Hard Sentences to Read With No Space

Abstract

Prior research points to efficient identification of embedded words equally a key factor in facilitating the reading of text printed without spacing between words. Hither we further tested the principal part of bottom-upward word identification by altering this process with a letter of the alphabet transposition manipulation. In two experiments, nosotros examined silent reading and reading aloud of normal sentences and sentences containing words with letter transpositions, in both normally spaced and unspaced weather. We predicted that letter transpositions should be specially harmful for reading unspaced text. In line with our prediction, the majority of our measures of reading fluency showed that unspaced text with letter transpositions was unduly hard to read. These findings provide further support for the claim that reading text without between-word spacing relies principally on efficient lesser-up processing, enabling accurate word identification in the absence of visual cues to place give-and-take boundaries.

Introduction

A number of studies have investigated the ability of skilled readers to read text in which the extra interword spacing has been removed (e.g., Dreighe, Fitzsimmons & Liversedge, 2017; Epelboim, Berth, Ashkenazy, Taleghani & Steinman, 1997; Morris, Rayner & Pollatsek, 1990; Perea & Acha, 2009; Rayner, Fischer, & Pollatsek, 1998; Veldre, Drieghe & Andrews, 2017). The results of this research signal that reading unspaced text is slower by about xl–70% relative to reading normally spaced text (Rayner & Pollatsek, 1996; Rayner et al., 1998). Readers make shorter saccades accompanied past longer fixations and more regressions when reading unspaced text, and the effect of word frequency on fixation durations is greater with unspaced text (Rayner et al., 1998). Furthermore, given the overall shorter saccade lengths, initial landing positions are closer to the start of words in unspaced text (Paterson & Hashemite kingdom of jordan, 2010; Perea & Acha, 2009).^{Footnote 1}

The conclusion that has emerged from this research is that removing the spacing between words disrupts two singled-out processes: saccade programming and word identification (Perea & Acha, 2009; Rayner et al., 1998). Firstly, given the primal part for interword spaces in guiding eye movements during the reading of normally spaced text (e.one thousand., Inhoff, Eiter, Radach, & Juhasz, 2003), removing interword spaces will logically affect saccade programming. The results of prior research propose that readers adopt a more cautious oculomotor strategy when reading unspaced text, leading to a greater number of saccades per judgement (both forward and regressive) that are shorter in length. Secondly, the longer time spent inspecting each give-and-take when reading unspaced text (as reflected by longer fixation durations) is well-nigh probable due to the absence of visual cues for word beginnings and endings, and also perchance due to crowding effects occurring not simply for the give-and-take'due south inner-positioned letters, equally is the case in normal (spaced) reading (due east.thou., Tydgat & Grainger, 2009), but also for the word's outer-positioned letters.

In some other study on reading unspaced text (Mirault, Snell, & Grainger, 2018) nosotros investigated the role of sentence-level structures. In that study we compared reading of grammatically right sentences and shuffled versions of the same words presented both with normal spacing and without spaces. In line with prior research, nosotros establish that reading was hampered by removing sentence structure (Schad, Nuthmann, & Engbert, 2010). Furthermore, there was merely express evidence that sentence structure facilitated the reading of unspaced text more than then than reading spaced text. This pattern of results suggests that our power to read grammatically correct unspaced text is not principally due to a greater involvement of top-downward feedback from judgement-level structures.

On the other paw, our prior research did betoken to a key part for word identification processes in reading unspaced text, not only for linguistic processing, but also for guiding middle movements. Nosotros constitute that the length of the currently fixated discussion determined the amplitude of frontwards saccades leaving that word during the reading of unspaced text. This result suggests that readers of unspaced text use length information about the currently fixated word in society to program a saccade across that discussion's rightward boundary. In the absenteeism of visual cues, such length information can only be obtained by word identification providing admission to data almost give-and-take length. We therefore concluded that the relative ease with which skilled readers can read unspaced text is mainly due to efficient bottom-upwardly discussion identification processes standing to operate, and that support from judgement-level structures can facilitate these processes in certain weather condition. Further support for this conclusion was found in the significantly greater touch on of word frequency in the unspaced condition compared with normal spacing (see besides, Veldre et al., 2017).

The present study was designed to farther investigate the hypothesized importance of bottom-upwardly word identification processes when reading unspaced text. Why might discussion identification exist more important for reading unspaced text? Offset of all, nosotros have shown that word identification guides eye movements when reading unspaced text, whereas the visual cues provided by interword spacing are the principal guiding factor when reading normally spaced text. Secondly, when reading unspaced text, discussion identification provides discussion club information that is necessary for the structure of a sentence-level representation. That is, the lodge in which words are identified is the main source of word order information, whereas with normally spaced text, the construction of a sentence-level representation benefits from the presence of interword spaces that facilitate the consignment of society information to discussion identities (Grainger, 2018; Snell & Grainger, 2017; Snell, Meeter, & Grainger, 2017). In line with this reasoning is the bear witness obtained from readers of Thai, a linguistic communication with an alphabetic script that does not utilise between-discussion spacing. Information technology has been shown that Thai readers benefit from the artificial insertion of interword spaces, and the eye-movements of these readers advise that this facilitation arises mainly at the level of give-and-take identification and sentence-level comprehension (Winskel, Perea, & Ratitamkul, 2012; Winskel, Radach, & Luksaneeyanawin, 2009).

In the present study, we tested the hypothesized greater role for word identification in reading unspaced text by selectively perturbing this process. We did so by introducing letter transpositions in certain words in each sentence. In a seminal written report, Rayner, White, Johnson, and Liversedge (2006) recorded eye movements while participants read sentences that could either be formed of commonly written words or contained a number of words with letter transpositions (e.g., The boy cuold non slove the probelm so he aksed for help). Rayner et al. reported that although reading text containing letter transpositions was relatively fluent, in line with prior findings from the single word recognition literature (e.g., Perea & Lupker, 2004; see Grainger, 2008, for a review), there was nonetheless a cost. That is, reading text containing letter transpositions induced longer fixation durations and more refixations and regressions compared with ordinarily written text (come across besides Blythe, Johnson, Liversedge, & Rayner, 2014; White, Johnson, Liversedge, & Rayner, 2008).

The specific aim of the present written report was to exam the prediction that alphabetic character transpositions should have a significantly greater impact on reading unspaced text compared with normally spaced text. Two prior studies have conjointly manipulated betwixt-give-and-take spacing and letter transpositions and have produced contradictory findings. Winskel et al. (2012) investigated the effects of alphabetic character transpositions and interword spacing in Thai. These authors reported an interfering outcome of alphabetic character transpositions that did not interact with the spacing manipulation. However, the lack of an interaction in this study is probable due to the fact that Thai readers have developed efficient mechanisms for word segmentation in the absence of interword spacing, plus the fact that the presence of interword spaces is not natural for Thai readers. More direct related to the present written report is the work of Johnson and Eisler (2012), who investigated effects of letter transpositions and interword spacing in English. The central results are those obtained in their Experiment iii, where rather than replacing interword spaces with filler stimuli, the typical greater spacing between words compared with inter-letter spacing was cancelled past increasing inter-alphabetic character spacing. The specific aim of that study, however, was to investigate effects of the position of letter transpositions, and the critical interaction between transposition effects (measured relative to a no-transposition condition) and spacing was not tested. Nevertheless, the condition ways revealed much greater transposition effects in the absenteeism of extra between-word spacing in all reading time measures except for get-go fixation durations. All the same, the selection to increase betwixt-letter spacing rather than reducing between-word spacing might accept impacted on their results. Therefore, in the nowadays report we provide a further test of the predicted interaction betwixt transposition effects and interword spacing in two experiments where normal inter-alphabetic character spacing was retained and the additional space betwixt words was removed. In Experiment 1 we recorded eye movements while participants silently read sentences, and in Experiment 2 we nerveless audio recordings while participants read aloud the same set of sentences.

Experiment 1: Silent reading

Method

Participants

Thirty-two participants (24 female person)^{Footnote ii} from Aix-Marseille University, Marseille, France, received either €x per 60 minutes or class credit for their participation. The participants were all native French speakers and gave written consent prior to the experiment. They reported having normal or corrected-to-normal vision, ranged in historic period from 18 to 28 years (One thousand = 22.07, SD = 2.46), and were naïve with regard to the purpose of the experiment. French language skills were assessed using a Spelling Dictation test (Beyersmann, Casalis, Ziegler, & Grainger, 2015) and the LexTale vocabulary test (Brysbaert, 2013). Participants' average scores were 78.59% (SD = thirteen.75) on the dictation test, and 88.37% (SD = 4.86) on the vocabulary test.

Design and stimuli

Nosotros constructed 104 sentences in French, each containing seven words. The sentences ranged in length from 37 to 57 characters including spaces (Thou = 47.94, SD = iii.77), and the average word frequency was ane,825 parts per one thousand thousand (ppm) (based on the Lexique2 film frequency counts: New, Pallier, Brysbaert, & Ferrand, 2004), which is equivalent to 6.26 Zipf (van Heuven, Mandera, Keuleers, & Brysbaert, 2014). Following a ii × 2 factorial design, we manipulated between-word spacing (Spacing: spaced vs. unspaced) and word letter club (Transposition: ordinarily written words vs. words containing transposed letters). The introduction of letter of the alphabet transpositions was constrained by five criteria: (i) the messages were next consonants^{Footnote iii}, (ii) the first 2 and the concluding ii letters of words were never transposed, (iii) the messages did not form a complex grapheme, (iv) the discussion containing the transposed letters was at least 5 letters long, and (v) words containing the transposed messages were always at the second position (verb), the quaternary position (noun), and the 5th position (describing word) in sentences (i.e., 3 disquisitional words per sentence independent letter transpositions in the transposed-letter condition). These disquisitional words had an average frequency of four.55 Zipf and an average length of eight letters. Words containing letter transpositions were never repeated across the different sentences seen by a given participant. Sentences were presented in lower case, except for the initial uppercase letter, and but independent letters without accents (see Appendix for a complete list the sentences and their transposed-letter versions). A Latin-square design was used with iv groups of participants to ensure that all sentences were tested in all four weather condition, but were seen just one time per participant. Therefore a given gear up of iii critical words were seen ordinarily written and written with letter of the alphabet transpositions in both the spaced and unspaced atmospheric condition simply by different participants.

Apparatus

Stimuli were displayed using OpenSesame (Mathôt, Schreij & Theeuwes, 2012), with each judgement occupying a single line. Eye movements were recorded with an EyeLink one thousand arrangement (SR Research, Mississauga, ON, Canada) with loftier spatial resolution (0.01°) and a sampling rate of one,000 Hz. Viewing was binocular, but only the right eye was monitored. The sentences were displayed on a gamma-calibrated 20-in. ViewSonic CRT monitor with a refresh rate of 150 Hz and a screen resolution of 1,024 x 768 pixels (thirty 10 40 cm). Stimuli were presented in black (0.15 Cd/m²) on a gray background (21.70 Cd/kⁱⁱ). Participants were seated 86 cm from the monitor, such that 3.6 characters equaled approximately 1° of visual angle. A chin-residuum and a brow-rest were used to minimize caput movements.

Process

At the showtime of the experiment, the participant'south heart position was calibrated using a ix-point calibration filigree. Each trial started with a drift correction dot located 200 pixels to the right of the left border of the display (Fig. 1). Participants were instructed to focus on this dot, which would trigger the onset of a sentence stimulus. The distance between the fixation bespeak and the commencement of the sentence was randomly adamant, within a range of -54 to +32 pixels. Participants were instructed to read from left to right for comprehension. An invisible purlieus was divers at the finish of the sentence, such that the sentence disappeared when the eyes crossed that boundary. Next, participants were shown a question that allowed u.s.a. to check whether they had paid attention to the discussion sequence. Participants were instructed to indicate whether they had seen a given word (due east.g., "Did you encounter the word 'tabular array'?") by means of a two-push button response for, respectively, "yes" and "no" responses (probe give-and-take classification). One-half of these questions concerned a word that was present in the sentence, and the other half a word that was not present in the sentence. The probe words never contained a alphabetic character transposition. Finally, a feedback dot was presented over ii,000 ms afterward the probe word classification response (green if the response was correct or red if the response was incorrect). The sentences were presented in a different random order for each participant. Participants received ten practise trials to familiarize them with the experimental procedure.

Analyses

We used linear mixed-effects models (LMEs) to clarify our data, with items and participants as crossed random effects (including by-item and by-participant random intercepts) (Baayen, Davidson, & Bates, 2008) and with random slopes (Barr, Levy, & Tily, 2013), and with Spacing and Transposition plus their interaction as stock-still effects. The model successfully converged under this maximal random-effects structure in some just not all cases. In case of a failure to converge, we excluded the by-item random slopes (a Chi-square exam indicated that a model including the by-particular random slopes did not differ significantly from a model including the past-participant random slopes, so this was an capricious choice); and if a model so still failed to converge, we included only random intercepts. Generalized (logistic) linear mixed-effects models (GLMEs) were used to clarify the error rate and fixation probabilities. The models were fitted with the lmer (for LMEs) and glmer (for GLMEs) functions from the lme4 packet (Bates, Maechler, Bolker, & Walker, 2015) in the R statistical computing environment. The condition with normal spacing and without letter of the alphabet transpositions was used as a reference and we reported regression coefficients (b), standard errors (SE),s and t-values (for LMEs) or z-values (for GLMEs) for all factors. Fixed furnishings were deemed reliable if |t| or |z| > i.96 (Baayen, 2008). All duration measures were inverse-transformed (-1,000/elapsing) prior to analysis for the purpose of normalization.

Results

The eye-movement information of ane participant were removed prior to analysis due to a big number of heart blinks. All other participants depicted normal eye-motility behavior and responded with accuracy higher than 90% (M = 94.10, SD = 23.54) on the probe word classification trials. Response accurateness was significantly higher (b = 2.91; SE = 0.41; t = 6.97) with unremarkably spaced sentences (98.47%) compared with unspaced sentences (89.75%). Prior to analysis nosotros excluded trials containing blinks (5.04%) and trials with wrong responses on the probe give-and-take classification task (5.89%). For the local word-based analyses, we used the data apropos the three critical target words in each sentence while excluding words that were skipped during first pass (one.81%). Nosotros measured and analyzed target word fixation durations and saccade blazon probabilities (skips, refixations, regressions), initial landing positions (ILPs; the location of the first forward fixation on a word), sentence reading speed, and estimated reading difficulty (evaluated by participants during post-experiment debriefing).

Fixation durations

From the eye-tracking data, nosotros computed three fixation duration variables: Beginning Fixation Duration (FFD), which represents the duration of the fixation immediately following the first forward saccade into a word; Gaze Duration (GD), which is the sum of all fixation durations on a word before the eyes leave that discussion (first pass fixations); and Total Viewing Time (TVT), which is the sum of all fixation durations on a discussion (thus including fixations made following a regressive saccade dorsum to the discussion). These values were computed for the three critical target words in each judgement (i.e., words that involved a letter of the alphabet transposition manipulation in the transposition condition) and the average value per sentence entered in the analysis. From these data, we excluded words with values beyond 2.5 SD from the m mean (FFD: two.38%, GD: ii.46%, TVT: 2.98%). The mean duration values (in milliseconds) per experimental condition are presented in Fig. ii.

All the elapsing measures revealed pregnant effects of Spacing and Transposition, and in total viewing times there was also a significant interaction between these variables (see Tabular array 1). Transposition effects were greater in the unspaced condition (83 ms) compared to the spaced condition (67 ms). We also analyzed second-pass reading times, which represent the amount of time spent re-reading a give-and-take later on start-pass reading (Juhasz & Pollatsek, 2011). Here, we institute a significant event of Transposition (b = 0.10; SE = 0.03; t = 2.85) with longer reading times in the transposed-letter of the alphabet condition, simply neither the effects of Spacing (b = 0.02; SE = 0.05; t = 0.40) nor the interaction were meaning (b = 0.09; SE = 0.05; t = i.73).

Table ane Fixed effects from the LMEs for the Fixation Elapsing measures in Experiment 1

Total size table

Saccade-type probabilities

We calculated the probability of skipping a word (when a word is not fixated during first-pass forward eye movements), of refixating a word prior to leaving the word (within-give-and-take saccade), and of refixating a give-and-take later on leaving that word (between-word regressive saccade). The average probabilities per experimental condition are shown in Fig. iii and the results of the statistical analyses are reported in Table 2. Nosotros found that the absence of interword spaces caused a subtract in skipping probability accompanied past an increase in the probability of refixations and regressions. Letter transpositions had a meaning upshot in all three measures (Table 2), decreasing the skipping charge per unit in the spaced condition and decreasing the skipping rate in the unspaced condition and increasing refixation and regression probabilities. For skipping probabilities, we observed an interaction between Spacing and Transposition, with a greater influence of letter transpositions in the unspaced condition compared to the spaced status.

Table 2 Fixed furnishings from the GLMEs for the different measures of saccade type probabilities in Experiment 1

Full size table

Initial landing position (ILP)

Prior to statistical assay of the initial landing positions (ILPs) we start excluded values lying beyond 2.five SD from the mean (ane.99%). Table 3 provides the mean ILP per experimental condition expressed in normalized values between the beginning (0) and the stop (one) of words. The distributions of ILPs in each status are shown in Fig. 4. There was a significant effect of Spacing (b = 0.06; SE = 0.00; t = 7.54), with ILPs being closer to the beginning of words in the unspaced condition, and a meaning effect of Transposition (b = 0.01; SE = 0.00; t = 2.04), with the presence of transpositions causing the ILPs to shift slightly toward the beginning of words. The interaction between Spacing and Transposition was not significant (b = 0.00; SE = 0.00; t = 0.60).

Tabular array iii Hateful initial landing positions from 0 (the offset of the word) to 1 (the end of the word) in Experiment ane

Full size tabular array

Sentence reading times

Judgement reading fourth dimension was measured every bit the fourth dimension between presentation of the stimulus and the moment participants' eyes crossed the end boundary of the judgement. Thus, this measure gathered duration values for all the vii words of the sentence. Prior to analysis nosotros excluded values across 2.5 SD from the hateful (2.61% of trials). The average reading times (in ms) per experimental condition are shown in Table four.

Tabular array 4 Hateful sentence reading times (ms) in the iv experimental conditions of Experiment ane

Full size tabular array

We constitute a significant consequence of Spacing (b = 0.23; SE = 0.01; t = sixteen.72) and Transposition (b = 0.05; SE = 0.01; t = 5.21). The interaction between Spacing and Transposition was likewise pregnant (b = 0.02; SE = 0.01; t = 2.06) with greater transposition effects in the unspaced status compared to the spaced condition (see Tabular array four).

Estimated reading difficulty

In order to evaluate the subjective difficulty of reading in the different conditions, at the finish of the experiment we asked participants to approximate their experienced reading difficulty in each condition. To practice so, they were instructed to move a cursor on a scale from 0 to 100, and the corresponding number of the location of the cursor was ever visible. Responses were collected without time limit, and no data were excluded prior to analysis. The boilerplate values for each condition are reported in Table five.

Tabular array 5 Mean of estimated reading difficulty on a scale from 0 (very like shooting fish in a barrel) to 100 (very difficult) in Experiment 1

Total size table

We plant significant effects of Spacing (b = 28.29; SE = 3.32; t = 8.51) and Transposition (b = 17.lxxx; SE = 2.60; t = 6.83), and also a significant interaction (b = 21.29; SE = 3.68 t = v.87), with letter transpositions having a stronger effect when reading unspaced sentences compared to the usually spaced sentences (see Table 5).

Effects of vocabulary and spelling

Hither we examined the impact of participants' scores on the vocabulary and spelling dictation tests on the different dependent measures of Experiment 1, and whether these scores influenced the furnishings of Spacing and Transposition. Nosotros only report pregnant effects from LME and GLME analyses that successfully converged.

Gaze durations, total viewing times and initial landing positions were significantly influenced past vocabulary level (GD: b = 0.03; SE = 0.01; t = ii.05; TVT: b = 0.04; SE = 0.01; t = 2.27; ILP: b = 0.00; SE = 0.00; t = ii.62), with college vocabulary scores leading to shorter viewing durations and a shift of the ILP toward the middle of words. Get-go fixation elapsing, gaze duration and total viewing times were significantly influenced past spelling ability (FFD: b = 0.01; SE = 0.00; t = two.30; GD: b = 0.01; SE = 0.00; t = 2.05; TVT: b = 0.01; SE = 0.00; t = ane.98), with higher spelling scores leading to shorter viewing durations.

More interesting is the fact that vocabulary level interacted with the effects of letter transpositions in gaze durations (b = 0.02; SE = 0.01; t = 1.98), such that the interfering effect of transposing letters was greater in participants with higher vocabulary scores. Letter of the alphabet transposition effects also interacted with spelling power in gaze durations (b = 0.00; SE = 0.00; t = 2.62) and full viewing times (b = 0.01; SE = 0.00; t = 3.06).

Effects of boundary letter frequency

In these analyses nosotros report on the effects of boundary letter frequencies. Boundary letter frequency refers to the position-specific token frequency of the first and last letters in words. These analyses are motivated by the findings of Kasisopa, Reilly, Luksaneeyanawin, and Burnham (2013) showing an bear upon of such variables when reading in Thai and suggesting that these letter frequencies might act as a cue to word boundaries when reading unspaced text. Averages of the first letter and last letter frequency values (in Zipf) across the 3 critical words in each sentence were used in the LME and GLME analyses. Offset and terminal letter frequency were entered as separate variables given that Kasisopa et al. institute more robust furnishings of these two variables when analyzed separately every bit opposed to a combined bigram frequency measure out. Here we only report significant effects obtained in analyses that successfully converged.^{Footnote 4}

There was a significant three-way interaction involving starting time letter of the alphabet frequency in the gaze durations (b = 0.37; SE = 0.15; t = 2.32) and total viewing times (b = 0.34; SE = 0.xiv; t = 2.42). The 2-way interaction betwixt Spacing and Transposition (i.due east., the greater consequence of letter of the alphabet transpositions in the unspaced condition) was found to be stronger with depression get-go letter of the alphabet frequencies (GD: b = 0.29; SE = 0.10; t = 2.78; TVT: b = 0.32; SE = 0.09; t = 3.57) compared to high kickoff letter frequencies (GD: b = 0.00.; SE = 0.11; t = 0.00; TVT: b = 0.11; SE = 0.10; t = 1.03). There was also a significant three-way interaction involving last letter of the alphabet frequency in the total viewing times (b = 0.34; SE = 0.14; t = 2.42). Once again, the critical interaction between Spacing and Transposition was stronger when terminal letter frequency was low (b = 0.32; SE = 0.13; t = 2.46) compared to loftier (b = 0.thirteen; SE = 0.14; t = 0.96).

Discussion

The results of Experiment 1 showed clear effects of both the Spacing factor and the Transposition gene on the majority of our measures of reading difficulty, both in terms of judgement-level measures (sentence reading speed and estimated reading difficulty), and in terms of local eye-motility behavior concerning the three critical target words in each sentence. The eye-move results are in line with prior reports of effects of letter transpositions on fixation durations, and number of regressions and refixations (Blythe et al., 2014; Rayner et al., 2006; White et al., 2008), as well as prior reports of the influence of removing interword spaces on fixation durations, saccade-blazon probabilities, and initial landing positions (eastward.chiliad., Mirault et al., 2018; Perea & Acha, 2009; Rayner et al., 1998). Crucial, with respect to the hypothesis under test, is that we observed a significantly stronger influence of alphabetic character transpositions when reading unspaced text compared with normally spaced text in total viewing times (per disquisitional word) as well as for the sentence reading time and the estimated sentence reading difficulty. We besides establish that words containing letter transpositions were skipped more when reading unspaced text, whereas the reverse pattern was seen with unremarkably spaced text. We return to talk over this finding in the General give-and-take. Overall, this pattern of results is in line with the hypothesized greater office for word identification processes when reading unspaced text, with letter transpositions selectively perturbing this process during reading.

In additional analyses we examined how the vocabulary scores and spelling power of our participants influenced their reading beliefs. The full general pattern nosotros observed was that college vocabulary or spelling scores led to faster reading times in diverse measures. However, simply vocabulary level afflicted initial landing positions, with a shift toward the centre of words for participants with college scores. We also observed that the influence of vocabulary and spelling scores on certain elapsing measures was most pronounced in the condition with no letter transpositions. Vocabulary and spelling level had a much-reduced bear upon when reading text containing letter transpositions because participants with higher scores on these tests were more affected by interference from alphabetic character transpositions.

Finally, we found that differences in the frequency of the beginning and last letters of critical words impacted on the key interaction between Spacing and Transposition. The greater influence of letter transpositions in the unspaced condition significantly increased when first or last letter frequency was low. Low first and final letter of the alphabet frequencies increase doubt with respect to word boundaries, hence increasing the interference caused by introducing letter transpositions when there are no visual cues to word boundaries.

Experiment 2: Reading aloud

Middle-movement recordings do non actually tell us if words are correctly identified in the different weather condition, and more precisely, whether or not participants were actually identifying the basewords from which the transposed-letter stimuli were generated. Experiment 2 was therefore run in order to measure how well participants can really identify words, including the basewords of transposed-letter of the alphabet stimuli, in the different experimental conditions. To do so, we asked participants to read aloud the same set of sentences as tested in Experiment 1, and we recorded the vocal output.