The Role of Linguistic Ambiguity in Understanding and Improving Children's Text Comprehension Nicola Yuill and Kate Easton CSRP 296 Running Head: Linguistic Ambiguity in Text Comprehension Acknowledgements This work was supported by a Research Development Grant from the University of Sussex. Thanks to Donnacha O'Donovan for help with reading tests, David Hitchin for advice on statistics, Jane Oakhill for help with coding and Graham Hole for comments on the manuscript. Requests for reprints should be addressed to the first author at the above address. Email address: nicolay@uk.ac.susx.cogs 1 Linguistic Ambiguity in Text Comprehension _T_h_e _R_o_l_e _o_f _L_i_n_g_u_i_s_t_i_c _A_m_b_i_g_u_i_t_y _i_n _U_n_d_e_r_s_t_a_n_d_i_n_g _a_n_d _I_m_p_r_o_v_i_n_g _C_h_i_l_d_r_e_n'_s _T_e_x_t _C_o_m_p_r_e_h_e_n_s_i_o_n _A_b_s_t_r_a_c_t Previous work suggests that poor text comprehenders differ from good ones in linguistic awareness. Two studies investigated the relation between appreciation of linguistic ambiguity and children's text comprehension skill. In Experiment 1, 29 8-11 year-olds recalled, explained and rated funniness of 16 riddles requiring either high or low levels of linguistic awareness. For high-awareness riddles, recall and explanation were significantly related to reading comprehension skill with reading accuracy and age partialled out. There was no such relation for low-awareness riddles. This shows that the relation is due specifically to appreciation of language ambiguity, rather than to general verbal skills. Funniness was unrelated to all other measures. Experiment 2 used riddles and other word games involving ambiguity to train 7-8 year-olds with good or poor comprehension. Control groups were given training with humorous materials not involving linguistic ambiguity. After 7 sessions, trained children at both comprehension levels showed significantly greater improvements in comprehension skill than the control groups. This finding reinforces the claim that comprehension skills are linked to appreciation of linguistic ambiguity. 2 Linguistic Ambiguity in Text Comprehension Donaldson (1978) argued that one of the fundamental changes that occurs when children learn to read is that they come to treat language as an object of thought, whereas previously they had just used language to communicate. Similarly, Tunmer and Bowey (1984) have hypothesised that children's initial focus on meaning in using spoken language has to be set aside when they learn to read. In mastering the skills needed to decipher words, they have to ignore meaning. Once decoding skills are secured, children have to 'put humpty-dumpty back together again' (ibid, p. 162) by paying attention to the sentential context of words, rather than the words in isolation. A series of studies by Oakhill and colleagues (e.g. Oakhill, 1982; Yuill & Oakhill, 1991) have identified a group of 7-8 year-old children who seem to find this transition from text decoding to text comprehension difficult. Despite having decoding and vocabulary skills at or above their chronological age, they show deficits in inferential skills. In particular, these authors suggested that poor comprehenders do not monitor their comprehension and seem unaware of the need to construct meanings from text. Given these deficits in linguistic awareness, such children can be expected to have difficulties in recognising and dealing with linguistic ambiguity. The present paper investigates the link between text comprehension ability and understanding of linguistic ambiguity using two methods. Experiment 1 investigates the correlation between comprehension skill and the ability to understand word play in riddles and jokes, and Experiment 2 is a training study using riddles and linguistic ambiguity as a means to improve children's comprehension. _E_x_p_e_r_i_m_e_n_t _1: _R_e_l_a_t_i_o_n _o_f _R_i_d_d_l_e _U_n_d_e_r_s_t_a_n_d_i_n_g _t_o _T_e_x_t _C_o_m_p_r_e_h_e_n_s_i_o_n The ability to reflect on different interpretations of a text is a _m_e_t_a_l_i_n_g_u_i_s_t_i_c skill. The riddle is a linguistic device that makes explicit use of ambiguity for humorous effect, and many studies have charted age changes in the understanding of riddles of various types. For example, Yalisove (1978) asked children to explain riddles and to choose the best answer for them. Riddles involving conceptual tricks were easiest, understood by many children in 1st-3rd grade, followed by those with linguistic ambiguity, at 3rd to 6th grade, and hardest were riddles with absurdity, which even some 10th graders and college students struggled to explain. Shultz (1974) looked in more detail at linguistic ambiguity and showed that lexical ambiguities were understood at a younger age than syntactic ambiguities. Hirsh-Pasek, Gleitman & Gleitman (1978) presented some evidence that ambiguities hinging on semantic properties (e.g. lexical ambiguity) were easier to explain than those resting on more superficial syntactic representations, e.g. phonological ambiguity. There has been much speculation that riddle comprehension is related to reading skill, but less in the way of strong evidence. Hirsh-Pasek et al. (1978) compared the ability to explain riddles in good and poor readers from grades 1 to 6. Reading skill was assessed by the school reading specialist, although the authors acknowledge that the range of skill was restricted, and 'poor' readers were probably about average when compared to the general population. Poor readers in all age groups were less able to explain riddles than the good readers were. Similar conclusions were reached by Fowles and Glanz (1977), who found an age-independent relation between reading ability and riddle comprehension. They asked fourteen children aged between 6 and 9 to retell and explain riddles read out by an experimenter. Half the 3 Linguistic Ambiguity in Text Comprehension children were above-average readers, as identified by their teachers, and half were below average. No statistical tests were reported, and the sample size was small, but the good readers were better at explaining riddles than the poor ones. Both these studies suggest a relation between riddle comprehension and reading skill, but in both cases, reading skill was related to explanation for all types of riddle equally. It is likely that poor readers would perform more poorly on a whole range of verbal tasks, perhaps because they are less able to understand instructions and to give the kinds of articulate explanations required. The authors of both studies postulate that riddles somehow tap metalinguistic awareness, but the precise relation between this and general reading skill is not clear, particularly as the results might be explained in terms of general verbal talent. A more precise account of the relation between reading ability and metalinguistic awareness would need to specify the relation between a particular aspect of reading skill and particular types of riddle, and would need to demonstrate that appreciation of some riddles and not others is related to reading skill. The two previous studies mentioned used a global, subjective measure of reading skill, probably including aspects of decoding, comprehension and general verbal ability. Tunmer and Bowey's model of metalinguistic skill in relation to learning to read proposes that when children have mastered decoding, and are beginning to develop text comprehension skills, they need to refocus their attention onto meaning. To the extent that riddles demand an explicit awareness of semantic ambiguity, then, they should be related to reading comprehension ability. There is already good evidence that basic decoding skills are related to understanding of riddles involving orthographic word play. Mahony and Mann (1992) showed that reading skill (word attack and word identification) in second graders correlated with the ability to pick the correct punchline for phoneme/morpheme riddles but not for a group of 'control' riddles which did not involve morphophonological ambiguity. There is also some preliminary evidence that reading comprehension is related to riddle understanding from two pilot studies reported by Yuill and Oakhill (1991). In the first, two groups of four 7-8 year-old children matched in age and decoding skills but differing in comprehension ability were given 2 riddles in each of 7 categories of ambiguity, and had to retell them and explain them to a peer who claimed not to get the joke. Skilled comprehenders were more accurate than less-skilled ones at retelling all types of riddle, and better at explaining them, although this varied according to the type of riddle. In a second study, an unselected group of 20 7-8 year-olds were given 16 riddles divided into four types, to retell and explain to the experimenter. Because of methodological differences, the children in this study were much poorer at providing explanations, and the results for riddle retelling were of greater interest. There was a significant correlation between riddle recall and comprehension skill when the effects of age and reading accuracy were partialled out, whereas correlations of riddle retelling with reading accuracy were not significant. These two studies suggest that riddle appreciation is correlated specifically with comprehension skill, rather than with general reading ability. 4 Linguistic Ambiguity in Text Comprehension However, the relation between comprehension skill and riddles of particular types is unclear, as the two studies gave inconsistent results. This may have been because the sample sizes and number of riddles of each type were very small. Also, the authors presented no predictions about relative difficulty of riddle types, as they had no clear account of the demands of different riddles. The concept of 'linguistic awareness' or 'metalinguistic skill' is notoriously hard to define (see e.g. Gombert, 1992), and Gombert seems to suggest that all linguistic humour involves metalinguistic skill. However, different kinds of riddles draw on the awareness of different features of language, and may require different degrees of metalinguistic sophistication. Some riddles require metalinguistic awareness in the strict sense of involving terms for linguistic entities such as 'word': e.g. 'What word is loud, even when you say it softly? Loud.' These involve a contrast between the use and mention of a word, and are here termed 'metalinguistic' riddles. Other riddles, although not metalinguistic in the same strict sense, play on listeners' expectations about language. For example, the studies reported by Yuill & Oakhill (1991) used a category called 'pragmatic', which involves violations of expectations about the speaker's intent. These riddles pose a question that gives too much specific information, violating Grice's (1975) maxim of quantity: the speaker must be as explicit as the situation demands but not more so, e.g. 'Why do firemen wear red braces? To keep their trousers up'. A sufficient question for the given answer would be 'Why do people wear braces?'. A third category of riddle (more usually given in joke form) with apparently high metalinguistic demands is a type not identified in previous studies, termed here the 'word compound' joke. These rest on the interpretation of word compounds such as 'Swiss roll' and 'watch dog'. In jokes involving such compounds, the relation between the two words has to be 'unpacked' and re-assembled in a different way, e.g. 'Would you like to buy a pocket calculator, sir? No thanks, I know how many pockets I've got.' These three types of riddle were all judged to make high demands on metalinguistic skills. In word-compound riddles, listeners have to construct an entirely new relation between two words that is inconsistent with their usual meaning. Lexical riddles too involve flipping between two different meanings of a word, but seem to involve a lesser degree of linguistic awareness than word-compound jokes. In lexical riddles, unlike word- compound ones, the ambiguous word has two equally well-known meanings, and the listener has to construct different interpretations of the sentence for each meaning, e.g. 'Why was the crab arrested? Because he kept pinching things.' Another category of riddle not involving any kind of linguistic ambiguity is the 'absurd' riddle, which entails treating a silly question as if it were sensible, e.g. 'How do you get 6 elephants into a mini? Three in the front and three in the back.' Because linguistic ambiguity is not involved, explanation of this type should not demand metalinguistic skills: they depend more on knowledge of the cultural assumptions about riddles. In summary, metalinguistic, pragmatic and word-compound riddles, which require a high degree of linguistic awareness, should relate more closely to comprehension skills than lexical and absurd riddles, which do not do so. Using the latter set of riddles acts as a control against the possibility that appreciation of riddles of all types rests equally 5 Linguistic Ambiguity in Text Comprehension on comprehension ability, and that children with poorer comprehension simply do not understand the task. We would expect that comprehension should be more highly related to appreciation of riddles requiring linguistic awareness than to other types of riddle. Various different measures of riddle appreciation have been used in the previous literature -- explanation, recall and humour ratings. As Yuill and Oakhill (1991) pointed out, there is no reason to expect that humour ratings should be related to understanding of a riddle, because children can find a riddle amusing for many different reasons. For example, many children in their study liked the 'fireman' riddle (see 'pragmatic' category, Table 1), because they were amused by the slapstick element of trousers falling down, but did not understand how the listener is deceived into thinking that the colour of the trousers is relevant. Their results supported this assumption: funniness ratings were not related to ability to explain the riddles, particularly for less-skilled children. The utility of recall and explanation scores seems to depend on the design of the study. Yuill & Oakhill found that reading comprehension was related to riddle _e_x_p_l_a_n_a_t_i_o_n_s when the method involved children explaining the riddle to an uncomprehending peer, but that comprehension skill was more strongly related to _r_e_c_a_l_l when the children had to explain the riddle to an adult. They suggest that explaining to a peer, who presumably did not understand, made more sense to the children than explaining to an adult who had told them the riddle, and who presumably understood it. In the present study, as children explained the riddle to the experimenter, we expected that recall would be the more sensitive measure. However, explanations were also sought in case the children in the present study, who were slightly older than those in Yuill & Oakhill's work, found the task any easier. _M_e_t_h_o_d _S_u_b_j_e_c_t_s. Twenty-nine 8-11-year-olds (mean age 9yrs 9 months, s.d. 9.0 months; 16 girls, 13 boys) from a local primary school participated in the study. All had received the Neale Analysis of Reading Ability, Form C (Neale, 1966), which requires children to read aloud short stories and then to answer comprehension questions from memory. The test gives separate measures of reading accuracy (number of words read correctly) and comprehension (questions correct). Mean accuracy score was 58.1, s.d. = 20.5 and mean comprehension score was 18.9, s.d. = 7.7. _M_a_t_e_r_i_a_l_s. Sixteen riddles and jokes from 5 categories, shown in Table 1, were used. There was complete agreement on categorisation between the first author and a blind rater. The first 3 categories in the table -- metalinguistic, word-compound and pragmatic -- were judged to make high metalinguistic demands, unlike the last 2 categories, lexical and absurd. There were four examples of each riddle, except for pragmatic and absurd ones. A potential difficulty with these two categories is that they rest on the listener's ignorance of a general convention, and that, given a series of such riddles, a listener could predict the general type of answer. That is, the tricks they play on a listener cannot be repeated very often before they cease to be a surprise. For this reason, we used only two examples of each of these. -------------------------- Table 1 about here -------------------------- 6 Linguistic Ambiguity in Text Comprehension _P_r_o_c_e_d_u_r_e. Each child was seen individually in a separate room. The riddles were written on cards which were shuffled before each session. The female experimenter, who was blind to the child's comprehension score and to the hypotheses, asked the child to help in judging and explaining some riddles and jokes. She read out each riddle, giving the child a chance to guess the punchline before revealing it, then asked for a rating of funniness (thumbs down = not funny, thumbs level = quite funny or thumbs up = very funny). The child then re-told the riddle and was asked to explain why it might be funny. At the end of the session children were asked about their favourite jokes and thanked for their participation. _S_c_o_r_i_n_g: Two raters showed complete agreement on scores for a sample of 3 protocols. _F_u_n_n_i_n_e_s_s _r_a_t_i_n_g_s: 'Not funny' scored 0, 'quite funny' scored 1 and 'very funny' scored 2. This yielded a maximum possible score of 8 for metalinguistic, word-compound and lexical riddles, and 4 for pragmatic and absurd riddles. _R_e_c_a_l_l: Children's retellings were scored 1 if they maintained the joke, even with changes in wording, and 0 if they missed the point or if no response was given. This yielded a maximum possible score of 4 for metalinguistic, word-compound and lexical riddles, and 2 for pragmatic and absurd riddles. _E_x_p_l_a_n_a_t_i_o_n_s: These were scored 3 if they explicitly mentioned both meanings or the absurdity, 2 if children could give an explicit account after prompting for 'another meaning', 1 if there was some implicit awareness (e.g. the child used both forms of an ambiguous word or phrase but did not comment on their relation) that could not be explicated after prompting, and 0 for no response or a completely inadequate answer. The maximum possible score for each riddle category was 12 for metalinguistic, word-compound and lexical, and 6 for pragmatic and absurd. _R_e_s_u_l_t_s The main issue of interest was whether appreciation of each category of riddle was correlated with comprehension skill independently of accuracy score and chronological age. We expected that reading comprehension would be significantly related to appreciation of 'high awareness' riddles (metalinguistic, word-compound and pragmatic), but not to 'low awareness' riddles (lexical and absurd). Inspection of the data suggested that the scores for some of the riddle categories were not normally distributed, so Kendall's partial correlation was computed. Since Kendall's r only allows a single variable to be partialled out, parametric correlations, partialling out both accuracy score and age, were also computed, to provide some check on any differences between the two methods of partialling out the variables. _R_e_c_a_l_l 7 Linguistic Ambiguity in Text Comprehension The mean recall score over all riddle types was .75 (where 1 = correct recall). Kendall's partial correlation was computed for each of the 5 categories of riddles separately, and for the combined categories of high and low awareness, with reading accuracy score partialled out, and with chronological age partialled out. The partial correlations for recall of high awareness riddles were significant with accuracy partialled out, r (27) = .36, p<.005, and with age partialled out, r (27) = .32, p<.01. Those for recall of low awareness riddles were not significant, r<.11 for both categories. For the individual categories, both metalinguistic and word-compound riddle recall correlated significantly (all p<.05) with comprehension when accuracy and age were partialled out. The correlations for metalinguistic riddles were r (27) = .25 and .23 partialling out accuracy and age respectively. The correlations for word-compound riddles were r (27) = .36 and .31 with accuracy and age partialled out. The correlations with pragmatic riddles were not significant, r<.15. Although there is no available test of the difference between nonparametric correlations, the parametric analysis showed that the partial correlation for high awareness riddles, r(27) = .63, was significantly greater than that for low awareness riddles, r (27) = .17, t (26) = 2.38, p<.025. It was not the case that low awareness riddles failed to correlate with anything: for example, recall of absurd riddles correlated negatively with reading accuracy at p<.02. _E_x_p_l_a_n_a_t_i_o_n _S_c_o_r_e_s Despite being older than the children in Yuill & Oakhill's study, children were still rather poor at explaining riddles: the mean score was .92 out of the maximum possible score of 3. Correlations with comprehension age were in general lower than for the recall scores. The combined high awareness riddles correlated with comprehension at r (27) = .25, p<.05 with age partialled out, but the correlation with accuracy partialled out did not reach significance, r = .20, p<.07. Correlations for combined low awareness riddles were low and nonsignificant, both <.10. Of the individual categories of high awareness riddles, metalinguistic and pragmatic riddles correlated significantly with comprehension when age was partialled out, r = .22, p<.05 and r = .28, p<.01 respectively, but word compound riddles did not, r = .16, and when accuracy was partialled out, only pragmatic riddles showed a significant correlation, r = .23, p<.05. The individual categories of low awareness riddles showed no significant correlations with comprehension, r <.16 for each category. _F_u_n_n_i_n_e_s_s _r_a_t_i_n_g_s The average rating was 1 (quite funny). As expected, these ratings did not correlate with comprehension score, either with age or accuracy partialled out, with Kendall's partial rank r <.17, p>.10 in each case. _D_i_s_c_u_s_s_i_o_n As predicted, comprehension ability was significantly related to riddle recall independently of reading accuracy and age, but only for riddles judged to make high demands on linguistic awareness. There was no such relation for riddles not demanding such awareness, demonstrating that children's general recall abilities were not a simple function of 8 Linguistic Ambiguity in Text Comprehension comprehension skill. Earlier studies suggesting relations between reading skill and riddle appreciation did not provide clear evidence of a link: Fowles and Glanz (1977) used a very small sample and report no statistics, while Hirsh-Pasek et al. (1978) found such a relation for a sample of average and very good readers for all kinds of riddles. It is possible that this overall correlation reflects differences in verbal articulation skills and intelligence in general. The present study shows that better comprehenders were no more likely to recall riddles correctly than poorer comprehenders when the riddles required little or no linguistic awareness, but were better at recalling riddles that did make such demands. Previous studies have not distinguished between reading accuracy and reading comprehension, while the present study makes it clear that riddle recall was related to comprehension skill, rather than reading accuracy. This complements nicely the study by Mahony and Mann (1992), showing a relation between reading accuracy and appreciation of morphophonological riddles. The pattern of relations found for recall was repeated, but more weakly, in the explanation scores. This weaker relation supports the earlier suggestion by Yuill & Oakhill (1991) that recall may be a more sensitive measure of riddle appreciation than explanation when children are constrained by having to explain riddles to a knowledgeable adult experimenter. The finding of no relation of comprehension skill to funniness rating confirms the idea that funniness ratings cannot be seen as a clear measure of the understanding of ambiguity. Overall, funniness ratings failed to correlate with any other measure, and it was clear from children's responses that reasons other than the intended ones, such as pure incongruity between question and answer, could make a riddle funny. Shultz and Horibe (1974) separated out the incongruous elements of jokes from the resolution of ambiguity, and found that children from the age of 6 preferred 'jokes' with incongruity but no resolution over those with neither element, although 8 year olds preferred jokes with resolutions to unresolved but incongruous 'jokes'. In the present study, some children enjoyed the incongruity even if they did not understand the joke: for example, one child found the idea of a sausage rolling down a hill (see Table 1) 'weird' and therefore quite funny. Conversely, other children perceived the incongruity but did not find it amusing: for example, one child understood the misleading implication about the question of firemen's braces (see Table 1), but thought it was 'silly' and rated it not funny. _E_x_p_e_r_i_m_e_n_t _2: _U_s_i_n_g _L_a_n_g_u_a_g_e _A_m_b_i_g_u_i_t_y _t_o _I_m_p_r_o_v_e _T_e_x_t _C_o_m_p_r_e_h_e_n_s_i_o_n In their study of riddle comprehension, Fowles and Glanz (1977) speculated that riddles could be used as a technique for improving children's reading skills. Given that Experiment 1 showed riddle appreciation to be related specifically to comprehension skills, any remedial effect of using riddles should influence comprehension skill rather than other aspects of reading. Word games in general seem a very useful technique for improving cognitive skills in young children, perhaps because such games address awareness of linguistic features without requiring explicit discussion of study skills. Commonly, comprehension skills are addressed by training explicit study skills, but this approach may be less useful for young children than the use of word games they engage in naturally at around this age. For example, Bryant & Bradley (1985) have shown how the everyday activity of rhyming games can play a part in learning to decode the written word because it 9 Linguistic Ambiguity in Text Comprehension increases children's awareness of _s_o_u_n_d_s and corresponding letter patterns. In a similar way, it is possible that practice with word games involving linguistic ambiguity, such as riddles and puns, is a way of fostering awareness of alternative _m_e_a_n_i_n_g_s, and hence comprehension skills. Although there are several studies of children's comprehension of riddles and ambiguity, there appears to be no work that uses ambiguity as a training technique. Given that riddles are thought to tap syntactic and pragmatic awareness and that this type of awareness has been found to be an important factor in children's text comprehension (Tunmer, 1989), training in such skills should be effective. There are several ways of focusing children's attention on alternate interpretations of text. One method is to help children develop their skills in explaining ambiguity in riddles, and to make up their own riddles. However, it is also possible to use ambiguous stories (e.g. see Yuill & Oakhill, 1988) and communication games of the type that have been used in research on children's understanding of linguistic ambiguity (e.g. Glucksberg, Krauss & Weisberg, 1966). An appropriate control group for a training study needs to involve all the elements of the training except for the crucial ingredient of fostering linguistic awareness of alternative meanings. In addition to individual attention, familiarity with the experimenter and other special features experienced by children involved in the study, the ambiguity training group were being exposed to humorous material. The control group therefore read and made up funny stories, as well as focusing on linguistic awareness in relation to _s_o_u_n_d_s rather than meanings. It was hypothesised that children in the ambiguity training group would improve their comprehension skills more than children in the control group, independently of any increase in decoding skills. In fact, decoding skills might even be fostered more in the control group, given the focus on awareness of sounds. We also expected poor comprehenders given training to improve more than good ones, since Yuill and Oakhill (1991) suggest that the first group are particularly deficient in metalinguistic skills. _M_e_t_h_o_d _S_u_b_j_e_c_t_s: Eighteen good and eighteen poor comprehenders were selected using the method devised by Oakhill (1982): i.e., they were matched on word recognition skills using the Gates-MacGinitie Primary One Vocabulary test and on decoding skills (regressed scores) using the Neale Analysis of Reading Ability (Neale, 1966), Form C, and were significantly different in comprehension, again on the Neale Analysis. The characteristics of each group are shown in Table 2. Children in each of the two skill groups were allocated to the ambiguity training group or a control group to give four groups: trained less-skilled, control less-skilled, trained skilled and control skilled. Allocation was random subject to the constraints that the two treatment groups were equated on average reading ability and, for logistic reasons, that in each school there were subgroups of 3-4 children in the same condition. Despite the strong constraints, the trained and control groups within each skill group were not significantly different on any of the selection criteria, all ts <1, with the sole exception of the control good comprehender group, who tended to be slightly younger than their trained counterparts, t (17) = 2.07, p<.06. Furthermore, there were seven females and two males in each group except for the trained less-skilled 10 Linguistic Ambiguity in Text Comprehension group, which had three males and six females. -------------------------- Table 2 about here -------------------------- _P_r_o_c_e_d_u_r_e: The experimenter saw children in groups of 2-6, once a week over 7 weeks. Composition of each group was largely dictated by which class the children were in. Children participated in the same subgroup and good and poor comprehenders were seen separately. The average length of each session was about 30 minutes. Within two weeks of completing the training, children were given a series of post-tests (see below). _A_m_b_i_g_u_i_t_y _t_r_a_i_n_i_n_g _g_r_o_u_p_s The first session was devoted to explaining double meanings of single words, and some examples of their use in jokes, followed in the second session by double meanings in whole sentences. In Session 3, children made up their own jokes using word compounds which were provided. The next two sessions involved a communication game with ambiguous messages derived from sets of pictures (e.g. see Pratt & Bates, 1982) and a clue-construction game in which children had to think of a clue that would help someone choose between two words that were either similar in meaning (e.g. river-ocean) or dissimilar (wash-give), as used by Asher and Parke (1975). Session 6 dealt with riddles of the type defined above as 'metalinguistic' and the final session involved highly abstract and ambiguous stories (see Yuill & Joscelyne, 1988, for examples). _C_o_n_t_r_o_l _g_r_o_u_p_s In the first two sessions, children read funny stories, following which they made up their own 'silly stories'. The third session consisted of finding rhyming words and inventing limericks, followed in the fourth session by a rhyme detection task, picking out words from a list that did not rhyme. Session 5 used the phoneme awareness task devised by Tunmer, Herriman & Nesdale, 1988, in which children tap out the phonemes of words. The sixth session involved matching middle sounds of words and the treatment finished with a final funny story. All the funny stories involved elements of slapstick and incongruity rather than specifically linguistic humour. _D_e_p_e_n_d_e_n_t _m_e_a_s_u_r_e_s All children had previously been given the Neale test (Form C) for selection purposes. The main measure of change was performance on an alternate form of this test (Form B). We also wished to establish whether children improved on the tasks given in training, so we devised two more specific post-tests. The 'jokes' task was a series of ten jokes using linguistic ambiguity of various types, which the children had to re-tell after rating their funniness, in a similar way to the method of Experiment 1. The rhyming task was a list of 74 words, divided into 11 groups of 6-10 words each. The words in each group had a similar final sound although they varied in spelling, e.g. snare, fair, lair, pear, share. In each group there were 1-3 exceptions, e.g. 'mere' for the above group. Children had to find the non-rhyming words in each list as quickly and accurately as possible. 11 Linguistic Ambiguity in Text Comprehension _R_e_s_u_l_t_s _C_o_m_p_r_e_h_e_n_s_i_o_n There was quite a high overall increase in comprehension score for children in both treatment groups and both skill groups, with an average gain of 14 months. This general improvement presumably occurred because of the passage of time, practice on the test and on the reading tasks done in small groups. The differences between Neale comprehension age before and after training in the two treatment groups and two skill groups, shown in Table 3, were analysed using analysis of variance with skill and treatment groups between subjects. There was a main effect of treatment group, F (1, 32) = 4.18, p<.05. As expected, children given ambiguity training showed greater increases in comprehension age than those in the control group, means = 17 months and 11 months respectively. The same effect occurred when increases in raw comprehension scores were used. There was also a marginal effect of skill group, F (1, 32) = 3.73, p<.06, although this did not appear for the raw comprehension scores. The less skilled comprehenders tended to improve more than the skilled ones, means = 16.8 and 11.2 months respectively. There was no significant interaction between skill group and training, F <1. -------------------------- Table 3 about here -------------------------- There was clearly individual variation in the extent of the increase in comprehension scores, and it is therefore also useful to examine the comprehension outcomes for individual children. Children were classified as 'high' if their comprehension age was above both chronological and accuracy age, 'low' if it was below both of these, and 'mixed' if it was below either chronological or accuracy age. The resulting classification of each group is shown in Table 4. If children who showed any improvement ('mixed' or 'high' comprehenders) are combined, more of them appear in the less-skilled group given training than in the control group, Fisher's exact test p <.05. There was unsurprisingly no such significant relation for the skilled children, as all but one child continued to be classified as high or mixed. -------------------------- Table 4 about here -------------------------- _A_c_c_u_r_a_c_y Accuracy ages increased overall by about 8 months. The effects on comprehension did not seem to be moderated by changes in accuracy scores, as an anova on the accuracy age differences showed no significant effects or interactions, all Fs <1. Furthermore, there was no correlation between comprehension improvement and accuracy improvement, r (35) = -.05. 12 Linguistic Ambiguity in Text Comprehension _O_t_h_e_r _p_o_s_t-_t_e_s_t_s The 'jokes' post-test required children to retell a series of 10 jokes. Two subjects from the trained less-skilled group were absent. The scores were analysed by anova, with treatment condition and skill group as between-subjects factors. Although ambiguity-trained children did better than control children, means = 5.19 and 3.78 out of 10 respectively, the main effect of treatment was not significant, F (1, 30) = 1.90, p<.18. There was no significant effect of skill group, F (1, 30) = 1.23, means = 3.87 for less-skilled and 4.94 for skilled children. Children also completed a 'rhymes' post-test, described above. Using analysis of variance as above, there was a marginal effect of treatment, F (1, 32) = 3.81, p<.06. Children in the control group tended to score higher on this task than ambiguity-trained children, means = 59.67 and 55.28 respectively. The effect of skill group did not reach significance, F (1, 32) = 2.74, p<.11, although skilled children tended to score higher than less-skilled ones, with means of 59.33 and 55.61 respectively. _D_i_s_c_u_s_s_i_o_n This study provides preliminary evidence that riddles and ambiguity can indeed be useful in advancing children's text comprehension. However, the effects of ambiguity training were not specific to poor comprehenders: children in the trained groups improved regardless of their initial comprehension level. This is surprising, because the work by Yuill and Oakhill (1991) on such groups of children shows that poor comprehenders have poorer metalinguistic skills in general: they are less likely to pick out the main point of a story, less likely to notice inconsistencies in stories and find it harder to resolve anomalies in text when processing demands are high. The present Experiment 1 also shows a link between comprehension skill and ability to recall riddles involving complex linguistic ambiguity. Furthermore, Yuill and Oakhill report training studies showing a specific benefit for poor comprehenders. It seems that while good comprehenders are superior in linguistic awareness, they can show further gains given training of the sort used here. Although this tells us less about the differences between good and poor comprehenders, it means that the training technique is applicable more widely than the techniques used previously by Yuill and Oakhill. Rates of improvement were quite variable, and more work is needed to try and specify the processes by which comprehension increases occur. This would lead to a better theoretical understanding of the link between linguistic awareness and normal comprehension processes, and would help in developing more effective training programmes. The two studies reported here provide convergent evidence, using different techniques, that linguistic awareness is implicated in the development of text comprehension skills. More specifically, the results suggest that awareness at the syntactic and pragmatic levels is of greater importance in this respect than awareness at the lexical level. This complements the study by Mahony and Mann (1992), which indicated that basic reading accuracy skills are related to the ability to provide appropriate punchlines for morphophonological jokes. In their paper, they called for evidence of the relation between riddle appreciation and comprehension skills. The present paper provides that evidence, but 13 Linguistic Ambiguity in Text Comprehension further work will be needed to examine the degree to which accuracy and comprehension skills are independently related to riddle appreciation, and to investigate how the relations between reading skill and riddles change as reading skills develop. 14 Linguistic Ambiguity in Text Comprehension _R_e_f_e_r_e_n_c_e_s Asher, S.R. & Parke, R.D. (1975). Sampling and comparison processes. _6_7, 64-75. Bryant, P.E. & Bradley, L. (1985). _C_h_i_l_d_r_e_n'_s _R_e_a_d_i_n_g _P_r_o_b_l_e_m_s: _P_s_y_c_h_o_l_o_g_y _a_n_d _E_d_u_c_a_t_i_o_n. Oxford: Blackwell. Donaldson, M. (1978). _C_h_i_l_d_r_e_n'_s _M_i_n_d_s. Glasgow: Fontana. Ferguson, G.A. (1976). _S_t_a_t_i_s_t_i_c_a_l _A_n_a_l_y_s_i_s _i_n _P_s_y_c_h_o_l_o_g_y _a_n_d _E_d_u_c_a_t_i_o_n. McGraw-Hill. Fowles, B. & Glanz M.E. (1977). Competence and talent in verbal riddle comprehension. _J_o_u_r_n_a_l _o_f _C_h_i_l_d _L_a_n_g_u_a_g_e, _4, 433-452. Glucksberg, S., Krauss, R.M. & Weisberg, R. (1966). Referential communication in nursery school children: Method and some preliminary findings. _J_o_u_r_n_a_l _o_f _E_x_p_e_r_i_m_e_n_t_a_l _C_h_i_l_d _P_s_y_c_h_o_l_o_g_y, _3, 333-342. Gombert, J.E. (1992). _M_e_t_a_l_i_n_g_u_i_s_t_i_c _D_e_v_e_l_o_p_m_e_n_t. Harvester Wheatsheaf. Grice, H.P. (1975). Logic and conversation. In P. Cole & J.L. Morgan (eds.), _S_y_n_t_a_x _a_n_d _S_e_m_a_n_t_i_c_s: _V_o_l. _7. _S_p_e_e_c_h _a_c_t_s. New York: Academic Press. Hirsh-Pasek, K., Gleitman, L.R. & Gleitman, H. (1978). What did the brain say to the mind? A study of the detection and report of ambiguity by young children. In A. Sinclair, R.J. Jarvella & W.J.M. Levelt (Eds.) _T_h_e _C_h_i_l_d'_s _C_o_n_c_e_p_t_i_o_n _o_f _L_a_n_g_u_a_g_e. New York: Springer. Mahony, D.L. & Mann, V.A. (1992). Using children's humour to clarify the relationship between linguistic awareness and early reading ability. _C_o_g_n_i_t_i_o_n, _4_5, 163-186. Neale, M.D. (1966). _T_h_e _N_e_a_l_e _A_n_a_l_y_s_i_s _o_f _R_e_a_d_i_n_g _A_b_i_l_i_t_y. (2nd ed.). London: Macmillan Education. Oakhill, J. (1982). Constructive processes in skilled and less skilled comprehenders. _B_r_i_t_i_s_h _J_o_u_r_n_a_l _o_f _P_s_y_c_h_o_l_o_g_y, _7_3, 13-20. Pratt, M.W. & Bates, K.R. (1982). Young editors: Preschoolers' evaluation and production of ambiguous messages. 18, 30-42. Shultz, T.R. (1974). Development of the appreciation of riddles. _4_5, 100-105. Shultz, T.R. & Horibe, F. (1974). Development of the appreciation of verbal jokes. _1_0, 13-20. Tunmer, W.E. (1989). The role of language-related factors in reading disability. In Shankweiler & I. Liberman (eds.), _P_h_o_n_o_l_o_g_y _a_n_d _R_e_a_d_i_n_g _D_i_s_a_b_i_l_i_t_y: _S_o_l_v_i_n_g _t_h_e _R_e_a_d_i_n_g _P_u_z_z_l_e. Ann Arbor: University of Michigan Press. 15 Linguistic Ambiguity in Text Comprehension Tunmer, W.E. & Bowey, J.A. (1984). Metalinguistic awareness and reading acquisition. in W.E. Tunmer, C. Pratt & M.L. Herriman (eds.), _M_e_t_a_l_i_n_g_u_i_s_t_i_c _A_w_a_r_e_n_e_s_s _i_n _C_h_i_l_d_r_e_n. New York: Springer. Tunmer, W.E., Herriman, M.L. & Nesdale, A.R. (1988). Metalinguistic ability and beginning reading. _R_e_a_d_i_n_g _R_e_s_e_a_r_c_h _Q_u_a_r_t_e_r_l_y, _2_3, 134-158. Yalisove, D. (1978). The effect of riddle structure on children's comprehension of riddles. _D_e_v_e_l_o_p_m_e_n_t_a_l _P_s_y_c_h_o_l_o_g_y, _1_4, 173- 180. Yuill, N. & Joscelyne, T. (1988). Effects of organisational cues and strategies on good and poor comprehenders' story understanding. Journal of Educational Psychology, 80, 152- 158. Yuill, N. & Oakhill, J. (1988). Effects of inference awareness training on poor reading comprehension. _J_o_u_r_n_a_l _o_f _A_p_p_l_i_e_d _C_o_g_n_i_t_i_v_e _P_s_y_c_h_o_l_o_g_y, _2, 33-45. Yuill, N. & Oakhill, J. (1991). _C_h_i_l_d_r_e_n'_s _P_r_o_b_l_e_m_s _i_n _T_e_x_t _C_o_m_p_r_e_h_e_n_s_i_o_n: _A_n _E_x_p_e_r_i_m_e_n_t_a_l _I_n_v_e_s_t_i_g_a_t_i_o_n. Cambridge: Cambridge University Press. 16 Linguistic Ambiguity in Text Comprehension _T_a_b_l_e _1: _R_i_d_d_l_e_s _U_s_e_d _i_n _E_a_c_h _C_a_t_e_g_o_r_y _o_f _L_i_n_g_u_i_s_t_i_c _A_w_a_r_e_n_e_s_s HIGH AWARENESS: _M_e_t_a_l_i_n_g_u_i_s_t_i_c: Why is there so little honey in Brighton? Because there's only one B in it. Constantinople is a very long word. Can you spell it? I-T. What is at the end of everything? G What word is loud, even when you say it softly? Loud. _W_o_r_d _c_o_m_p_o_u_n_d_s: Have you heard the one about the man who bought a paper shop? It blew away. Sales clerk: Would you like to try our new oatmeal soap? Customer: No thanks, I never wash my oatmeal. Sales clerk: Would you like to buy a pocket calculator? Customer: No thanks, I know how many pockets I've got. How do you make a sausage roll? Push it down a hill. _P_r_a_g_m_a_t_i_c: Why do firemen wear red braces? To keep their trousers up. What would you do if you swallowed a pen? Use a pencil. LOW AWARENESS: _L_e_x_i_c_a_l: What's black and white and red all over? A newspaper. How do you know there was fruit on Noah's ark? Because the animals came in pairs. Why was the crab arrested? Because he kept pinching things. How can hunters in the woods best find their lost dogs? By putting their ears to a tree and listening to the bark. _A_b_s_u_r_d: How do you get an elephant into a matchbox? Take the matches out first. How do you fit 6 elephants into a mini? Three in the front and three in the back. 17 Linguistic Ambiguity in Text Comprehension _T_a_b_l_e _2 _M_e_a_n_s _a_n_d _s_t_a_n_d_a_r_d _d_e_v_i_a_t_i_o_n_s (_i_n _p_a_r_e_n_t_h_e_s_e_s) _o_f _c_h_a_r_a_c_t_e_r_i_s_t_i_c_s _o_f _c_h_i_l_d_r_e_n _i_n _e_a_c_h _s_k_i_l_l _a_n_d _t_r_a_i_n_i_n_g _g_r_o_u_p n Chronolog. Vocab. Neale Neale Age Score Accuracy Age Comprehension Age (regressed) Less-skilled comprehenders: Ambiguity 9 92.6 34.0 93.2 87.9 (4.2) (3.3) (5.3) (2.4) Control 9 92.7 33.0 93.7 86.3 (4.8) (3.5) (5.8) (2.9) -------------------------------------------------------- Mean 92.7 33.5 93.5 87.1 -------------------------------------------------------- Skilled comprehenders: Ambiguity 9 93.9 32.0 94.9 100.9 (4.9) (5.4) (5.5) (7.4) Control 9 89.4 32.4 94.3 98.7 (4.3) (4.7) (4.5) (5.9) -------------------------------------------------------- Mean 91.7 32.2 94.6 99.8 -------------------------------------------------------- Difference:good-poor 0.4 0.6 0.4 5.2,p<.005 (related t, 17 df) ------------------------------------------------------------------------ 18 Linguistic Ambiguity in Text Comprehension Table 3 _M_e_a_n _i_m_p_r_o_v_e_m_e_n_t_s (_m_o_n_t_h_s) _i_n _c_o_m_p_r_e_h_e_n_s_i_o_n _a_g_e _a_s _a _f_u_n_c_t_i_o_n _o_f _t_r_e_a_t_m_e_n_t _g_r_o_u_p _a_n_d _c_o_m_p_r_e_h_e_n_s_i_o_n _s_k_i_l_l (_s._d._s _i_n _p_a_r_e_n_t_h_e_s_e_s) Treatment Group ------------------------- Ambiguity Control Difference Less-skilled 20.56 13.11 7.45 (8.07) (7.92) Skilled 13.44 8.89 4.55 (6.73) (10.13) 19 Linguistic Ambiguity in Text Comprehension Table 4 _P_o_s_t-_t_r_e_a_t_m_e_n_t _c_o_m_p_r_e_h_e_n_s_i_o_n _s_t_a_t_u_s _a_s _a _f_u_n_c_t_i_o_n _o_f _t_r_e_a_t_m_e_n_t _a_n_d _s_k_i_l_l _g_r_o_u_p Comprehension status Combined data low mixed high low mixed/high Less-skilled Ambiguity 0 4 5 0 9 Control 4 3 2 4 5 Skilled Ambiguity 0 1 8 0 9 Control 1 1 7 1 8 _N_o_t_e_s. low = Neale comprehension age below accuracy and chronological age, mixed = Neale comprehension age above either accuracy or chronological age, high = Neale comprehension age above accuracy and chronological age. 20