Journal of Research in Reading

What is already known about this topic

• Reading skills influence linguistic processing in children and adults.
• Retrieval of information is cascaded from childhood.
• Learning to read modifies previous word representations.

What this paper adds

• Children's reading skills influence attention to specific information in both image and printed word modalities.
• Skilled and less‐skilled young readers access phonological information when reading words and semantic information when looking at pictures.
• Only skilled readers are sensitive to semantic relationships between spoken and printed words.

Implications for practice and/or policy

• The speed and type of information children can access in a visual task depend on their reading skills.
• Understanding that children with similar age and IQ and in the same grade at school have different reading skills can help us to design reading programs.
• Longitudinal evaluations will help us determine the effects of reading skills on children's ability to access perceptual, phonological, and semantic information.

Studies have found differences in the processing of linguistic stimuli attributable to reading skills and years of literacy. Literacy modulates the prediction of linguistic stimuli: adults with a mean of 15 years of formal education, but not those with 2, predict a visual target before it is named, based on semantic, associative, and syntactic cues (Mishra, Singh, Pandey, & Huettig, 2012). Mani and Huettig (2014) also reported a correlation between reading skills and the ability to predict linguistic input in 8‐year‐old children exposed to a spoken semantically related verb (‘eat’), a visual target (‘cake’), and a visual distractor: good readers were better at predicting a target than poor readers. The authors suggest that the acquisition of orthographic representations of words in learning to read shapes children's preexisting lexical representations and affects their subsequent processing of linguistic stimuli.

Unlike oral language, reading acquisition requires formal instruction in decoding written alphabetic language. Learning to read modifies previous word representations and is related to phonological as well as orthographic ability. Phonological representations are likely to be adjusted when the orthographic code is acquired. Children not only establish relationships among the letters that constitute written words, but also among oral language phonemes (Mani & Huettig, 2014; Schild, Röder, & Friedrich, 2011).

Phonological awareness – the discovery that discourse can be fragmented into phonemes, which can be reordered to form different words – precedes the acquisition of reading skills (Bowyer‐Crane et al., 2008). It has also been suggested that reading acquisition changes the approach to solving phonological awareness tasks, given that orthographic and phonological information are simultaneously available (Castles & Coltheart, 2004). Although both high and low literates show similar performance in phonemic discrimination, syllable rhyme detection, and phonological sensitivity involving the manipulation of sounds, illiterates have difficulty in identifying and interchanging phonemes and syllable order, which are also phonological awareness abilities (Petersson, Reis, Askelöf, Castro‐Caldas, & Ingvar, 2000). Thus, although phonological abilities exist before reading, specific phonological abilities may be more accurate after the acquisition of a written code. The connectionist model of Plaut, McClelland, Seidenberg, and Patterson (1996), which allows word‐reading simulations of readers with and without reading difficulties, explains the processing of orthographical representations during reading. The model suggests an interaction between the phonological and semantic routes for reading: it shows how phonological and semantic aspects interact and how the nature of such interaction can be modified as readers learn and gain experience.

Beginning in childhood, hearing a word (‘cat’) activates its phonemes (/kat/) (Huang & Snedeker, 2011), visual features (Arias‐Trejo & Plunkett, 2010), and semantically related words (e.g., ‘dog’) (Mani, Durrant, & Floccia, 2012); literate adults also activate orthographic representations (Huettig & McQueen, 2007), which are linked to phoneme detection (Perre, Pattamadilok, Montant, & Ziegler, 2009; Ziegler, Ferrand, & Montant, 2004). Adults find it easier to detect orthographically consistent words, where the orthographic representation of the phoneme is the same (e.g., /b/, ‘boat’, ‘bowl’), than those that vary (e.g., /f/, ‘farm’, ‘pharmacy’) (Cutler & Davis, 2012). Adults are slower at deciding whether two words rhyme when orthography varies (e.g., ‘boat‐note’) than when it is consistent (e.g., ‘bank‐tank’) (Ziegler et al., 2004); and they are efficient at detecting phonological similarities (e.g., in the onset and coda). This work aims to shed light on whether children with different reading skills but the same years of literacy also respond to phonological similarities within written and auditory modalities.

It is unclear whether children's degree of literacy has equal effects on semantic and shape‐similarity processing. Reis and Castro‐Caldas (1997) found more difficulties in semantic and phonological memory tasks in illiterate adults than in literate ones (though the differences were significant only in phonological memory tasks). At the electrophysiological level, there are differences between adults with high and low comprehension abilities in the amplitude of the P200 and the N400 wave forms (associated with phonological and semantic processing, respectively), yet there are no differences in phonological processing (Landi & Perfetti, 2007). Low comprehension skills could therefore be explained in part by semantic processing difficulties (Landi & Perfetti, 2007).

A model proposed by Harm and Seidenberg (2004) sheds light on the processes involved in determining meaning from printed words. The mechanism can be direct lexical access (orthography to meaning), phonologically mediated lexical access (orthography to phonology to meaning), or both routes simultaneously. The activation of meaning is built in time based on a continuous input of available information, mainly from these two sources. In this model, there is a distributed semantic pattern rather than the instant access to meaning.

Reading abilities also correlate with response time in lexical decision tasks: fourth‐ and sixth‐grade children with poor reading comprehension have slower reaction times than their average counterparts (Choi & Hwang, 2010). In verbal fluency tasks, there are significant differences between literates and illiterates in producing words with a specific phoneme, but not in producing words from a category (e.g., animals) (Kosmidis, Tsapkini, Folia, Vlahou, & Kiosseoglou, 2004). Literacy also has an effect on the perceptual abilities needed to rapidly identify and name black‐and‐white as opposed to color images (Petersson, Reis, & Ingvar, 2001), mirror images as opposed to originals (Kolinsky & Verhaeghe, 2012), and two‐dimensional as opposed to three‐dimensional images (Reis, Petersson, Castro‐Caldas, & Ingvar, 2001).

Words produce a cascaded activation of related phonological, orthographic, and perceptual information in children and adults (Huang & Snedeker, 2011; Huettig & McQueen, 2007; Mani et al., 2012). Huettig and McQueen (2007) investigated adults' access to long‐term memory after exposure to an auditory word and images or words. Participants looking at images tended to retrieve phonological, shape, and semantic information, while those looking at words retrieved primarily phonological information. The competition among these types of information suggests the existence of cascaded processing, the incremental propagation of information across multiple levels of analysis (Huang & Snedeker, 2011), at least for images.

Linguistic input can activate phonological, orthographical, perceptual, and semantic representations of the processed word. The identification of auditory input favours the activation of phonological information first, followed by perceptual and semantic information (McClelland, 1979). The identification of written words facilitates access to orthographic and phonological information, followed by semantic and perceptual information (Coltheart, Rastle, Perry et al., 2001). Finally, the identification of images allows rapid access to perceptual and semantic characteristics, and then to phonological activation (Levelt, Schriefers, Vorberg, Meyer, Pechman, & Havinga, 1991). The activation can be propagated to other words through characteristics shared within the same or different sensory modality. Huettig and McQueen (2007), for example, observed that participants presented with an auditory stimulus (‘balloon’) looked first at phonologically related images (‘bath’), and then at those with a perceptual relation (of shape: ‘sun’) and a semantic relation (‘doll’). However, when the images were replaced with written words, participants looked only at the phonologically related competitor (‘bath’). This difference suggests that auditory words enhance coactivation of different representations than written ones, and that there are discrepancies in the timing of the coactivation.

The ability to retrieve information from visual stimuli is related to reading skills. Huettig, Singh, and Mishra (2011) found that highly literate adults, but not those with lower literacy, retrieve phonological information from exposure to images in an auditory–visual cross‐modal eye‐tracking task. In Experiment 1, participants listened to spoken sentences containing a critical word and looked at a visual display of four objects: a phonological competitor, a semantic competitor, and two unrelated objects. In Experiment 2, the semantic competitor was replaced by an unrelated object. In both experiments, high literates shifted their attention to the phonological competitor as soon as phonological information was available. However, low literates preferred to look at the phonological competitor only in Experiment 2. Low‐literacy participants did not engage in a struggle between different types of cognitive representations.

In image processing, the spontaneous fixation on a visual referent upon hearing a word can be the product of the partial semantic overlap between a word and an object. Ocular movements respond to the degree of coincidence between the word and the mental representation of the objects present in the visual field (Huettig & Altmann, 2005). However, in word identification, the activation of information depends on the precision and flexibility of the mental representations of words in relation to their orthography, phonology, and meaning (Perfetti, 2007). The lexical high‐quality hypothesis suggests that the quality of word representations affects reading skills such as comprehension (Perfetti, 2007). High‐quality representations consist of phonological, orthographic, and meaning characteristics that are fully specified and strongly linked, activating different types of information in relation to a particular word (Richter, Isberner, Naumann, & Neeb, 2013). Such representations are required for reading comprehension (Perfetti & Hart, 2001; Perfetti, 2007). Richter et al. (2013) reported that the efficiency and precision of orthographic, phonological, and semantic representations in a group of children (mean age 8.27 years) explained individual variability in a computerised test of text comprehension. There is a link between lexical quality and reading skills: readers with poor lexical representations are at risk of retrieving imprecise or incomplete lexical information during reading (Richter et al., 2013).

The present study uses an eye‐tracking experiment to identify the recovery of phonological, semantic, and perceptual knowledge of third‐grade children (average age 8 years) upon hearing a critical word while exposed either to images or printed words. It aims to establish whether the retrieval of information by these skilled and less‐skilled children is partially modulated by their reading skills, depending on the presentation modality (images or words). Differences between these groups would suggest an early impact of literacy on the manner in which information is retrieved, although its influence may be stronger with words than with images. Skilled readers (SRs) might possess high‐quality representations that allow them to retrieve precise information from auditory linguistic stimuli.

Method

Results

Discussion

The aim of this study was to identify the recovery of phonological, semantic, and perceptual knowledge of 8‐year‐old SRs and LSRs upon hearing a critical word while exposed to either images or printed words in an eye‐tracking experiment.

In the image modality, both SRs and LSRs preferred to look at the semantic competitor over both the phonological and shape competitors and the distractor. The proportion of looking time and the tracking trajectory showed that their next preference was the shape competitor. SRs, but not LSRs, exhibited a marginal late preference for the phonological competitor. Time‐course analysis showed that attentional shifts in SRs were initially to the shape competitor, then to the semantic competitor, and finally to the phonological competitor. However, these preferences showed a delay that may reflect a strategic effect. In LSRs, attentional shifts were initially to the semantic competitor and then to the shape competitor. This group did not show an attentional shift to the phonological competitor.

The time‐course analysis for the image modality is partially consistent with the findings in adults of Huettig and McQueen (2007), but only for SRs. Although in the present study the previsualisation stimulus time (800 ms) before the spoken critical word was similar to that of Huettig and McQueen's Experiment 1, our results were similar to those of their Experiment 2, in which the previsualisation time was shorter (200 ms). However, we did not find early attentional shifts to the phonological competitor in the image modality, as in their Experiment 2. It is possible that because children's processing is slower than that of adults (Jescheniak, Hahne, Hoffmann, & Wagner, 2006), the previsualisation time was insufficient to retrieve image names. The late retrieval of phonological similarity by SRs was possibly accomplished through a postlexical coactivation.

An alternative or complementary explanation for our results is related to children's short memory span (Vogan, Morgan, Powell, Smith, & Taylor, 2016), which causes difficulties in recalling and maintaining labels. However, children's retrieval of image names is an automatic process, and the free propagation of information causes phonological coactivation. In addition, it has been shown that 8‐year‐olds can recall four verbal items (Vogan et al., 2016). Thus, a postlexical phonological coactivation due to slow retrieval of picture names is the likely explanation for our results. Although this explanation might be valid for SRs, LSRs presented a temporal retrieval of information morphologically different from that of SRs or the adults in Huettig and McQueen (2007). The LSRs showed a faster attentional shift to semantic information, at 200 ms, and a delayed attentional shift to shape information, at 1,600 ms; they did not show attentional shifts to the phonological competitor.

The influence seen here of reading skills on retrieving the names of images coincides with reports employing other techniques, such as rapid automatised naming (RAN) tasks. In RAN tasks, participants must name images of familiar objects, digits, or letters as quickly as possible. Poor performance in RAN is related to dyslexia, due to participants' difficulties in retrieving phonological information (Di Filippo, Zoccolotti, & Ziegler, 2008). Georgiou, Parrila, Cui, and Papadopoulos (2013) examined why RAN is related to reading in children, through the manipulation of processes at different production stages: input, processing, and output. Their explanation for the relationship is that both require serial processing and the pronunciation of the stimulus names.

The ability to identify an image depends both on detection of perceptual, physical properties, and semantic processing of meaning (Farah, 1997). Exposure to words also involves activation of different components, including those that are phonological, perceptual, and semantic. The divergence between our results and those of Huettig and McQueen (2007) could be explained by difficulties for LSRs in the activation of a semantic competitor, which requires a link between a word (spoken critical word), a second word (name of semantic competitor), and an image (semantic competitor). Activation of a shape competitor requires a link between a word (spoken critical word), an image (spoken critical word), and another image (shape match). In the latter case, a comparison between the physical properties of two images is required: one not available (spoken critical word) and another available (target); this comparison involves online processing of the physical characteristics of the stimuli at a lexical representation level.

With the aim of assessing the similarity in shape of our images, an image validation study with 20 adults was conducted, after completing the experiment, to assess the perceptual similarity between a referent for the critical word and the shape competitor. This study used a Likert scale from 0 (less similar) to 7 (more similar). Although for these adults the shape competitor was more similar to the referent of the spoken critical word than the other competitors (semantic, phonological, and distractor), the scores (M = 2.88, range: 2.20–3.57) were not different from chance (3.5). These ratings suggest that the shape competitor was not highly similar to the referent of the critical word and that children thus found it difficult to detect the perceptual similarity. However, Huettig and McQueen (2007) also reported similarity ratings around chance level (M = 5.6 on a scale from 0 to 11). LSRs, unlike SRs, might have experienced difficulty in creating an online mental image of the referent of the spoken critical word to compare with the shape competitor. However, this explanation is not parsimonious, given evidence about the early categorisation of visual stimuli based on perceptual features (Colombo, McCollam, Coldren, Mitchell, & Rash, 1990) and the extension of words to perceptually similar referents (Graham & Poulin‐Dubois, 1999; Landau, Smith, & Jones, 1998). In these studies, children are generally presented with two images (target‐distractor), not four; shape similarity may be more evident in a simplified scenario in which the shape target is paired with a distractor, but not in a challenging scenario in which the shape competitor is in conflict with semantic and phonological competitors.

It has been suggested that learning to read is related to phonological processing (Blaiklock, 2004; Bowyer‐Crane et al., 2008; González, Romero, & Blanca, 1995); people with high‐level reading skills and literacy also have good phonological skills (Kosmidis et al., 2004; Reis et al., 2001). Thus, it is possible that phonological decoding abilities improve as reading and writing abilities develop. The differences between SRs and LSRs in the present study point to the existence of divergences in phonological processing in the early stages of reading acquisition, suggesting an incipient ability to transform the visual input into a phonological representation of the stimulus to compare it with the spoken critical word. The phonological processing of SRs is more similar to that previously described in low literacy adults than that of high literacy adults (Mishra et al., 2012).

In the written word modality, both SRs and LSRs preferred to look at the phonological competitor over the other options during the first 1,000 ms of the trials. SRs looked at the phonological competitor earlier (600 ms) than LSRs (1,200 ms). These results coincide with those of Huettig and McQueen (2007), who found that in reference to a previously spoken critical word, adults prefer to look at a phonological competitor early in the trial (300–600 ms in Experiments 3 and 4). In the present study, LSRs were slower at processing phonological information (1,200 ms). As previously noted, in Spanish, the onset of a phonological competitor is generally an orthographic competitor; both SRs and LSRs could benefit from this natural coincidence in the language.

In a later time window (1,100–2,000 ms) of the written word modality, SRs showed visual preference for the semantic competitor, but the visual attention of LSRs remained with the phonological competitor. Thus, although both groups detected phonological similarity between the onset of the spoken critical word and that of the phonological competitor, better reading skills facilitated identification of alternative semantic relationships. High‐quality representations might have allowed SRs in the written modality to activate different types of information, such as phonology, orthography, and semantics, upon hearing the critical word (Perfetti, 2007). Learning and experience with reading can also influence the interaction of phonology and semantics (Plaut et al., 1996). The greater fixation of SRs on the semantic competitor in the written modality indicates that they are in the process of developing high lexical representations. Incomplete retrieval of lexical information by LSRs – attention to phonological but not semantic information – may be the consequence of poor lexical representation. Although both groups of readers develop their reading abilities and refine the necessary skills to access meaning, in the written modality only SRs employ equally the direct route (orthography–meaning) and the mediated route (orthography–phonology–meaning) (Harm & Seidenberg, 2004) to determine the meaning of a word.

The sustained attention of LSRs to the phonological competitor in the written word modality reduced their probability of fixating on another competitor. Although Huettig and McQueen (2007) did not find a preference for the semantic competitor in adults, they analysed only an interval of 1,000 ms, as compared with the 2,000 ms in the present study; it thus remains unknown whether those adults had a preference for the semantic competitor after 1,000 ms. SRs in the present study might have been using reading strategies similar to those of more experienced readers, who use simultaneous grapheme decoding (lexical–semantic route), while LSRs were still decoding graphemes consecutively (sublexical route) (Dehaene, 2009). The reading scores of SRs were above the average reading percentile for their age, while the scores of LSRs were below average.

Existing evidence suggests that semantic processing abilities are not affected by reading acquisition because innate mechanisms allow for the establishment of semantic similarities, making it possible to form word associations without an orthographic code (Kosmidis et al., 2004). This evidence comes from verbal fluency tasks that show differences between literate and nonliterate adults in phonological but not semantic abilities: nonliterate adults evoke a greater number of semantically associated words than phonologically related ones (Reis & Castro‐Caldas, 1997). In the printed word modality of the present study, associating the spoken critical word with a semantically related written word required decoding and identification of the written word in order to access the mental lexicon and retrieve the associations stored in semantic memory. LSRs experienced difficulty in identifying the semantic relationship between the critical word and the semantic competitor. Choi and Hwang (2010) reported longer reaction times in a lexical decision task for low‐comprehension children than for their high‐comprehension counterparts. The authors argued that low‐comprehension participants are slower at, but not incapable of, accessing semantic information.

In the printed word modality, SRs but not LSRs had a significant preference for the semantic competitor, followed by the phonological competitor. SRs might have benefited from their reading speed, as this was the criterion by which they were categorised; however, given the correlations among reading speed, comprehension, and precision, it is likely that all three factors contributed. If the performance of SRs was due mainly to their speed, then a random pattern in their visual attention to words after the preference to the phonological competitor would be expected. However, SRs showed a systematic subsequent preference for the semantic competitor, while LSRs showed only a preference for the phonological competitor. In addition, participants had a display preview of 800 ms before the onset of the stimuli, so both groups should have had time to track the words on screen.

According to lexical processing models, phonological, orthographic, and semantic representations are important to the visual identification of words (Coltheart, Rastle, Perry et al., 2001). It is possible that school‐age children are capable of retrieving phonological representations in a rapid and precise manner, but that their retrieval of semantic knowledge is not fully developed (Nation & Snowling, 1998). Reading comprehension involves extracting meaning from written words, which relies strongly on the quality of the semantic representation (Richter et al., 2013). In reading words, LSRs did not find a relation between the spoken critical word and the displayed written competitors.

It is possible that SRs in this study had closely unified phonological, orthographic, and semantic representations – that is, high‐quality lexical representations – as is suggested by their preference for the semantic competitor. SRs may be successful at reading because they have accurate lexical representations that can be retrieved, while LSRs may have poor lexical representations, causing the retrieval of inaccurate or incomplete lexical information.

Although the preference of LSRs for the shape competitor, in the printed words condition, between 800 and 1,100 ms was not statistically significant, this result could be related to variability in fixations to distractors among individual LSRs. The preference for the shape competitor meant that children with reading difficulties needed to retrieve physical features or mental images of the words in order to retrieve semantic information and facilitate the reading comprehension process. In another cross‐modal lexical decision task (auditory‐reading), Moss, McCormick, and Tyler (1997) found semantic coactivation preceding shape activation; however, LSRs in the present study only coactivated shape information.

This study provides evidence of the differences among SRs and LSRs, at approximately 8 years of age, in their access to different representations of words in two modalities: images and printed words. Both groups showed a looking preference for the semantic competitor in the image modality and for the phonological competitor in the printed word modality. However, in the printed word modality, only SRs found a semantic relationship between the spoken critical word and the semantically related printed word, while in the image condition, only SRs were able to marginally detect a phonological relationship between the spoken critical word and the image of a phonologically related word.

This study cannot offer conclusions related to reading development in children, but research including prereaders and expert readers, as well as longitudinal evaluations, may allow analysis of the development of the ability to access perceptual, phonological, and semantic representations. Although it is beyond the scope of this study, it is worth noting that an equal number of years of formal education in children with similar age and IQs does not necessarily imply acquiring the same reading abilities. Future research should help us to understand the factors implicated in these differences and the possible consequences of classroom disparities.

Acknowledgments

We would like to thank parents, teachers, and students from the participating schools: Professor Jorge Casahonda Castillo, Alberto Lenz, Guadalupe Victoria, and Gral. Juan N. Álvarez. We are also grateful to the members of the Laboratorio de Psicolingüística for their help in testing participants.

This research was supported by a grant awarded to the third author: Investigación Científica Básica CONACyT (Grant No. 167900).