Understanding indirect requests for information in high-functioning autism

The experimental literature on typical development (TD) seems to suggest that understanding forms of conventionalized IRs such as Can you…?, May you…?, Would you…?, and Why don’t you…? emerges early, from 2;6 years of age. For instance, Shatz (1978) tested 2-year-old children in a cooperative game in which the experimenter commented on some toy props and occasionally asked participants to perform some action with one of them. The experimenter’s requests could be presented in several forms that included IRs such as Why don’t you…? (e.g., Why don’t you put the chair in the house?). Children provided evidence that already at 2 years of age they seem to grasp the IR by behaving accordingly. Reeder (1980) tested 2; 6- and 3-years-olds’ IRs understanding in a paraphrases-choice task. Children were presented with a request such as Would you like to do x? and were asked to choose one out of 2 paraphrases that provided the interpretation of the request either as an indirect offer (e.g., I’ll let you do x), or as a request (e.g., I want you to do x). Data collected suggested that 3-year-olds performed better than 2; 6-year-olds in interpreting these forms of IRs.

Other studies exploring the developmental trajectory of IRs comprehension, testing different forms of IRs, provided a more precise view of the development of the phenomenon. Carrell (1981) asked 4–7-year-olds to color a circle either red or blue, following the experimenter’s requests. These could be formulated in 40 different ways, ranging from interrogatives such as Can you color the circle blue? to assertions such as The circle really needs to be painted blue and conditionals such as I’ll be very happy if you make the circle blue. Even though none of these might be properly defined as a hint, hence as a HIR, half of the IRs presented in this study displayed a positive surface form but conveyed a negative meaning—e.g., Should you color the circle blue? to suggest Don’t color the circle blue. Color it red. The results revealed that all age ranges were above chance in IRs understanding. However, a developmental pattern was observed: 4-year-olds provided 64.5% of correct interpretations, 5-year-olds 73.5%, 6-year-olds 78% and 7-year-olds 92%. Interestingly, with IRs that had a positive surface form and conveyed a negative meaning, 4- and 5-year-olds were at chance. The author links this pattern to children’s difficulty in understanding the mismatch between positive/negative surface forms and negative/positive conveyed meanings.

The findings on the developmental trajectory for IRs comprehension have been corroborated by other studies on highly indirect requests that used story completion tasks. Bernicot and Legros (1987) asked 3–4 and 5–6-years-olds to complete a story that ended with either a direct request or an IR. The interlocutor in the story would not comply with these. To complete the story, children could choose one out of 3 pictures, in which the speaker uttering the IR was shown very angry, unhappy, or okay. The type of the picture alternative selected by the children would provide an indication of the children’s ability to attribute to the speaker the intention of requesting. For instance, if the children chose the very angry characters, they were taken to attribute to the character a strong intention of request. The context was either strong (i.e., strongly biasing toward a directive interpretation) or weak (i.e., the object of the request was not explicitly mentioned or depicted). Five-to-six-year-olds performed better than the younger children and were sensitive to the context manipulation. However, they still exhibited greater difficulties in interpreting IRs as directives than direct requests. On the contrary, 3–4-year-olds seemed not to perceive any manipulation: they treated IRs similarly to direct requests and did not discriminate between contexts. According to the authors, younger children’s interpretation of IRs would not take into account the linguistic form of the request yet. Instead, children at this age would rely on the context, even though their ability to distinguish a strong vs. a weak context would not be mature enough yet.

Similar findings can be found in a study by Elrod (1987): children between 3; 3 and 6; 5 years of age were presented with illustrated stories that ended either with an IR (e.g., Those cookies are for our guests tonight) or a direct request (e.g., Please don’t eat the cookies). Children were then presented with three cards (and their verbal explanation) to complete the story with either a compliance to the IR (e.g., putting the cookie back); a non-compliance with a reference to the literal meaning of the IR (e.g., saying These are for our guest and eat a cookie); and a non-compliance without any reference to the literal meaning of the IR (e.g., eating a cookie). Elrod found that children younger than 5 years of age could not understand the IRs.

A similar developmental trajectory was found by Bucciarelli et al. (2003), who tested children between 2;6 and 7 years of age with a range of IRs (together with direct) that included conventionalized indirect (e.g., Sorry, could you close the window?) and highly indirect requests (e.g., Excuse me, I am studying, as a request to stop making noise). All age groups exhibited higher difficulties with HIRs than both direct and conventionalized IRs. Yet, 6–7-year-olds correctly interpreted HIRs 68% of the times, against the 43% of 4;6–5;6-year-olds, the 42% of 3;6–4-year-olds and the 38% of 2;6–3-year-olds. The authors associate children’s difficulty at interpreting HIRs with the complexity of the process of comprehension for such cases; the process, according to the authors, requires the ability to construct mental representations and build chains of inference, which increases with age. In addition to this, the development of IRs comprehension seems to unfold during school age, at least up to 8 years of age (Bernicot et al. 2007).

Overall, the literature on IRs comprehension in TD children seems to show relatively high construct validity, as most of the studies above operationalized their research questions with a cooperative task, which warrants children’s engagement with the task. However, some aspects are noteworthy that might constrain the external validity of these studies. For instance, some studies used purely metalinguistic tasks (e.g., a paraphrase-choice task in Reeder 1980) or presented the task by openly stating it would involve requests (e.g., Carrel 1981). Additionally, some of these studies, especially the earlier works, were conducted on a small sample size and this poses issues on the generalization of results.—see Table 5 in the appendix for experimental details of the studies. Nevertheless, this body of research suggests two main findings. First, children show some understanding of IRs already from around age 3. However, second, the development of the ability to interpret IRs, particularly highly indirect ones, seems to go well beyond pre-school years. In fact, it is only at around 7–8 years of age that children start mastering HIRs.

Similarly to other pragmatic impairments, research on IRs understanding in HFA provided mixed findings for both adults and children.

As for adults, some studies suggested that the ability to understand IRs is impaired in ASD. Paul and Cohen (1985) replicated Carrell (1981) and asked adult ASD participants (mental age: 4–7 years) and controls with an age-matched cognitive disability to color a circle either in blue or in red (e.g., I’ll be happy if you color this circle blue). The different forms of IRs (e.g., conventionalized and non-conventionalized) were made in two experimental sessions that could or could not explicitly inform participants that they would be presented with requests. In the structured session, participants were clearly instructed that the experimenter would tell them to color some circles. In the unstructured session, IRs were made throughout a conversation between the experimenter and the participants while they were drawing. ASD participants performed worse than controls in both sessions, but—contrary to the control group—they performed worse in the unstructured than in the structured session. Moreover, similarly to TD children in Carrell’s (1981) study, ASD participants exhibited particular difficulties with negative items such as You shouldn’t color the house blue: they tended to respond by choosing the color named rather than its opposite. Evidence for HFA individuals’ impairment with IRs was also found in Ozonoff and Miller (1996). In this study, HFA adults (vs. controls matched for age and IQ) saw short vignettes that ended with a question (e.g., Can you see that house number?). Items appeared in two contexts that could either elicit a literal interpretation of the question (i.e., a question on the interlocutor’s ability) or an interpretation of this as a conventionalized IR (e.g., tell me that house number). Participants could choose one response (out of 4). The indirect context was found easier than the literal context for both HFA individuals and the control group. However, HFA participants provided overall less correct responses than the control group in both contexts. Interestingly, in incorrect trials, independently of the context, HFA participants interpreted as indirect requests the questions that in fact were not an indirect request—which the authors explain as an impairment in using context and meaning appropriately.

Counterevidence is also available. Deliens et al. (2018) tested ASD adults and neurotypical controls in two act-out tasks on IRs and irony. In the IRs task, participants were presented with some colored shapes that appeared in a grid. Upon hearing some requests, they could either move the shapes around the grid or press a yes or no button. The requests were imperatives or conventionalized/non-conventionalized IRs in two interrogative forms (e.g., Can you… or Is it possible to… move the blue circle to the left of the red triangle?). ASD participants interpreted the conventionalized IRs as much as the control group. Instead, they interpreted non-conventionalized IRs directively more often than the control group. The authors concluded that IRs understanding seems to be preserved in ASD.

As for HFA children, the available results provide a more homogeneous picture. MacKay and Shaw (2004) tested HFA children and TD controls aged between 8 and 11 years in a task that assessed the understanding of both the “meaning” and the “intention” of IRs. Children heard short stories that could end with non-conventionalized IRs such as That cake looks delicious. Afterward, they were asked two questions, one about the surface meaning of the utterance (e.g., What does x mean?) and one about its intent (e.g., Why A utters x rather than asking for x directly?). Both HFA and TD children correctly understood the IRs. However, contrary to TD children, the HFA participants exhibited more difficulties at explaining the intent behind the IRs, for instance they used more I don’t know. The authors suggest that HFA children grasp an IR, but they cannot explain this either because of their linguistic competence or because of a pragmatic impairment. Within a naturalistic scenario that avoided metalinguistic tasks, Kissine et al. (2012) videotaped the behavior of ASD children aged 4;3–12;5 years while they were involved in interactions with adults, who uttered 4 types of requests: imperative (e.g., Pour the milk!), declarative (e.g., You are going to put the bottle in your bag; You forgot the water in your bag), interrogative (e.g., Can you throw this in the bin?) and sub-sentential (e.g., Your place, with the meaning of Get back to your place). It was found that ASD children complied well with all types of adults’ requests and this was taken as evidence in favor of the fact that IRs understanding seems preserved in HFA children. Yet, the authors acknowledge the possibility that HFA children’s interpretive strategies might be fairly simple, based on their participants’ very low IQ. Overall, the earlier literature on IRs comprehension in ASD seems to show some of the same weaknesses in validity as the literature on TD children. In fact, some studies used metalinguistic tasks, which might decrease construct validity (e.g., the question about intentions by Mackay and Shaw, 2004) or involved a small sample size (Paul and Cohen, 1985; Kissine et al. 2012), thus reducing the statistical power of the study—see Table 6 in the Appendix for the experimental details of these studies.

Using a more fine-grained experimental method, Kissine et al. (2015) provided very interesting results. A group of ASD children and control TDs were tested in a semi-structured task involving a toy called Mr. Potato Head. Participants were asked to “dress” Mr. Potato Head with different items, such as a hat, glasses, etc. In the first phase, the children were invited to put the hat on the toy with a highly indirect request (e.g., Oh, he has no hat!); in the following two phases, the same sentence was uttered as a comment on a picture by a collaborator and then by the experimenter, to make sure the utterance would not stimulate a bias to action regardless of the context. ASD children complied with the IR in the first phase and inhibited the directive interpretation in the other two, while TD children had difficulties in comprehension in the first phase. The authors conclude that IRs comprehension is preserved in ASD. However, we think that of such a conclusion might be argued. First, the two groups of participants were not age-matched (HFA: 7–12-year-olds; TD: 2;7–3;6-year-olds). As such, it is unclear whether HFA children’s better performance reflected age group-related differences rather than genuine IRs comprehension in ASD. Moreover, the authors explain the dissociation between the two age groups pointing to the possibility that the younger TD group might not have sufficient mind-reading abilities to comply with the semi-structured task. Crucially, no ToM measure was collected in this study. In our view, the authors’ remark is partly controversial. In fact, there are at least two alternatives: first, if it is true, as they apparently assume, that IRs comprehension relies on ToM, then we should expect this ability to be compromised in HFAs, as it is for their skills in other pragmatic phenomena that notably rely on ToM. However, this conflicts with Kissine et al.’s conclusion that IRs are preserved in HFA. Second, if IRs comprehension does not rely on ToM, then the observed group difference reflects a mere developmental effect. To put it simple, TD children performed worse than HFA children because they are younger; thus, not only their ToM skills but also their general cognitive functioning is not developed enough yet. Hence, there might be reasons to think that the authors’ conclusion that IRs are preserved in HFA might remain a speculation. To rule this issue out, developmental biases need to be avoided: TD and HFA children at the same developmental stage have to be tested. Moreover, a measure of their ToM ability should be collected in order to verify whether (un)preserved mind-reading skills can explain IRs understanding. In this respect, it is worth mentioning that only a few studies on IRs comprehension have taken ToM into account, which, however, did not involve ASD or TD children. In fact, these studies were conducted on healthy adults (Trott and Bergen 2018; van Ackeren et al. 2012) and on a range of clinical populations such as Alzheimer disease (Cuerva et al. 2001), traumatic brain injuries (Muller et al. 2010) and on right-hemisphere lesions (Champagne-Lavau and Joanette 2009). Crucially, they overall suggest that ToM might play a role in understanding IRs.

All experimental studies on IRs understanding referred to in this section are also schematically reported in the Appendix Tables 5, 6, 7 with details on population characteristics, experimental tasks and results.

Sixty-one Italian monolingual children between 5; 2 and 12 years of age participated in the experiment. Children were divided into three groups. The first group consisted of 14 children who received a diagnosis of high-functioning autism (age range: 9–12 years; mean age(SD) = 10.6 (1.17); 2 F). They had all received their diagnosis of HFA (according to the DSM-IV criteria) by a team of trained neuropsychologists. The second group was composed of 28 typically developing children (TD), who were age-matched with the HFA children (age range: 9–12 years; mean age (SD) = 11.03(0.61); 11 F). The third group consisted of 19 younger TD children (age range: 5; 2–6; 3 years; mean age (SD) = 5.35(0.48); 6 F). All children were also tested for structural language through a standardized test for Italian, the Batteria per la Valutazione del Linguaggio (BVL_4-12) test for children aged 4–12 years (Marini et al. 2015). Participants in the HFA group were recruited in an ASD support center located in Genoa. Participants in the TD group were recruited in a primary school in Vicenza. The younger TD children were recruited in a kindergarten in Genoa. All participants were tested in a quiet area of their support center/school. None of them had neurological, linguistic, or hearing disorders.

Written informed consent was obtained from the parents/guardians of the participating children; since the Ethics Committee of our institution had not been constituted yet, at the time of the experiment, we could not ask for their approval. Nonetheless, the research has been conducted in full compliance with the ethical standards of APA ethical guidelines as well as of the 1964 Declaration of Helsinki.

Children were administered four tasks: an IRs comprehension task; one task that assessed for their linguistic abilities; and two ToM tasks. The order of administration of the tasks was randomized across children to control for biases related to the order of the tasks. For each child, the overall experimental session was about 30 min long.

All experimental sessions were audio-recorded, and children’s responses in the three tasks were subsequently coded by the experimenter.

Three types of statistical analyses were conducted. First, children’s accuracy in the IRs tasks was analyzed. Second, group-related differences in the BVL and the two ToM tasks were analyzed. Third, an analysis of predictors was conducted to assess whether children’s linguistic and/or mind-reading abilities play a role in their understanding of IRs.

Children’s accuracy in the IRs task was analyzed with Linear-Mixed Models statistics (LMMs), using the lme4 (Bates et al. 2015) and the lmerTest packages in the R environment, to provide F statistics with degrees of freedom. Tukey contrasts with the R package emmeans were used for post hoc comparisons. For each child, a composite score in each of the three experimental conditions of the IRs task was computed by counting the number of correct trials.⁴ In each condition, each child could be assigned a composite score between 0 and 4, i.e., each child had three composite scores between 0 and 4 (one for the direct, one for the indirect, and one for the highly indirect requests). The fixed-effects structure of the LMM model included Group (HFA, TD, and younger TD) and Condition (DIR, IR, HIR), along with the resulting interactions. The random structure of the model included random intercepts for subjects.

Group differences in structural language and mind-reading skills were analyzed with the Kruskal–Wallis rank sum test statistics, and Dwass–Steel–Critchlow–Fligner contrasts for pairwise comparisons.

As for the analysis of predictors, a LMM statistics was carried out in which participants’ composite score in the IRs task was the outcome variable and participants’ scores in the structural language task (the BVL) and in the ToM tasks were treated as predictors, together with Group, Condition and the resulting interactions. For each child, a composite ToM score was computed by counting the score in each of the two false-belief tasks.⁵ This returned 3 possible outcome scores: 0, if the child failed both tasks; 1, if the child passed only one of the two tasks; and 2, if the child passed both tasks.

All statistical analyses were conducted using R software (R Core Team, 2020). Data and code used for the analyses are available in the Supplementary Material hosted on the Open Science Framework web platform (link to the project: https://​osf.​io/​p2dn5).

Overall, all children performed well in the IRs task (see Fig. 1). Yet, their performance does not seem the same independently of group and condition. The LMMs statistics (see Table 2 for all statistical details) revealed a significant main effect of condition, thus indicating that children provided significantly more correct responses to indirect than highly indirect requests (mean composite score (SD) in each condition: DIR: 3.55 (0.84); IR: 3.57 (1.05); HIR: 3.08 (1.34)). Additionally, and most important, the interaction Group by Condition was significant, too, and simple effects analysis (Table 2) revealed interesting group differences. First, older TD children exhibited significantly higher accuracy in direct than both indirect and highly indirect requests. Second, an opposite pattern emerged for HFA children, who provided significantly more correct responses to indirect than both highly indirect and direct requests, while no difference emerged between DIR and HIR in this group. Finally, no significant differences of condition emerged in the group of younger TD children.

Taken together, these data suggest different patterns of IRs comprehension in the three groups of participants.

Significant differences emerged in children’s score in the BVL test depending on group: children in the older TD group scored significantly higher than both HFA children and children in the younger TD group. No differences emerged between HFA and younger TD children—see Table 3 for all statistical details.

As for ToM, significant group differences emerged for the first-order ToM task: as with the BVL test, children in the older TD group scored significantly higher than both the HFA group and the younger TD group, while the HFA group and the younger TD group did not significantly differ (see Table 3 for the statistical details). Instead, the pattern of results

slightly changed for the second-order ToM task (Table 3). Significant group differences emerged, with older TD children scoring significantly higher than younger TD children. Yet, this time, the difference between older TD children and HFA children only approached significance, and so did the difference between HFA and younger TD children, too.

Most important, when children’s scores in the two ToM tasks were combined into a composite score to obtain an overall measure for ToM, the significant main effect of group was confirmed and TD children exhibited overall better ToM skills than both HFA and the younger TD children. On the contrary, this time, no significant difference was found between HFA children’s scores and those of the younger TD children—Table 3.

All results from the LMMs statistics for the analysis of predictors are reported in Table 4. This analysis confirmed that children’s scores in the IRs comprehension task are significantly predicted by the type of request: when this is an IR, but not a HIR (vs. DIR), the effect is significant (i.e., significant effect of Condition IR only). Most interesting, children’s linguistic abilities as assessed by the BVL test did not reveal a significant predictor of IRs comprehension (i.e., no significant BVL-related effects). Even more interesting, mind-reading skills differently predicted the scores in the IRs comprehension task depending on participants’ group and the type of request. In fact, even though ToM marginally predicted children’s accuracy in the IRs task regardless of group, this revealed a significant predictor of children’s performance in the IRs task only in the group of older TDs and only with HIRs (i.e., significant 3-ways interaction Condition HIRX Group TD X ToM Composite score)—see Table 4 and Fig. 2.

Overall, then, children’s performance in the ToM tasks seems to play a role in IRs comprehension only for TD children (as compared to HFA children and the younger TD children) and when highly indirect requests are involved.