Generating inferences from written and spoken language: A comparison of children with visual impairment and children with sight

The two experiments reported here investigated the ability of sighted children and children with visual impairment to comprehend text and, in particular, to draw inferences both while reading and while listening. Children were assigned into 'comprehension skill' groups, depending on the degree to which their reading comprehension skill was in line with that predicted by their decoding skill. They then read (either print or Braille) and listened to a series of novel short stories. These were followed by a series of questions, which required either the generation of inferences, or an answer that could be taken literally from the text. The results suggest that children with and without sight are comparable in their ability to draw inferences, and that children with visual impairment show an advantage for literal questions under auditory presentation. Introduction Comprehension processes have not been studied in any great detail in children with visual impairment (VI), partly because it is assumed that they will simply mirror those observed in sighted children. However, it is especially important to understand the comprehension processes of Braille readers, because greater resources are required for teaching Braille in terms of equipment, time, and teacher support. This is reflected in the delay of approximately 2 years in the development of reading in children with VI (Nolan & Kederis, 1969). The aim of the present study was to explore comprehension in children with VI in more detail: firstly to investigate whether, as is the case for children with sight, comprehension skill in children with VI is associated with better inference generation; and secondly, to examine whether an advantage for auditorially presented information shown by children with VI for digit recall is also shown for more meaningful information. Reading Braille is both like and unlike reading print. Braille has a letter-letter mapping from print and is comprised of patterns of raised dots presented in a 2 × 3 matrix. However, Braille (or Braille Grade II to be precise) employs many contractions, where commonly occurring groups of letters or words are represented by individual symbols; for example, there are symbols that represent letter strings such as 'and', 'ea', and 'ing'. Even with the use of contractions, experienced Braille readers are slower than sighted people reading print, averaging 70-100 words per minute, compared with just under 300 words per minute for sighted print readers (Foulke, 1982; Nolan & Kederis, 1969). Reading Braille is similar to reading print in that reading times in both media are affected by word length, word frequency, repetition, and semantic priming (Carreiras & Álvarez, 1993; Pring, 1984). While the literal interpretation of text involves searching for explicitly stated information, successful comprehension of text often requires the inference of information that is not explicitly stated, but implicitly implied. Full comprehension of text often requires the reader to fill in missing information by applying general knowledge, and to make links between different sections of text. While this ability to generate inferences has been extensively studied in children with sight (Cain & Oakhill, 1999; Oakhill, 1984; for a summary see Yuill & Oakhill, 1991), to date, no research has been conducted that explores the ability of children with VI to generate inferences from text. In the case of children with sight, comprehension performance has been examined under different presentation modalities, with reading and listening comprehension ability correlating positively (Nation & Snowling, 1997; Smiley, Oakley, Worthen, Campione, & Brown, 1977). The modality in which information is presented may differently influence the comprehension ability of good and less good comprehenders. Some studies have observed similarities in the performance of able and less able comprehenders whilst reading and listening (Nation & Snowling, 1997; Stothard & Hulme, 1992). However, others have reported discrepancies; for example, Miller and Smith (1990) found that poor readers showed no difference between reading aloud and listening, average readers performed better on a listening comprehension test compared with a read aloud version, and good readers showed an advantage for reading aloud over listening. In sighted children, comprehension success is associated with a greater ability to generate inferences. Many studies have shown that children who are skilled at comprehension differ from their less skilled counterparts in their inference-making skill (Cain & Oakhill, 1999; Oakhill, 1984; for a summary see Yuill & Oakhill, 1991). Most of the research exploring inference generation by skilled and less skilled comprehenders has concentrated on reading comprehension; in the present study, we aimed to investigate the role of inference making in comprehension success in both reading and listening, in children with sight, and children with visual impairment. While modest correlations have been found between reading and listening comprehension in Braille reading adults with VI (Daneman, 1988), no research has considered the ability of either adults or children with VI to generate inferences whilst reading or listening. Children with VI have shown an advantage for auditorially presented information that might extend to text comprehension. Children with VI show superior performance on the Digit span subtest of the Wechsler Intelligence Scales for Children (WISC; Wechsler, 1974), a test that requires the serial recall of verbally presented digits (Smits & Mommers, 1976; Tillman & Osbourne, 1969; Warren, 1984). This advantage is not shown when items are presented tactually using Braille (Millar, 1974). This auditory advantage may extend to other types of information, and may be affected by the complexity of the information, or by how much processing is required. In terms of comprehension, an auditory advantage may be restricted to simple information, such as literal information in text, and may not be observed when information requires additional processing, as is the case with information generated by inferences. In the present paper, an adaptation of Oakhill's (1983, 1984) study design of skilled and less skilled comprehenders was used. The conventional method of assigning children to comprehension skill groups is to match them on decoding age using the Neale Analysis of Reading Ability (NARA; Neale, 1966), and categorize them as skilled comprehenders if their NARA comprehension age matches, or is better than, their decoding age; and as less skilled comprehenders if their comprehension age is 6 months below their decoding age (e.g. Oakhill, 1983; Oakhill, Yuill, & Parkin, 1986). With special populations, like children with VI, whose numbers are small, an approach that requires selecting subsets of participants after initial screening is impractical. A common approach with many assessment tests is to examine discrepancies between observed skill, and that predicted from general ability (e.g. Wechsler Objectives Numerical Dimensions, 1996; Wechsler Objectives Reading Dimensions, 1993); we adopted a similar method here. Linear regression was used to predict comprehension age from decoding age, and children were assigned to comprehension skill groups on the basis of the discrepancy between their measured comprehension age, and their predicted comprehension age. This method is advantageous with small samples because it allows every child initially assessed to be assigned to a group. In this paper, children whose comprehension skill was less than that predicted from their decoding ability are referred to as less advantaged comprehenders, while those whose comprehension ability exceeded that predicted from their decoding age are referred to as advantaged comprehenders. Data from two experiments are reported here. Both compared the comprehension performance of children with sight and children with VI when answering literal questions and inferential questions in response to stories. Additionally, both experiments compared children who were advantaged comprehenders with those who were less advantaged comprehenders. Experiment 1 examined these factors in a reading comprehension task (printed text and Braille), and Experiment 2 used auditorially presented stories to examine listening comprehension. If the comprehension success of children with VI is akin to that of children with sight, it would be expected that children who were advantaged comprehenders would make fewer errors when answering questions for which an inference must be generated, compared with children who were less advantaged comprehenders, both when reading and listening, and regardless of vision group. If the advantage that children with VI have for auditorially presented digit strings is extended to other relatively simple types of auditory information, it might be expected that they would show an advantage for literal information, compared with information for which an inference must be generated in Experiment 2 (listening), an advantage that would not be shared by children with sight.