Content area
This paper introduces a new database of eye-tracking data on English derived words, DerLex. A total of 598 unique derived suffixed words were embedded in sentences and read by 357 participants representing both university convenience pools and community pools of non-college-bound adults. Besides the eye-movement record of reading derived suffixed words, the DerLex database provides the author recognition test (ART) scores for each participant, tapping into their reading proficiency, as well as multiple lexical variables reflecting distributional, orthographic, phonological, and semantic features of the words, their constituent morphemes, and morphological families. The paper additionally reports the main effects of select lexical variables and their interactions with the ART scores. It also produces estimates of statistical power and sample sizes required to reliably detect those lexical effects. While some effects are robust and can be readily detected even in a small-scale typical experiment, the over-powered DerLex database does not offer sufficient power to detect many other effects—including those of theoretical importance for existing accounts of morphological processing. We believe that both the availability of the new data resource and the limitations it provides for the planning and design of upcoming experiments are useful for future research on morphological complexity.
Introduction
Derivational morphology takes a central place in psycholinguistic research on complex words, with dozens of monographs, handbooks, and edited volumes (e.g., Booij, 2012; Halle & Marantz, 1993; Haspelmath & Sims, 2013; Kotowski & Plag, 2023; Lieber & Štekauer, 2014; Plag, 1999, 2018; Sproat, 1992) and thousands of journal papers dedicated to the subject (see reviews by Amenta & Crepaldi, 2012; Feldman & Milin, 2018). A Web of Science search across all databases using keywords “derivational morphology,” “derived words,” and “derivational suffixes” retrieved 2085 bibliographic items between the years 1976 and 2023. Among factors attracting researchers’ interest in studying derivation is the high productivity of this linguistic process, its ubiquity, and the diversity of its realizations across languages of the world (e.g., Lieber & Štekauer, 2014). As defined by Lieber (2017), “derivational morphology is concerned with forming new lexemes, that is, words that differ either in syntactic category (part of speech) or in meaning from their bases.” Often, this formation is realized through the addition of affixes to the root word. Indeed, so far, most experimental and computational work on how derived words are recognized has focused on affixation. Much of this work studies behavioral implications of (i) statistical properties of morphemes and their paradigms, (ii) affix ordering and other distributional properties of morphemes, (iii) semantics of the root morphemes and full derivations, (iv) orthographic and phonological cues towards the segmentation of derived words into morphological constituents, and (v) allomorphy and polysemy, among many others. The present paper contributes to this empirical base of research on visual recognition of derived words through the provision of the first behavioral eye-movement database of derived suffixed word reading in English.
There are very few eye-movement studies on English derived suffixed word reading (for a literature review on other languages, such as Dutch, see Kuperman et al., 2010). English studies that have been conducted focused on how eye movements are affected by the statistical and distributional properties of morphemes, i.e., point (i) above (e.g., Niswander et al., 2000; Ulicheva et al., 2020). For instance, Niswander et al., (2000; Experiment 1) manipulated root frequency (e.g., the frequency of elect in elective) and whole-word frequency (e.g., the frequency of elective) of derived words that were embedded into sentences in English. They found effects of both root frequency and whole-word frequency, with root frequency affecting earlier eye-movement measures (i.e., first fixation duration) than whole-word frequency (second fixation duration). More broadly, as with many experimental fields, the inquiry into visual recognition of derived words is grounded in relatively small-scale individual studies. Each study tends to use a specific experimental paradigm to record responses (e.g., lexical decision, word naming, eye-tracking, electroencephalography), and manipulates the orthographic or phonological form of those words, or the mode of their presentation to the reader (e.g., isolated word presentation vs. in-context word presentation, with or without priming; see Amenta & Crepaldi, 2012). Moreover, many studies, including the eye-movement studies of derived word reading cited above, test relatively small samples of participants compared with those in larger mega-studies (and typically recruit participants solely from university student populations). The resulting lack of statistical power of these smaller studies may render some of their findings spurious (e.g., Bakker et al., 2012; Brysbaert, 2019). Consequently, the outcomes described in the existing literature on the identification of derived words are at times difficult to compare and reconcile across methodologies, languages, and participant samples.
An effective (though partial) remedy to this issue is to create and use mega-studies, i.e., experimental studies in which behavioral data are collected for a large number of lexical stimuli and from a large number of participants, using the same methodology and apparatus. Typically, psycholinguistic mega-studies do not implement a specific manipulation but rather pursue a well-powered representation of behavioral responses to the broad range of linguistic stimuli, in a specific response paradigm. Prime examples of mega-studies that have thus far aided research on derived words are the family of Lexicon Projects. Initiated by Balota et al.’s (2007) English Lexicon Project, lexical decision studies have been conducted in a number of languages (see an overview in Ferrand et al., 2018), making accessible responses to tens of thousands of words from hundreds or thousands of participants. Given the highly productive nature of derivation as the cross-linguistic mechanism of word formation, mega-studies often include a substantial number of derivations as well (Mailhot et al., 2020; Sánchez-Gutiérrez et al., 2018).
Similarly, an important advancement of the large-scale research on derivations comes from the development of databases that collect or compile lexical characteristics for a large number of words. For instance, morphological databases of the MorphoLex and MorphoLex-F projects (Mailhot et al., 2020; Sánchez-Gutiérrez et al., 2018) match the English and the French Lexicon Projects (Balota et al., 2007; Ferrand et al., 2010), respectively, in that they offer a series of lexical variables for all words found in the respective lexical decision databases. The variables contain multiple estimations of frequency for the root word and derived word, affix productivity, family size, length, and several other formal and semantic properties. A similar morphological database exists for English compound words as well, LADEC (Gagne et al., 2019), which reports ratings and computational estimates of semantic similarity between the compound word and its constituents, along with multiple other distributional, formal, and semantic characteristics. Jointly, behavioral mega-studies and lexical databases enable researchers to conduct so-called virtual experiments without resorting to additional data collection on a variety of topics, including for instance topics (i)–(v) listed above for derived words. Specifically, researchers can draw samples of lexical items that vary across desired linguistic characteristics and analyze behavioral responses to these items as if they were obtained in an experiment that implemented this variability through a targeted manipulation (Keuleers et al., 2010; Kuperman, 2015).
The current paper adds to the collection of available mega-studies by presenting a database of eye-tracking data on English derived suffixed words which were embedded in sentences and read by a large number of participants. The eye-movement record while participants read texts substantially expands the empirical base of studies on complex word recognition (Bertram, 2011). Since sentences present a natural linguistic environment for derived words, this experimental paradigm has greater ecological validity than the single-word presentation techniques, including lexical decision. In fact, once the effects of benchmark predictors of word recognition—word length and frequency—are partialled out, lexical decision response times and fixation times to the same words show little to no correlation (Kuperman et al., 2013). For this reason, introducing an eye-tracking megastudy on visual recognition of English derived words in sentence context contributes to the desiderata of psycholinguistic research on derived words: it reports behavioral data on hundreds of derived suffixed words, collected from a relatively homogeneous and large sample of participants, using the same methodology and apparatus. This project largely replicates the structure and goals of CompLex, the eye-tracking megastudy of English compounds (Schmidtke et al., 2021).
The specific goals of this paper are to (a) make available the new megastudy and report its descriptive characteristics and reliability, (b) make use of the data to conduct power analyses and estimate the statistical power required to successfully examine specific lexical effects, and (c) recommend promising new venues for studying the effects of formal, semantic, and distributional properties on the visual recognition of derived suffixed words. Goals (b) and (c) are aided by the link between the present megastudy and the MorphoLex database (Sánchez-Gutiérrez et al., 2018).
Method
A subset of the stimuli lists used in the CompLex database (Schmidtke et al., 2021) was constructed such that the sentence frames contained derived suffixed words as well as compound words. The sentence frames were created with the specific aim of examining derived suffixed words and compound word processing. Therefore, all the derived word reading data in the present database were obtained in the same wave of data collection as CompLex.
Materials
A total of 605 unique derived words were manually generated and compiled for the study. A small number (seven) of these derived words also included a prefix, as identified using the morphological parse provided by the MorphoLex database (Sánchez-Gutiérrez et al., 2018). These words were configuration, enlargement, enforcement, enjoyable, unbearable, disrespectful, and awaken. Figure 1 displays the counts of the items broken down by suffix.
[See PDF for image]
Fig. 1
Barplot displaying the distribution of suffixes across the DerLex items broken down by part of speech
All 605 derived words were embedded in the same number of unique sentence frames. The derived word never occupied the first or the last position in the sentence frames; for example, the sentence for the derived word camper was The young camper sat on the doorstep and waited to leave. Other restrictions included that the context up to the derived word was kept neutral to minimize predictability effects, and that every sentence was restricted to a maximum of 90 characters so that it occupied no more than one line on the screen during the experiment. The stimuli were distributed across three lists, the names of which correspond to the list names in the CompLex database (List 2 = 211; List 4 = 198; List 5 = 196). Please note that there are slightly fewer items in the lists in DerLex than in CompLex because some sentences in these lists did not include derived words.
Participants
Data were collected from a total of 357 participants across six studies. Participants were recruited either from McMaster University (Hamilton, Ontario, Canada) or from non-university (community) participant pools (Hamilton, Ontario, Canada or New Haven, Connecticut, USA). All participants had English as their first language, did not have a diagnosed visual impairment or learning disability, and had normal or corrected-to-normal vision. All studies were approved by the relevant local ethics boards (McMaster Research Ethics Board, McMaster University, and the Yale University Institutional Review Board). We summarize the basic participant profiles of each study in Table 1. Please note that the study names match those reported in CompLex (Schmidtke et al., 2021). This is why trials from Study 4 are not present in DerLex: derived word stimuli were not in the stimuli lists shown to participants in Study 4. In Studies 1 and 2, stimuli lists were intermixed (see Table 1), and for these studies the sentences containing derived words served as fillers for trials with a different syntactic manipulation. The data considered here comprise sentences with only the derived word stimuli and no additional manipulation. Data from studies 3, 5, and 6 were reported in Schmidtke et al. (2017) on the time course of lexical processing.
Table 1. Individual study properties
Study | Sample | Population | List(s) | N | No. Female | No. Male | Age range (years) |
|---|---|---|---|---|---|---|---|
Study 1 | Community | New Haven | 2, 4, 5 | 77 | 54 | 23 | 16–24 |
Study 2 | Community | New Haven | 2, 4, 5 | 121 | 66 | 55 | 16–26 |
Study 3 | Community | Hamilton | 4 | 45 | 23 | 22 | 18–31 |
Study 5 | University | McMaster U | 2 | 35 | 26 | 9 | 18–37 |
Study 6 | University | McMaster U | 4 | 38 | 27 | 11 | 18–30 |
Study 7 | University | McMaster U | 5 | 41 | 34 | 6 | 18–26 |
The gender of one participant was declared undisclosed in Study 7
Apparatus
Eye movements in all studies were recorded with an EyeLink 1000 desk-mounted eye tracker (SR Research Ltd; Ontario). The eye-movement camera and the conjoined infrared illuminator were positioned on a desk beneath the stimulus display screen. The sampling rate of the pupil location and size was 1000 Hz. The EyeLink 1000 has an average gaze position error of < 0.05°, a resolution of 0.01° (root mean square error), and a micro-saccade resolution of 0.01°. Participants’ gross head movements were minimized using a chin support and a forehead rest, and all recordings were monocular (right eye).
Stimuli presentation
The stimuli were presented on a 10.75-inch x 13.25-inch screen with a refresh rate of 60 Hz. The sentences were displayed in Arial 14-point font. The font height was 0.5 cm and the distance from the screen to the participant’s eyes was 60 cm. Thus, one character space subtended approximately 0.48° of visual angle.
Procedure
Setup
In all studies, the eye tracker was first calibrated using a series of nine fixed targets distributed around the display, followed by a nine-point accuracy validation. Calibration was repeated after any breaks or whenever the experimenter judged necessary. Eye movements were recorded only when the calibration accuracy had an average error of < 1°. Participants were asked to read each sentence silently and were told that comprehension questions would follow some trials. Participants were told that they could take a break at any point during the study. The order of sentences was pseudo-randomized in each study. Each experimental session began with a series of 4–8 practice trials.
Experimental trials
Each trial began with a fixation point positioned in the middle left of the screen. The experimenter pressed a button which would display a sentence on the screen once the participant had fixated on this point. Participants were given 10 s to read the stimulus sentence. After participants finished reading a sentence, they were instructed to look at a rectangle at the bottom right corner of the screen. Once the participants’ gaze entered the area of the triangle, this triggered either the next trial or a comprehension question to appear. The next trial would automatically appear if the participant had not completed reading the sentence.
Comprehension questions
Comprehension of critical sentences in some studies was measured with yes/no comprehension questions that were asked for 25% to 30% of the trials. For example, the sentence, The fairytale ended with bountiful harvests for the whole kingdom was followed by the question, Did the kingdom have good harvests? Each comprehension question appeared in the center of the screen and the two possible answers were positioned three lines below. One answer was positioned to the left of the center of the screen and one answer was positioned to the right of the center of the screen. Participants responded by pressing a button on the left or right side of the keyboard. The side of the keyboard corresponded to the position of the answer. The position of the correct answer was counterbalanced throughout the experiment. The next trial would be presented automatically if the participant had not registered a response within the 10-s limit.
Databases
The DerLex database files (version 1) are formatted as.csv files and are hosted on the Open Science Framework at the following link: https://osf.io/fw9v2. The fully compiled DerLex database (
Derived word properties
The database containing the derived word characteristics includes the sentence frame ID (
The database also contains 20 numeric variables that quantify various properties of derived words at the orthographic, morphological, and semantic level. The motivation for focusing on these variables as predictors of morphological processing can be found in Schmidtke et al. (2017), and the definitions for these variables can be found in Table 2. The variables using UK frequency counts (derived word frequency–UK, stem frequency–UK, and lemma transition probability) were obtained from SUBTLEX-UK (Van Heuven et al., 2014) and those using US frequency counts (derived word frequency–US, stem frequency–US) were taken from SUBTLEX-US (Brysbaert et al., 2012). Morphological information from the MorphoLex database (Sánchez-Gutiérrez et al., 2018) was used to compute family size, family frequency, derivational family entropy, suffix productivity, and suffix frequency. The frequency counts for these measures were also taken from MorphoLex, which were originally obtained from the Hyperspace Analogue to Language corpus of English (HAL; Lund & Burgess, 1996). Levenshtein distance and Coltheart’s N were computed with the help of the
Table 2. Lexical properties of the derived words and their definitions
Variable | Label | Definition |
|---|---|---|
Derived word frequency–US | DerivedFreqUS | US-based subtitle frequency-of-occurrence of the derived word |
Stem frequency–US | StemFreqUS | US-based subtitle frequency-of-occurrence of the stem |
Derived word frequency–UK | DerivedFreqUK | UK-based subtitle frequency of occurrence of the derived word |
Stem frequency–UK | StemFreqUK | UK-based subtitle frequency of occurrence of the stem |
Lemma transition probability | LemmaTransProb | The ratio of the derived word frequency to its lemma frequency |
Family size | FamSize | The number of morphologically complex word types that share the same stem as the derived word (non-derivational forms and compound forms are excluded from this count) |
Family frequency | FamFreq | The summed frequency of morphologically complex word types that share the same stem as the derived word (non-derivational forms and compound forms are excluded from this count) |
Derivational family entropy | DerEntropy | Shannon entropy calculated over the word frequency-based probability distribution of all complex word types that share the same stem as the derived word (non-derivational forms and compound forms are excluded from this measure |
Suffix productivity | Prod | The number of morphologically complex word types that share a suffix with the derived word |
Suffix frequency | ProdToken | The summed frequency of all morphologically complex word types that share a suffix with the derived word |
Levenshtein distance | LevDist | The average Levenshtein distance of the derived word (OLD20; the mean orthographic distance from the 20 nearest orthographic neighbors) |
Coltheart's N | ColtheartN | The number of orthographic neighbors that can be generated by changing one letter of the derived word |
Length | Length | The word length in characters |
Form suffix productivity | OrthSuffixFreq | The type frequency of the suffix string appearing at the end of a word, irrespective of whether the word is morphologically complex |
Bigram transition probability | TransitBigram | The frequency of the first two letters of the suffix given that the preceding two letters (i.e., the last two letters of the stem) appear in its position relative to the end of the word. For suffixes two letters in length, we calculated the probability for just that letter |
Initial bigram frequency | InitBigram | The bigram frequency of the first two letters of the derived word |
Boundary bigram frequency | BoundaryBigram | The bigram frequency of the last letter of the stem and the first letter of the suffix |
Semantic distance | SemDist | The cosine similarity of the stem and derived word estimated using the SNAUT distributional semantics model. Cosine similarity was computed based on the UKWAC and SUBTLEX-UK database, a 300-dimensional semantic space, CBOW embeddings, and a six-word window for calculating co-occurrence statistics |
Derived word valence | DerivedValence | The average valence rating of the derived word target |
Stem valence | StemValence | The average valence rating of the stem |
Table 3. Summary of lexical properties of the derived words in the full database
Property | n | Mean | SD | Median | Min | Max |
|---|---|---|---|---|---|---|
Derived word frequency (US) | 571 | 290.05 | 881.31 | 73 | 1 | 14,266 |
Stem frequency (US) | 605 | 4386.16 | 14,915.9 | 722 | 1 | 203,947 |
Derived word frequency (UK) | 590 | 1740.42 | 7084.41 | 315.5 | 2 | 116,509 |
Stem frequency (UK) | 590 | 15,975.07 | 50,731.06 | 3664.5 | 2 | 681,544 |
Lemma transition probability | 590 | 0.85 | 0.22 | 1 | 0.05 | 1 |
Family size | 590 | 4.77 | 3.57 | 4 | 0 | 23 |
Family frequency | 590 | 12,377.06 | 30,215.08 | 3222 | 0 | 458,084 |
Derivational family entropy | 590 | 0.89 | 0.69 | 0.9 | 0 | 2.89 |
Suffix productivity | 590 | 1428.01 | 1223.4 | 632 | 4 | 3498 |
Suffix frequency | 590 | 1,903,565.98 | 1,714,369.37 | 988,835 | 40 | 4,499,116 |
Levenshtein distance | 590 | 2.47 | 0.73 | 2.55 | 1 | 4.9 |
Coltheart's N | 590 | 2.06 | 3.83 | 0.5 | 0 | 20 |
Length | 590 | 8.63 | 1.89 | 9 | 5 | 14 |
Form suffix productivity | 590 | 2704.06 | 3239.68 | 1026.5 | 88 | 13,048 |
Bigram transition probability | 590 | 0.03 | 0.06 | 0 | 0 | 0.22 |
Initial bigram frequency | 590 | 5411.62 | 4176.64 | 4241 | 242 | 23,014 |
Boundary bigram frequency | 590 | 4592.91 | 4146.9 | 3108 | 34 | 13,422 |
Semantic distance | 585 | 0.63 | 0.14 | 0.62 | 0 | 1.03 |
Derived word valence | 458 | 5.2 | 1.59 | 5.34 | 1.6 | 8.6 |
Stem valence | 573 | 5.38 | 1.52 | 5.6 | 1.53 | 8.78 |
Participant characteristics
The variables for the participant characteristics include subject ID (
The ART (Acheson et al., 2008) is a test that estimates an individual’s exposure to print. Participants are provided with an intermixed list of 130 names. Half of the names (65) are real authors, and the other half are distractors. Participants are asked to mark down the names they think are real authors. One point is given to every real author name that is marked down and one point is lost for every distractor name that is marked down. The possible range of scores is between − 65 and 65. A score of 0 indicates that the participant identified as many real authors as distractor names. ART scores are available for 349 of the 357 participants in this database.
The comprehension scores represent the accuracy of responses to the comprehension questions, expressed as a percentage. These scores are available for 257 participants. Table 4 reports descriptive statistics of the participant characteristics in the full database.
Table 4. Summary of reader characteristics in the full database
Characteristic | n | Mean | SD | Median | Min | Max |
|---|---|---|---|---|---|---|
Age (years) | 356 | 21.01 | 2.82 | 21 | 16 | 37 |
ART | 349 | 11.56 | 8.92 | 9 | − 1 | 44 |
Comprehension (%) | 275 | 52.97 | 45.15 | 86 | 0.16 | 100 |
Eye-movement variables
The eye-movement component of the database includes the participant ID (
Table 5. Eye-movement variables and their definitions
Variable | Label | Definition |
|---|---|---|
First fixation duration | FFD | The duration of the first fixation on the target word |
Second fixation duration | SECFD | The duration of the second fixation on the target word |
Gaze duration | GD | The summed duration of all fixations on the target word before the gaze left the target word for the first time |
Total fixation time | TFT | The summed duration of all fixations on the target word |
Regression path duration | RPD | The summed duration of all fixations on the target word, and to the left of the target word, before the gaze left the target word to the right for the first time |
Selective regression path duration | SRPD | The summed duration of all fixations on the target word before the gaze left the target word to the right for the first time |
Regression-in | REGIN | A binary measure of whether a fixation on the target word resulted from a regressive saccade |
Regression-out | REGOUT | A binary measure of whether a regressive saccade was executed after fixating on the target word |
Fixation count | FIXC | The count of all fixations on the target word |
Landing position | LP | The initial landing position of the gaze on the target word expressed as a proportion of the width of the word (0 corresponds to fixations in the space before the target word) |
Refixation | REFIX | A binary measure of whether the target word was fixated more than once before the gaze left the target word to the right for the first time |
Skipping | SKP | A binary measure of whether the target word was skipped |
Eye-movement data preparation
With all studies combined, the initial dataset consisted of 38,671 trials. This number includes trials for which the eye-tracking signal was not lost on a critical derived word in the sentence (the signal may have been lost in other parts of a critical word’s sentence). The initial count of trials also includes instances where the derived word was skipped in the first pass of reading. We excluded any trials for which more than seven fixations (either within-word refixations or regressions) were made on a critical derived word (removing the top 1% of fixation count distribution). Any first or second fixation durations shorter than 50 ms were excluded. We did not consider fixations where a target derived word was fixated on for the first time after the first pass of reading. Finally, we removed fixation durations that fell within the top 1% of the distribution for each of the following fixation time measures: first fixation duration (first fixation durations longer than 530 ms), gaze duration (gaze durations longer than 1071 ms), and total fixation time (total fixation times longer than 1528 ms). These operations led to a loss of 1814 data points (4.7% of the original dataset). The final eye-movement database contains 36,857 trials and includes observations when a derived word was skipped (1122 trials). Table 6 reports descriptive statistics of the eye-movement measures in the full database.
Table 6. Summary of eye-movement variables in the full database
Measure | n | Mean | SD | Median | Min | Max |
|---|---|---|---|---|---|---|
First fixation duration | 35,735 | 224.94 | 75.22 | 213 | 51 | 529 |
Second fixation duration | 21,498 | 203.37 | 87.9 | 187 | 51 | 1123 |
Gaze duration | 35,735 | 306.11 | 152.97 | 265 | 51 | 1069 |
Total fixation time | 35,735 | 428.39 | 250.31 | 365 | 51 | 1526 |
Regression path duration | 35,735 | 398.05 | 284.64 | 319 | 51 | 6177 |
Selective regression path duration | 35,735 | 340.74 | 177.86 | 295 | 51 | 1519 |
Regression-in | 35,735 | 0.26 | 0.44 | 0 | 0 | 1 |
Regression-out | 35,735 | 0.2 | 0.4 | 0 | 0 | 1 |
Fixation count | 35,735 | 2.01 | 1.12 | 2 | 1 | 7 |
Landing position | 35,735 | 0.36 | 0.23 | 0.33 | 0 | 1 |
Refixation | 35,735 | 0.54 | 0.5 | 1 | 0 | 1 |
Supplementary materials
Table S1 in the supplementary materials reports the number of observations for each derived word, both in the entire database and by study. Table S2 provides the distribution of sentence frames belonging to stimuli lists across studies. Summaries of item characteristics (Table S3), participant characteristics (Table S4), and eye-movement variables (Table S5) by individual study can also be found in the supplementary materials. All supplementary materials are hosted on the OSF repository.
Results and discussion
We conducted two analyses on the DerLex database. The aim of Analysis 1 is to provide an overview of the effects of a subset of lexical variables on a selection of durational eye-movement measures. In this analysis, our focus is on (i) estimating the main effects that a set of established characteristics of English derived suffixed words has on eye movements during reading, and (ii) determining whether such effects vary as a function of an individual’s exposure to print. The goal of Analysis 2 is to provide estimates of statistical power for a range of lexical effects on durational eye movements and sample sizes. The motivation for the power analysis is to provide researchers with an approximation of the number of trials that are required to detect a reliable effect on fixation durations.
Analysis 1
This analysis considered the following seven lexical variables: word length, derived word frequency–US, stem frequency–US, suffix productivity, lemma transition probability, derivational family entropy, and semantic distance. The seven variables were selected so that they covered knowledge of the orthographic (word length, surface frequency, stem frequency), morphological (lemma transition probability, derivational family entropy, suffix productivity), and semantic (semantic distance) properties of derived words.
Word length and (surface) word frequency have extremely robust effects on word recognition across languages and recognition tasks, including eye movements during reading (Kuperman et al., 2024; Rayner, 1998). Broadly construed, word length indicates the linguistic complexity of the word (the longer, the more complex) and word frequency approximates the degree of the reader’s experience with the word. Since complex words embed multiple morphemes, frequencies of those morphemes (e.g., the stem and the affixes of a derived word) demonstrably affect visual recognition of those complex words as well. Thus, we consider both stem frequency (e.g., frequency of the stem morpheme as a stand-alone word) and suffix productivity. While several quantifications have been proposed for the latter lexical property (e.g., Baayen & Renouf, 1996), we estimated suffix productivity as the number of word types that shared a suffix with the target word. Furthermore, earlier research proposed the use of lemma transition probability as a variable which reflects processing effects of morphological parsing (Solomyak & Marantz, 2010). It is defined as the ratio of each word’s surface frequency to its lemma frequency and serves as a measure of the conditional probability of encountering the whole word given the stem. The final variable considered here was derivational family entropy, e.g., Shannon’s entropy calculated over the probability distribution p of the word’s morphological family: the family was defined as words that shared the stem with the target. Derivational entropy is theoretically important for processing models advocating full decomposition of complex words and reflects the effect a word’s morphological family has on lexical access to the word’s decomposed stem (Fruchter & Marantz, 2015). Semantic distance is defined as the cosine similarity of the stem and derived word (see above) and quantifies the degree of semantic change to the stem word incurred by affixation.
The effects of the selected lexical variables were examined across four eye-movement fixation durations: first fixation duration, gaze duration, selective regression path duration, and total fixation time (defined in Table 5). These variables are argued to reflect the time course of lexical processing, whereby each variable indexes a stage at which a particular aspect of lexical processing is more likely to occur (Liversedge & Findlay, 2000; Rayner, 1998, 2009). Detailed reviews of how different eye movements reflect temporal stages of lexical processing can be found in Conklin et al. (2018) and Godfroid (2019). First fixation duration is often used as an index of lexical access, and as such is the earliest time at which lexical effects, such as word frequency and concreteness, are expected to emerge (e.g., Juhasz & Rayner, 2003; Magnabosco & Hauk, 2024). Gaze duration, the sum of all first fixations in the first pass of reading, arguably reflects additional cognitive processing, during which lexical access and semantic processing are also expected to occur (Inhoff, 1984; Inhoff & Radach, 1998). Selective regression path duration, which is the time spent reading the target word in the first pass of reading, including time spent fixating on the target after re-inspecting earlier parts of the sentence, is argued to reflect the syntactic and semantic integration of preceding contextual information with the target word (Radach & Kennedy, 2013; Radach et al., 2008). Finally, total fixation time represents the overall cumulative cognitive effort of word processing (see Table 5).
An additional independent variable in our analysis was the individual’s score in the author recognition test (Acheson et al., 2008), defined above. This measure of the individual exposure to print can be viewed as a rough measure of the reader’s experience with the words of the language and specifically less frequent words (including morphological derivations). For earlier demonstrations of interactions between the ART score and lexical properties of complex words see, e.g., Falkauskas and Kuperman (2015) and Schmidtke et al., (2018a, 2018b).
Statistical method and data
The dataset for this analysis included all eye movements for trials that were not skipped, for trials where there were complete observations for all seven lexical variables and ART scores. In addition, we only considered trials for derived suffixed words, i.e., words that did not include a prefix. The final dataset contained 32,496 eye-movement trials for 555 derived suffixed words from 349 participants.
To estimate the main effect of each lexical variable, we fitted a series of linear mixed-effect regression models to the full dataset (all studies combined) and each individual study. Each model was fitted to one of the four log-transformed dependent variables (first fixation duration, gaze duration, selective regression path duration, or total fixation time). Every model included the seven lexical variables and ART as main effects (all independent variables were scaled). With this generic fixed effect structure, separate models were run for each of the seven lexical variables, wherein the random-effects structure included by-participant random slopes for the target lexical variable under consideration. Thus, the fixed effect of a lexical variable was recorded only in the model when that lexical variable was included in the random-effects structure (see also Schmidtke et al., 2021). Other constant random effects included intercepts for item (i.e., derived suffixed words) nested under suffix (e.g., -ment). This accounts for the fact that the same suffixes are combined with different stems (e.g., government, department). In all, this modeling procedure resulted in 196 effects of interest (4 eye-movement measures × 7 lexical variables × 7 studies including the full dataset).
The same procedure was used for the modeling of the interaction of each lexical variable with ART (print experience). The model formula included the same random-effects structure, but this time the effect of interest was the interaction for each lexical variable and ART. Again, this run of models produced 196 effects of interest.
Model criticism involved refitting all models after excluding absolute standardized residuals exceeding 2.5 standard deviations from all datasets (Baayen & Milin, 2010). The condition number (κ) for the matrix of the predictor variables was 1.91, which indicates that collinearity between predictor variables was very low (Baayen, 2008; see Table 7 for correlations). The modeling procedure resulted in a family of four p-values for each lexical variable per dataset (i.e., one p-value for each of the four eye-movement measures per lexical variable per dataset). To control for inflated type I error, we applied the Bonferroni p-value correction to each family. Data analysis was performed using the
Table 7. Correlation table of lexical properties considered in the analyses (N = 555). Values above the diagonal are Pearson’s correlation coefficients, values below the diagonal are p-values
1 | 2 | 3 | 4 | 5 | 6 | 7 | |
|---|---|---|---|---|---|---|---|
1. Length | − 0.134 | − 0.339 | − 0.283 | 0.214 | 0.026 | 0.001 | |
2. Derived word frequency (US) | 0.002 | 0.275 | − 0.019 | − 0.068 | 0.014 | − 0.136 | |
3. Stem frequency (US) | < 0.001 | < 0.001 | 0.032 | − 0.014 | 0.311 | 0.065 | |
4. Suffix productivity | < 0.001 | 0.649 | 0.449 | − 0.408 | − 0.218 | − 0.057 | |
5. Lemma transition probability | < 0.001 | 0.108 | 0.743 | < 0.001 | 0.216 | 0.073 | |
6. Derivational family entropy | 0.548 | 0.735 | < 0.001 | < 0.001 | < 0.001 | 0.131 | |
7. Semantic distance | 0.986 | < 0.01 | 0.128 | 0.181 | 0.085 | < 0.01 |
Main effects of lexical variables
Figure 2 summarizes the effects of select lexical variables on eye movements in the full dataset and specific studies. The two benchmark predictors—length and frequency of the derived suffixed word—showed robust effects on reading times both in the individual studies and in the full dataset, and across the entire time course of word processing. As we will illustrate below, the effects were in the expected direction: longer and less frequent words elicit longer reading times than shorter and more frequent counterparts. The apparent exception, i.e., the null effect of word length on first fixation duration, is also well attested in the eye-tracking literature (Kuperman & Van Dyke, 2011).
[See PDF for image]
Fig. 2
Summary of the main effects of lexical predictors across eye movements and studies. A shaded box indicates that the lexical variable had a significant effect after correcting for multiple comparisons. FFD = first fixation duration; GD = gaze duration; SRPD = selective regression path duration; TFT = total fixation duration. Standardized regression coefficient estimates are provided in the boxes
Figure 2 also reveals strong effects of stem frequency on all eye movement measures in the full dataset: higher stem frequency came with shorter reading times. However, stem frequency effects in specific studies were either absent or non-systematic. They only reached significance in some measures of Studies 5, 6, and 7, and did not reveal an apparent bias towards early versus late stages of the time course of word processing. This patchwork of effects is likely due to the relatively small effect of stem frequency, which is more likely to emerge in more highly powered studies; see Analysis 2 below. Similar considerations apply to the effect of suffix productivity. It reached nominal statistical significance (after correction) in the cumulative eye-movement measure of total fixation time: words with more productive suffixes are read faster. Individual studies showed effects of total fixation time in Studies 5 and 7 and on other eye movements in Studies 5 and 6. We discuss below the interplay of statistical power with the emergence of significant effects of a relatively small size.
All other variables under consideration—lemma transitional probability, derivational family entropy, and semantic distance—did not demonstrate significant effects on eye movements in either individual studies or the full dataset. These null effects, obtained from the largest available collection of eye-tracking data on derived suffixed words, indicate the limitations of the empirical base for theories that ground their hypothesis testing in the effects of these lexical variables.
Interactions between lexical variables and print experience
The interactions between the ART score as the metric of exposure to print and select lexical variables are summarized in Fig. 3. Consistent interactions of ART by word length were found in selective regression path duration (SRPD) in the full dataset and Studies 1 and 2. More experienced readers both were overall faster and had weaker effects of word length. SRPD differs from total fixation time in that it includes durations of fixations made after regressions from the target word. This suggests that the interaction is driven by the number of regressions from the derived word and the duration of those fixations.
[See PDF for image]
Fig. 3
Summary of the interactions between lexical predictors and print experience (ART) across eye movements and studies. A shaded box indicates that there was a significant interaction after correcting for multiple comparisons. FFD = first fixation duration; GD = gaze duration; SRPD = selective regression path duration; TFT = total fixation duration
The most robustly observed interaction is between ART and derived word frequency (Fig. 4A). In gaze duration, total fixation time, and selective regression path duration, more experienced readers (with higher ART scores) both were overall faster and showed weaker effects of derived word frequency than their less experienced counterparts. These interactions emerged as significant (after correction) in the full dataset in Study 1 and (for total fixation time) in Study 6. The final noteworthy interaction was of ART and stem frequency. More experienced readers had shorter SRPD durations and had a weaker effect of stem frequency in the full dataset (see Fig. 4B).
[See PDF for image]
Fig. 4
Panel A shows the partial effects of log derived word frequency (scaled) modulated by ART (scaled), on gaze duration. Panel B shows the partial effects of log stem frequency (scaled) modulated by ART (scaled), on selective regression path duration. Slopes are provided for the 10th, 30th, 50th, 70th, and 90th percentiles of the scaled ART values (reported in the margin). Darker-colored slopes correspond to lower percentiles
All observed interactions were conceptually similar in that they revealed greater automaticity of word recognition in more experienced readers, who rely less on distributional or orthographic properties of complex words and their morphemes, compared to less experienced readers (e.g., Ashby et al., 2005). None of the remaining lexical properties under consideration (suffix productivity, lemma transitional probability, derivational family entropy, and semantic distance) showed a reliable interaction with ART scores.
Analysis 2
A major factor determining differences in the distributions of lexical effects between individual studies and the full dataset is the difference in the respective sample sizes and statistical power. Analysis 2 estimates power for the effects of lexical predictors, computed across a range of sample sizes. This analysis serves a dual purpose. First, it establishes how likely this large-scale data collection is to detect effects of varying magnitude, especially the subtle ones. Second, it provides a benchmark estimate of sample size for researchers interested in pursuing specific lexical effects via traditional smaller-scale experiments.
Power analysis method
A power simulation analysis was conducted to ascertain the level of statistical power for a range of effect sizes and numbers of items in the DerLex database. We first determined the effect sizes of interest. These effect sizes were defined as each of the observed fixed effect (beta) estimates for the seven lexical variables on log total fixation time in the full dataset in Analysis 1 (see Fig. 2). The models in Analysis 1 were all computed using scaled predictors, so all effect sizes are comparable and are expressed in units of standard deviation of the predictor. To recap, the effect sizes were as follows (from largest to smallest in absolute values): length = 0.082; derived word frequency = − 0.0679; suffix productivity = 0.0274; stem frequency = − 0.0231; semantic distance = 0.0101; lemma transition probability = 0.0057; derivational family entropy = − 0.0003. In absolute units, there is a relatively large gap between 0.0274 and 0.0679 in the range of observed effect sizes, so we also included the effect size of 0.04 to provide full coverage of values. Next, ten sample sizes were determined, beginning at 10% of the total number of observations of the full dataset used in Analysis 1 (3250 trials) and increased in increments of 10% to the full dataset size (32,496 trials). To put these samples sizes into perspective, the smallest sample size is equivalent to a fairly typical experiment in which 32 participants read 100 sentences each, and the full dataset is equivalent to 32 participants each reading 1000 sentences or 100 participants reading 320 sentences. With these eight effect sizes and ten sample sizes, we conducted a power simulation using the
[See PDF for image]
Fig. 5
Power estimates for eight effect sizes. Error bars correspond to the 95% confidence interval. Jitter added for readability. Statistical power expressed as a proportion
Power analysis results
Figure 5 reveals that the strongest effects (derived word length and frequency) can be observed with 80% power even in the experiment with the smallest sample size (10% of the total, or 3250 trials). A somewhat weaker effect of 0.04, which does not correspond to any lexical effect that we considered above, requires 6500 trials (e.g., 65 participants reading 100 sentences) to reach 80% power. The even weaker effect of suffix productivity requires 12,999 trials (129 participants reading 100 sentences) to reliably achieve the 80% power. A weaker yet effect of stem frequency numerically reaches the 80% power threshold at 19,498 trials and exceeds it numerically, but not with statistical significance, even in large samples. For the effects of all remaining lexical variables—lemma transitional probability, derivational family entropy, and semantic distance—even the full dataset of over 32,000 trials does not offer sufficient statistical power. These findings imply that earlier reports of such lexical effects (in the eye-tracking paradigm of sentence reading) likely stem from underpowered studies and are subject to well-described limitations and biases (e.g., Bakker et al., 2012; Brysbaert, 2019). In sum, the present estimates enable researchers to critically assess the feasibility of experimental consideration of some lexical variables proposed in the literature as theoretically important.
General discussion
The primary goal for this paper is to provide the field of morphological research with a large database of eye-tracking data on the processing of English derived suffixed words. The need for such a resource originates from the sustained and growing interest that researchers show in word formation processes, including derivation. It also stems from the diversity of methods, tasks, stimuli, and populations that limits comparability of existing experimental work. A final reason is that such a resource helps combat the underpowered studies that often give rise to spurious, non-reproducible effects, which then lead to theorizing with questionable empirical support (e.g., Bakker et al., 2012; Brysbaert, 2019). The DerLex database introduces eye-tracking data from 357 participants of varying reading proficiency who read 605 unique derived words (598 derived suffixed words) embedded in single sentences in English. Compiled over six studies, these data were obtained using the same apparatus, methodology, and well-defined variability in the reading proficiency of the participant sample. We supplement the behavioral data with estimates of readers’ exposure to print, as well as multiple lexical variables that characterize the distributional, orthographic, phonological, and semantic features of the derived words in the database.
This paper also makes use of the newly available data for two additional objectives. First, we estimate the effects of a small selection of lexical variables on durational eye movement measures. The selection of lexical variables represents a set of benchmark variables that are of specific importance for theories of morphological processing. This analysis yielded robust effects of benchmark variables—word length and frequency—and of stem frequency on eye movements. Other variables, including semantic similarity of the stem and derived suffixed words, lemma transitional probability, and derivational family entropy, failed to reach the nominal threshold of statistical significance. Similarly, interactions of lexical variables with a measure of the reader’s exposure to print (the author recognition test) were reliably observed only for derived word length and frequency and stem frequency. Given the large size of the DerLex database, these findings help us understand what effects are and are not readily reproducible in typical smaller-scale experiments.
The second objective formalizes the role of statistical power in estimating the effects of different magnitudes. With over 32,000 data points, the DerLex database offers a solid testbed for the design of experiments that aim to explore relatively subtle lexical effects using eye-tracking. Our calculations above provide estimates for the number of participants and trials required to reliably detect effects of different sizes. Perhaps the most important finding is that even the overpowered DerLex database (relative to typical experiments) does not offer sufficient statistical power to reliably detect effects for many of the lexical variables, including those of theoretical importance to influential accounts of morphological processing. This finding may inform a researcher’s decision on both how to operationalize their theoretical accounts (ideally, using variables that produce robust effects on human behavior) and how to design their experiments from the consideration of statistical power. We acknowledge that the present findings are specific to the sentence reading while eye-tracking paradigm, which is known to produce different results from the single-word presentation paradigms (e.g., lexical decision; see Kuperman et al., 2013). A comparison of effect sizes produced by the same variables while processing the same words in the eye-tracking record during reading versus single-word presentation tasks is an interesting methodological question to be addressed in the future.
Overall, given the high ecological validity and utility of eye-tracking data in morphological research (Bertram, 2011), we believe that this dataset will complement existing datasets on other types of morphological complexity (e.g., compounds, Schmidtke et al., 2021) and boost research on visual processing of morphological complexity.
Acknowledgements
We are thankful to Noor Al-Zanoon, Morgan Bontrager, Emma Bridgwater, Kaitlin Falkauskas, Irena Grusecki, Brooke Osborne, Sadaf Rahmanian, Katrina Reyes, Aaron So, Heidi Sarles-Whittlesey, and Chloe Sukkau for data collection.
Funding
Daniel Schmidtke’s contribution was partially completed during his PhD studies, which was supported by the Ontario Trillium Award and a Graduate fellowship awarded by the Lewis & Ruth Sherman Centre for Digital Scholarship (McMaster University). The remainder of Daniel Schmidtke’s contribution was supported by a Research Associate appointment at McMaster English Language Development Diploma (MELD) program, Faculty of Humanities, McMaster University, Canada. Victor Kuperman’s contribution was partially supported by the Canadian NSERC Discovery grant RGPIN/402395–2012 415 (Kuperman, PI), the Ontario Early Researcher award (Kuperman, PI), the Canada Research Chair (Tier 2; Kuperman, PI), the SSHRC Partnership Training Grant 895–2016-1008 (Libben, PI), and the CFI Leaders Opportunity Fund (Kuperman, PI). Julie A. Van Dyke’s contribution was supported by the following NIH grants to Haskins Laboratories: R01 HD-073288 (Julie A. Van Dyke, PI), and P01 HD-01994 (Jay G. Rueckl, PI).
Data availability
Research data are freely available at Open Science Framework (https://osf.io/fw9v2/). None of the reported studies were preregistered.
Code availability
The R analysis code is available upon request to the authors.
Declarations
Ethics approval
This study was approved by McMaster Research Ethics Board (McMaster University) and The Yale University Institutional Review Board.
Consent to participate
Informed consent was obtained from all individual participants included in the study.
Consent for publication
Informed consent was obtained from all individual participants included in the study.
Conflicts of interest/Competing interests
There are no conflicts of interest or competing interests to declare.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
Acheson, DJ; Wells, JB; MacDonald, MC. New and updated tests of print exposure and reading abilities in college students. Behavior Research Methods; 2008; 40,
Amenta, S. and Crepaldi, D. (2012). Morphological processing as we know it: an analytical review of morphological effects in visual word identification. Frontiers in Psychology, 3
Ashby, J; Rayner, K; Clifton, C, Jr. Eye movements of highly skilled and average readers: Differential effects of frequency and predictability. The Quarterly Journal of Experimental Psychology Section A; 2005; 58,
Baayen, RH. Analyzing linguistic data: A practical introduction to statistics using R; 2008; Cambridge University Press: [DOI: https://dx.doi.org/10.1017/CBO9780511801686]
Baayen, RH; Renouf, A. Chronicling the times: Productive lexical innovations in an English newspaper. Language; 1996; 72, pp. 69-96. [DOI: https://dx.doi.org/10.2307/416794]
Baayen, RH; Milin, P. Analyzing reaction times. International Journal of Psychological Research; 2010; 3,
Bates, D; Mächler, M; Bolker, B; Walker, S. Fitting linear mixed-effects models using lme4. Journal of Statistical Software; 2015; 67,
Bakker, M; van Dijk, A; Wicherts, JM. The Rules of the Game Called Psychological Science. Perspectives on Psychological Science; 2012; 7,
Balota, D. A., Yap, M. J., Hutchison, K. A., Cortese, M. J., Kessler, B., Loftis, B., .. & Treiman, R. (2007). The English lexicon project. Behavior research methods, 39, 445–459.
Bertram, R. Eye movements and morphological processing in reading. The Mental Lexicon; 2011; 6,
Booij, G. (2012). The grammar of words: An introduction to linguistic morphology. Oxford University Press.
Brysbaert, M; New, B; Keuleers, E. Adding part-of-speech information to the SUBTLEX-US word frequencies. Behavior Research Methods; 2012; 44, pp. 991-997. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/22396136][DOI: https://dx.doi.org/10.3758/s13428-012-0190-4]
Brysbaert, M. How many participants do we have to include in properly powered experiments? A tutorial of power analysis with reference tables. Journal of Cognition; 2019; 2,
Cohen, J. Statistical power analysis for the behavioral sciences; 2013; Routledge: [DOI: https://dx.doi.org/10.4324/9780203771587]
Conklin, K., Pellicer-Sánchez, A., & Carrol, G. (2018). Eye-tracking: A guide for applied linguistics research. Cambridge University Press.
Falkauskas, K; Kuperman, V. When experience meets language statistics: Individual variability in processing English compound words. Journal of Experimental Psychology: Learning, Memory, and Cognition; 2015; 41,
Feldman, LB; Milin, P. Berthiaume, R; Daigle, D; Desrochers, A. Psycholinguistic studies of word morphology and their implications for models of the mental lexicon and lexical processing. Morphological processing and literacy development: Current issues and research; 2018; Routledge: pp. 16-47. [DOI: https://dx.doi.org/10.4324/9781315229140-2]
Ferrand, L., New, B., Brysbaert, M., Keuleers, E., Bonin, P., Méot, A., … Pallier, C. (2010). The French Lexicon Project: Lexical decision data for 38,840 French words and 38,840 pseudowords. Behavior Research Methods, 42, 488–496.
Ferrand, L., Méot, A., Spinelli, E., New, B., Pallier, C., Bonin, P., .. & Grainger, J. (2018). MEGALEX: A megastudy of visual and auditory word recognition. Behavior Research Methods, 50, 1285–1307.
Fruchter, J; Marantz, A. Decomposition, lookup, and recombination: MEG evidence for the full decomposition model of complex visual word recognition. Brain and Language; 2015; 143, pp. 81-96. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/25797098][DOI: https://dx.doi.org/10.1016/j.bandl.2015.03.001]
Gagné, CL; Spalding, TL; Schmidtke, D. LADEC: The large database of English compounds. Behavior Research Methods; 2019; 51, pp. 2152-2179. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/31347038][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6797637][DOI: https://dx.doi.org/10.3758/s13428-019-01282-6]
Godfroid, A. Eye tracking in second language acquisition and bilingualism: A research synthesis and methodological guide; 2019; Routledge: [DOI: https://dx.doi.org/10.4324/9781315775616]
Green, P; MacLeod, CJ. SIMR: An R package for power analysis of generalized linear mixed models by simulation. Methods in Ecology and Evolution; 2016; 7,
Halle, M., and Marantz, A. (1993). Distributed morphology and the pieces of inflection. In The view from building 20 (pp. 111–176). The MIT Press.
Haspelmath, M; Sims, A. Understanding morphology; 2013; Routledge: [DOI: https://dx.doi.org/10.4324/9780203776506]
Inhoff, AW. Two stages of word processing during eye fixations in the reading of prose. Journal of Verbal Learning and Verbal Behavior; 1984; 23,
Inhoff, AW; Radach, R. Underwood, G. Definition and computation of oculomotor measures in the study of cognitive processes. Eye guidance in reading and scene perception; 1998; Elsevier: pp. 29-53. [DOI: https://dx.doi.org/10.1016/B978-008043361-5/50003-1]
Juhasz, BJ; Rayner, K. Investigating the Effects of a Set of Intercorrelated Variables on Eye Fixation Durations in Reading. Journal of Experimental Psychology: Learning, Memory, and Cognition; 2003; 29,
Keuleers, E., Diependaele, K., & Brysbaert, M. (2010). Practice effects in large-scale visual word recognition studies: A lexical decision study on 14,000 Dutch mono-and disyllabic words and nonwords. Frontiers in psychology, 1, 2082.
Kotowski, S; Plag, I. The semantics of derivational morphology: Theory, methods, evidence; 2023; De Gruyter: [DOI: https://dx.doi.org/10.1515/9783111074917]
Keuleers, E. (2013). vwr: Useful functions for visual word recognition research. R package version 0.3.0. Retrieved from http://CRAN.Rproject.org/package=vwr
Kuperman, V. Virtual experiments in megastudies: A case study of language and emotion. Quarterly Journal of Experimental Psychology; 2015; 68,
Kuperman, V; Bertram, R; Baayen, RH. Processing trade-offs in the reading of Dutch derived words. Journal of Memory and Language; 2010; 62,
Kuperman, V; Drieghe, D; Keuleers, E; Brysbaert, M. How strongly do word reading times and lexical decision times correlate? Combining data from eye movement corpora and megastudies. Quarterly Journal of Experimental Psychology; 2013; 66,
Kuperman, V; Schroeder, S; Gnetov, D. Word length and frequency effects on text reading are highly similar in 12 alphabetic languages. Journal of Memory and Language; 2024; 135, [DOI: https://dx.doi.org/10.1016/j.jml.2023.104497] 104497.
Kuperman, V; Van Dyke, JA. Effects of individual differences in verbal skills on eye-movement patterns during sentence reading. Journal of Memory and Language; 2011; 65,
Kuznetsova, A; Brockhoff, PB; Christensen, RHB. lmerTest package: Tests in linear mixed effects models. Journal of Statistical Software; 2017; 82, pp. 1-26. [DOI: https://dx.doi.org/10.18637/jss.v082.i13]
Lieber, R. Derivational morphology; 2017; Oxford University Press, USA:
Lieber, R; Štekauer, P. The Oxford handbook of derivational morphology; 2014; Oxford University Press: [DOI: https://dx.doi.org/10.1093/oxfordhb/9780199641642.001.0001]
Liversedge, SP; Findlay, JM. Saccadic eye movements and cognition. Trends in Cognitive Sciences; 2000; 4,
Lund, K; Burgess, C. Hyperspace analogue to language (HAL): A general model semantic representation. Brain and Cognition; 1996; 30, 5.
Magnabosco, F; Hauk, O. An eye on semantics: A study on the influence of concreteness and predictability on early fixation durations. Language, Cognition and Neuroscience; 2024; 39,
Mailhot, H; Wilson, MA; Macoir, J; Deacon, SH; Sánchez-Gutiérrez, C. MorphoLex-FR: A derivational morphological database for 38,840 French words. Behavior Research Methods; 2020; 52, pp. 1008-1025. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/31676967][DOI: https://dx.doi.org/10.3758/s13428-019-01297-z]
Mandera, P; Keuleers, E; Brysbaert, M. Explaining human performance in psycholinguistic tasks with models of semantic similarity based on prediction and counting: A review and empirical validation. Journal of Memory and Language; 2017; 92, pp. 57-78. [DOI: https://dx.doi.org/10.1016/j.jml.2016.04.001]
Niswander, E; Pollatsek, A; Rayner, K. The processing of derived and inflected suffixed words during reading. Language and Cognitive Processes; 2000; 15,
Plag, I. Morphological productivity: Structural constraints in English derivation; 1999; Mouton de Gruyter:
Plag, I. Word-formation in English; 2018; Cambridge University Press: [DOI: https://dx.doi.org/10.1017/9781316771402]
R Core Team (2024). R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. <https://www.R-project.org/>
Radach, R; Huestegge, L; Reilly, R. The role of global top-down factors in local eye-movement control in reading. Psychological Research Psychologische Forschung; 2008; 72,
Radach, R; Kennedy, A. Eye movements in reading: Some theoretical context. Quarterly Journal of Experimental Psychology; 2013; 66,
Rayner, K. Eye movements in reading and information processing: 20 years of research. Psychological Bulletin; 1998; 124,
Rayner, K. Eye movements and attention in reading, scene perception, and visual search. Quarterly Journal of Experimental Psychology; 2009; 62,
Sánchez-Gutiérrez, CH; Mailhot, H; Deacon, SH; Wilson, MA. MorphoLex: A derivational morphological database for 70,000 English words. Behavior Research Methods; 2018; 50, pp. 1568-1580. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/29124719][DOI: https://dx.doi.org/10.3758/s13428-017-0981-8]
Schmidtke, D; Matsuki, K; Kuperman, V. Surviving blind decomposition: A distributional analysis of the time-course of complex word recognition. Journal of Experimental Psychology: Learning, Memory, and Cognition; 2017; 43,
Schmidtke, D; Van Dyke, JA; Kuperman, V. Individual variability in the semantic processing of English compound words. Journal of Experimental Psychology: Learning, Memory, and Cognition; 2018; 44,
Schmidtke, D; Gagné, CL; Kuperman, V; Spalding, TL. Language experience shapes relational knowledge of compound words. Psychonomic Bulletin & Review; 2018; 25, pp. 1468-1487. [DOI: https://dx.doi.org/10.3758/s13423-018-1478-x]
Schmidtke, D; Van Dyke, JA; Kuperman, V. CompLex: An eye-movement database of compound word reading in English. Behavior Research Methods; 2021; 53, pp. 59-77. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/32572845][DOI: https://dx.doi.org/10.3758/s13428-020-01397-1]
Solomyak, O; Marantz, A. Evidence for early morphological decomposition in visual word recognition. Journal of Cognitive Neuroscience; 2010; 22,
Sproat, R. W. (1992). Morphology and computation. MIT press.
Ulicheva, A; Harvey, H; Aronoff, M; Rastle, K. Skilled readers’ sensitivity to meaningful regularities in English writing. Cognition; 2020; 195, [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/30509872][DOI: https://dx.doi.org/10.1016/j.cognition.2018.09.013] 103810.
Van Heuven, WJ; Mandera, P; Keuleers, E; Brysbaert, M. SUBTLEX-UK: A new and improved word frequency database for British English. Quarterly Journal of Experimental Psychology; 2014; 67,
Warriner, AB; Kuperman, V; Brysbaert, M. Norms of valence, arousal, and dominance for 13,915 English lemmas. Behavior Research Methods; 2013; 45, pp. 1191-1207. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/23404613][DOI: https://dx.doi.org/10.3758/s13428-012-0314-x]
© The Psychonomic Society, Inc. 2024.