally, we examined the number of regressions made out of the context to other regions of the sentence. Participants made more regressions out of the context regions when they were more informative (M = .88) relative to less informative (M = .54), as indicated by a significant effect of Contextual Strength. Participants made more regressions out of the context regions when they appeared after the target word (M = .80) compared to when they came before the targets (M = .62), as indicated by a significant effect of Context Location. There was also a significant effect of depth (β = 0.09, p = .01) in that individuals with higher depth scores made more regressions out of the context.

While there was a significant interaction among Target Word, Contextual Strength, and Vocabulary Depth, there was also a 4-way interaction between Target Word, Contextual Strength, Context Location, and Vocabulary Depth. To investigate this 4-way interaction, we separated the data by Target Word, Context Location, and Contextual Strength. We then conducted contrasts to see if there were differences in the number of regressions out of the context region across the three Vocabulary Depth categories. See Table 4 for interaction contrasts. When less informative context followed the ambiguous target words, individuals with higher depth scores made more regressions out of the context than lower depth individuals, but no other differences were significant. See Figure 4.

Figure 4.

Regressions Out of Context Regions in Relation to Contextual Strength, Context Location, and Target Type for Each Depth Group (Means and standard error bars)

Note: **p < 0.01

Discussion

The present study examined how adult literacy learners utilize context to process biased lexically ambiguous words. To explore adult learners’ reading behavior, we monitored participants’ eye movements as they read sentences containing lexically ambiguous or control words, and we manipulated the strength and location of the relevant context. Participants showed sensitivity to lexically ambiguous words compared to unambiguous control words, as well as to changes in contextual strength and location. Additionally, four tasks were used to measure vocabulary depth, and a composite Depth score for each participant was created from these four tasks, where each participant was then separated into one of three depth groups (i.e., low, medium, and high). Ultimately, we found vocabulary depth to be significantly related to adult learners’ processing of biased ambiguous words at both initial and later processing stages.

Lexically Ambiguous Words and Vocabulary Depth

Eye movement measures, such as gaze duration, total reading time, and regressions to target words, demonstrated readers’ sensitivity to lexical ambiguity and to changes in context. In accordance with previous findings, our results showed an increase in initial processing time spent reading biased ambiguous words compared to unambiguous control words (Binder & Morris, 1995; Kambe et al., 2001; Sereno et al., 2006). Without consideration for context strength or location, these results suggest an inflation in processing time as readers attempt to access and decide between the multiple meanings of the ambiguous words. Upon encountering a biased ambiguous word, readers likely access the dominant meaning of the word first (Kambe et al., 2001). Readers must then suppress the dominant meaning of the word to effectively retrieve the other possible meanings, thus increasing overall initial processing time on biased ambiguous compared to unambiguous control words. Similar patterns emerged when factoring in vocabulary depth skill. Adult learners with higher depth scores exhibited shorter initial processing times on ambiguous words compared to participants with lower depth scores. This suggests that in the early stages of processing, depth plays a supportive role in the recognition of ambiguous words.

Accounting for the role of context in early processing, readers’ gaze duration on ambiguous words was relatively similar to control words when the available context was less informative. There are two possible explanations for these results. One reason could be that readers likely accessed the dominant meaning of the ambiguous words initially, due to the lack of contradictory context, thus their reading time was not affected by a need to reinterpret the target (Kambe et al., 2001). The other reason could be that adult learners were simply unaware that they needed to distinguish between multiple meanings, thus competition between meanings that would lead to an inflation of reading time would not occur (Davidson & Weismer, 2017; Kambe et al., 2001; Sheridan et al., 2009). Either way, this would suggest that the context was not informative enough to cue readers to access the subordinate meaning, thus they continued reading with the dominant meaning in mind.

When the available context was more informative, however, adult learners’ initial processing reveals a different pattern. Participants initially spent more time reading ambiguous words when more informative context was present in the sentence. Previous research involving the re-ordered access model demonstrates how influential context is when it precedes ambiguous words and biases the subordinate meaning (Binder & Morris, 1995; Binder & Rayner, 1998; Kambe et al., 2001; Sereno et al., 2006). Upon encountering the biased ambiguous word following the disambiguating contextual region, readers will have accessed both the subordinate and dominant meanings at similar points in time (Kambe et al., 2001). The two competing meanings pose a challenge for readers, which results in a greater amount of time necessary to process ambiguous words, hence an overall slower reading rate. Considering past research related to lexical processing, we suggest sentential regions with more informative contextual cues allowed readers to activate the subordinate meaning, rather than simply assuming the dominant meaning, thus increasing overall processing time. When more informative context is provided, the subordinate meaning may be more accessible to readers compared to when less informative context is available.

Regarding later stages of processing, findings for total reading time depicted readers’ sensitivity to changes in both the strength and location of the context when processing lexically ambiguous words. When less informative context preceded target words, adult learners spent more total reading time on ambiguous compared to control words. The inflated time suggests readers were attempting to identify the appropriate interpretation but struggled due to a lack of informative context. When vocabulary depth is accounted for, learners with higher depth scores similarly exhibited longer total reading times when less informative context preceded ambiguous words. This pattern of results is not limited to when the contextual cues were less informative, however. Higher depth readers also had longer total reading times on ambiguous words compared to control words when more informative context either preceded or followed. Unlike readers with low or medium depth scores, readers with higher depth scores were sensitive to changes in contextual strength and location, in addition to showing an awareness for the complexity of lexically ambiguous words. When less informative context followed target words, learners could be incorporating the dominant meaning into their representation, resulting in no differences in processing time. However, when less informative context preceded or more informative context preceded or followed, higher depth learners successfully accessed the subordinate meaning, but it just took longer, as was reflected in the total reading times on the target words. With the availability of more informative context, higher depth readers’ inflated reading time also likely suggests that they are using surrounding contextual cues to process the ambiguous words (Qian, 1999). This reasoning is further supported by the pattern of regressions, where adult learners with higher depth scores made more regressions to the target word, regardless of whether the word was ambiguous or unambiguous. Readers could be regressing back to the target word to rectify their understanding after encountering the informative context, thus shifting their interpretation.

Moreover, when more informative context followed ambiguous words, readers with lower depth scores spent more total reading time on ambiguous words, while time spent on ambiguous words decreased for readers with higher depth scores. Previous research has investigated how readers are influenced by their level of vocabulary depth when attempting to understand new words (Qian, 1999). Qian (1999) found that readers with higher depth skills made greater use of contextual cues to access the meaning of an unknown word. Individuals with lower depth skills, in contrast, fixated on the target word instead of turning to the context for support (Qian, 1999). In accordance with the present study’s findings, this suggests that adult learners with higher depth scores likely recognize and use context more effectively than those with lower depth scores. Considering that readers with higher depth scores made more regressions out of contextual regions, this finding further supports that learners with greater vocabulary depth are better able to identify and return to useful context when determining the correct interpretation of a word (Qian, 1999). Learners with lower depth, compared to those with higher depth, do not seem to be as effective at utilizing context. Such reasoning is supported by lower depth readers’ inflated time spent on ambiguous words when more informative context followed, and a lack of regressions made out of contextual regions. With support from previous findings, it appears that lower depth readers are attempting to seek clues within the ambiguous word itself, rather than utilizing the surrounding sentential context to derive the appropriate interpretation (Qian, 1999).

Lexical Processing of Contextual Regions

Gathering a complete picture of adult learners’ reading behavior encompassed examining how they processed contextual regions embedded in the sentence. Focusing on the context’s location, findings demonstrated that readers spent more time and made more regressions to the contextual region when it preceded the target word, regardless of whether the target word was ambiguous or unambiguous. This suggests learners are likely re-reading sentences or regressing to relevant portions of context during processing. By re-reading or regressing to the context, adult learners are seeking and taking advantage of available cues as a means to gather support for determining the correct interpretation of the target word.

Lexical processing of contextual regions is not only influenced by changes in the context’s location, but also by changes in its strength. For example, readers made more regressions into more informative regions of context when an ambiguous word served as the target word. This reinforces that readers who are sensitive to lexical ambiguity are seeking relevant contextual support to access and retrieve the subordinate meaning. The patterns of regressions exhibited by our adult learners suggest an increased awareness of contextual cues as they made more regressions out of context regions when they were more informative. Even in cases where contextual cues were less informative, readers with higher depth scores, compared to those with lower depth scores, made more regressions out of the less informative context when it followed the ambiguous word. This illustrates that higher depth learners are likely more effective at seeking relevant contextual cues after encountering an ambiguous word, even in cases where the available context is not as helpful. Thus, this further supports higher depth readers’ sensitivity to lexical ambiguity and their ability to re-interpret the ambiguous word’s meaning after engaging with relevant context.

Limitations and Future Directions

One of the limitations associated with our study was that our depth factor was highly reliant on morphological awareness, even though we believe that construct, in accordance with the depth literature, is multifaceted (e.g., Binder et al., 2017; Ouellette, 2006; Proctor et al., 2011; Qian, 1999; Read, 1988; Stæhr, 2009; Tran et al., 2020; Wesche & Paribakht, 1996). However, we would like to note that one of our tasks, the Target Word Task (Richard, 2011) relies not only on morphological awareness, but some of the items also require that participants appreciate multiple meanings and multiple senses of the target words. In addition, some items require that the participants appreciate the different syntactic categories a target word can represent in a sentence. Thus, the addition of this task to our other measures that tapped morphological awareness was important to more fully measure vocabulary depth. However, in future studies, other measures that are thought to tap depth could also be used.

In addition, although we were able to report findings that support readers’ sensitivities to lexical ambiguity, changes in contextual strength and location, and their relationship to vocabulary depth, there was no assessment of vocabulary knowledge after the eye-tracking task. Future studies should consider incorporating a vocabulary acquisition task that would serve to assess whether readers acquired the correct meaning of the biased ambiguous words. Assessing readers’ knowledge of the target words after monitoring their reading behavior would illuminate whether readers can accurately infer meanings of lexically ambiguous words and how variations in context location and strength influence accuracy. Not only would this type of task reinforce the relationship between lexical ambiguity and vocabulary depth, but it could also emphasize the role context plays in accurately distinguishing multiple interpretations of biased ambiguous words. In addition, we were surprised that our vocabulary breadth measure did not correlate with any of our eye movement measures. We did only include one measure of breadth, so perhaps in future studies, breadth should also be measured using multiple assessments to get a more psychometrically stable variable.

Conclusion and General Implications

Overall, the present study’s findings suggest adult literacy learners are able to access and make use of supportive context to process biased lexically ambiguous words. There is also evidence to suggest that greater vocabulary depth is advantageous to word processing, especially in cases of lexical ambiguity where competition between meanings makes the use of context essential to determining the correct interpretation. Since research involving adult learners is still very limited, especially involving lexical ambiguity and context use, it was relevant to examine whether these readers showed similar processing trends consistent with previous research with other populations (e.g., Binder & Morris, 1995; Kambe et al., 2001; Sereno et al., 2006). Ultimately, a text becomes accessible to these adult learners when ambiguous words are embedded in informative context, and when they know how to successfully utilize contextual cues. Without context, readers would be unable to recognize or retrieve the proper meaning of an unfamiliar or lexically ambiguous word. For these adult learners, it may benefit them to learn how to better make use of context to recognize unfamiliar or ambiguous words in more difficult texts, thus increasing their overall understanding of the various materials they encounter in their everyday lives.

Table 5.

Contrasts for Interactions: Total Time on Targets Comparing Ambiguous to Control Words

		F	p
Higher Depth
More Informative
Before	6.44		0.01
After	6.16		0.01
Less Informative
Before	4.66		0.02
After	1.38		0.24
Medium Depth
More Informative
Before	1.24		0.27
After	1.82		0.18
Less Informative
Before	1.63		0.20
After	0.62		0.43
Lower Depth
More Informative
Before	0.14		0.71
After	10.61		0.001
Less Informative
Before	2.00		0.16
After	3.29		0.07

Table 6.

Contrasts for Interactions: Regressions to Target Comparing Ambiguous to Control Words

		F	p
Higher Depth
More Informative
Before	0.32		0.57
After	0.23		0.63
Less Informative
Before	0.05		0.82
After	6.60		0.01
Medium Depth
More Informative
Before	0.01		0.92
After	0.04		0.84
Less Informative
Before	0.11		0.75
After	0.66		0.42
Lower Depth
More Informative
Before	0.23		0.63
After	0.02		0.88
Less Informative
Before	3.28		0.07
After	1.05		0.31

Table 7.

Contrasts for Interactions: Regressions out of Context Comparing Across Vocabulary Depth Categories

		F	p
Ambiguous Words
More Informative
Before	1.17		0.32
After	0.06		0.95
Less Informative
Before	0.29		0.75
After	5.28		0.01
Control Words
More Informative
Before	0.01		0.99
After	0.87		0.43
Less Informative
Before	0.18		0.83
After	0.27		0.76