Pragmatic Tolerance and Theory of Mind in Children’s Comprehension of Under-Informativeness

(1)

Pragmatic Tolerance and Theory of Mind

in Children’s Comprehension

of Under-Informativeness

Alma Veenstra, S1328891 A.M.Veenstra.2@student.rug.nl University of Groningen Research Master Linguistics Supervisors: Dr. Bart Hollebrandse Dr. Napoleon Katsos

(2)

0. Abstract

It has been argued that certain experimental set-ups will obscure children’s true competence with pragmatics. We proposed a combination of a binary and a graded judgment task with the addition of a reaction time element to show the importance of a good methodology. This methodology managed to make a distinction between logical, Pragmatic Tolerant, and pragmatic behaviour towards under-informative utterances. We tested two hypotheses:

Hypothesis 1: Longer reaction times for under-informative sentences than semantically correct ones will reveal that both children and adults are sensitive towards under-informativeness, even though they accept it in binary settings, thus supporting the Pragmatic Tolerance account.

Hypothesis 2: Theory of Mind is necessary to be able to detect under-informativeness and to generate implicatures. Scores on a Theory of Mind task will positively correlate with rejections of under-informativeness.

The data support hypothesis 1, longer reaction times were found for under-informative sentences than for semantically correct sentences. We also found that Theory of Mind facilitates the cognitive processes associated with under-informativeness, supporting hypothesis 2. Yet, having an adult Theory of Mind does not necessarily promote the actual rejection of under-informativeness. (183 words)

(4)

1. Introduction

Picture a saleswoman carrying a basket with ten apples in it and none outside of the basket, and she tells you that some of the apples are in the basket. Your first reaction may be ‘no that is incorrect; all of the apples are in the basket.’ You have derived the scalar implicature of ‘some’ meaning ‘some, but not all.’ If the woman really had only some of the apples in her basket, there would have to be apples outside of the basket as well. The fact that the situation does not completely match the statement of the woman causes the hearer to penalise the utterance; she is not right. This reaction is based on the Gricean maxims of quantity: Your statements should not contain more information than needed, and your statements should not contain less information than needed (Grice, 1989). The woman in this example is clearly giving less information than needed; she is being under-informative.

(5)

be used. Children are reported to initially interpret sentences in the literal, semantic manner, whereas the narrowed interpretation, the scalar implicature is acquired later.

Katsos and Bishop (under review) are more nuanced in their conclusions. They found that the age at which children seem to acquire scalar implicatures can be largely influenced by the experimental setup. Providing a visual context, removing distracters from this context, using expressions that are familiar to children, giving training and explicit instructions, and using action-based experiments are found to improve the children’s performance. This effect is possibly due to the fact that performance is then relying less on working memory capacity (Pouscoulous, Noveck, Politzer, & Bastide, 2007). Katsos and Bishop argue that children approach under-informativeness along the principle of Pragmatic Tolerance: They are aware that an utterance is under-informative, but do not reject this pragmatic infelicity. In a binary judgment task year-old children accepted under-informative sentences, whereas other 5-year-olds were given a 3-point scale to grade the sentences and none of them graded these under-informative sentences maximally. It can be argued that children in earlier studies, who were categorized as oblivious to under-informativeness, were in fact just behaving Pragmatically Tolerant. Moreover, even adults are reported to be Pragmatically Tolerant, accepting under-informative sentences in a binary judgment paradigm but showing sensitivity in different tasks.

(6)

expect children with a Theory of Mind to be stricter towards speakers that do have the opportunity to consider their options.

Theory of Mind is also shown to be related to language development in other areas such as embedding and recursion (Hollebrandse & Roeper, 2007; Hollebrandse & Van Hout, 2008; Hollebrandse, Hobbs, De Villiers & Roeper, 2008) and labelling (Jacques & Zelazo, 2005). Performance on these linguistic aspects was found to be a predictor of performance in ToM-tasks. A relation between Theory of Mind and taking into account the speaker’s and hearer’s perspective is alluded to in studies about referential expressions (Wubs, 2008). Having an adult Theory of Mind is assumed to be a prerequisite for being able to optimize bi-directionally.

Since Katsos and Bishop did their experiments on different groups of children, the ultimate test would be to see how Pragmatic Tolerance is present in different experimental paradigms within one group of participants. By taking a range of participant’s ages from 4 to 9, it may become clear how the acquisition of informativeness develops. We might be able to answer the question what the role is of methodology on the age at which children seem to be able to interpret under-informative sentences with a narrowed meaning.

(7)

2. Implicatures and Informativeness

2.1 Generalized vs. Particularized Implicatures

Implicatures arise when there is additional information in an utterance by what is not said. Noveck (2001) describes the implicature as “an inference that consists of attributing to a speaker an implicit meaning that goes beyond the explicit linguistic meaning of an utterance” (p. 165). In the example (1) taken from Bott and Noveck (2004) John will interpret Robyn’s answer meaning that she has not in fact met all of his friends:

(1) John: Did you meet all of my friends?

Robyn: Some of them.

(8)

conversational implicatures. They are generated systematically, even without context, because the meaning of the implicature is lexicalized in our language system. Levinson (2000) claims generalized implicatures are always generated when an utterance contains a weak term from a scale, but can be cancelled afterwards in special circumstances. These special, grammatical, circumstances are negations, question forms and antecedents of conditionals (Chierchia, 2004).

Another type of implicature is the particularized implicature. Consider the example (2) from Breheny, Katsos and Williams (2006):

(2) John: Was the exam easy?

Mary: Some of the students failed.

a) Not all the students failed

b) The exam was not easy

The implicature of (2a) is a generalized implicature; it will be generated even without the context of John’s question. However, (2b) will only be generated in the context of John’s question. Because if the question was different, for example ‘is the teacher doing a good job?’ (2a) still makes sense, whereas (2b) does not. A particularized implicature also arises in a situation in which there is the possibility of two objects, an apple and a pear, being in a basket. Mentioning that the apple is in the basket implies that the pear is not, since on a scale <apple (or pear), apple and pear> the weaker term implies the stronger term is not true. Clearly this scale only holds within this specific context. The implicatures that will be studied in this study are of the particularized kind.

(9)

Chierchia, 2004) claim that generalised implicatures are generated by default and take effort to be cancelled, whereas particularized implicatures take effort generate because they are not generated by default. On the other hand, there are the proponents of the Relevance Theory, a unitary theory that treats all implicatures the same (Sperber & Wilson, 1986/1995). They claim that an implicature will only be generated if required by the context. Findings of Bott and Noveck (2004) support this theory: “Such an inference occurs in particular situations as an addressee makes an effort to render an utterance more informative” (p. 456), without distinguishing implicature type.

2.2 Processing Costs

According to Relevance theory generating an implicature is a costly process; it will be only executed if required by the context (Sperber & Wilson, 1985/1996). Consider the following examples (3) and (4) from Carston (1998):

(3) A: If you or some of your neighbours have pets, you shouldn't use this pesticide in

your garden.

B: Thanks. We don't have pets, but some of our neighbours certainly do.

(4) A: Do all, or at least some, of your neighbours have pets?

B: Some of them do.

(10)

Feeney and his colleagues found longer reaction times for adult participants on under-informative sentences than semantically correct ones (Feeney, Scrafton, Duckworth, & Handley, 2004). They claim that this longer reaction time is caused by both the time it takes to generate the implicature and then to inhibit to respond pragmatically, since their participants often accepted informativeness. A positive correlation between acceptances of under-informativeness and successfulness in a counting span task, leads them to conclude that sufficient cognitive capacity is needed to cancel the implicature.

Noveck and Reboul (2009), however, claim that rejecting an under-informative utterance takes longer than accepting it, whereas both take longer than responding to semantically correct or incorrect sentences. Evidence for this can be found in the study by Noveck and Posada (2003), who looked at reaction times and ERP data on under-informativeness in adults. The ERP data on the other hand, did not signal clear differences between under-informative utterances and semantically correct or incorrect ones. In addition, Bott and Noveck (2004) report a trend in which responding to a semantically true sentence takes less time than responding to an under-informative sentence.

Pouscoulous et al. (2007) point out that computing implicatures is a costly enterprise, something that is less likely to be invested in by speakers that do not have sufficient resources available. The fact that pragmatic interpretations of under-informativeness are not very common in young children they attribute to the processing capacity that is smaller in children than in adults.

2.3 Acquisition of Implicatures

(11)

weaker term to describe a situation in which a stronger term is more informative, consider a situation in which there all of the balls are blue: It is logically true that there are some balls that are blue, yet this statement would be under-informative. The frequency of pragmatic interpretations increases with age. Researchers have used a wide variety of experimental paradigms and found equally varied results: Noveck (2001) found that children at the age of 7 were not able to reject a weaker term on a scale in favour of the more informative term in a binary judgment task. Papafragou and Musolino (2003) tested, again in a binary judgment paradigm, 5-year-old children’s performance on quantified, numerical, and verbal scales, and found they often accepted under-informative statements. Training and explicit instruction improved the children’s performance, increasing the successfulness on the numerical scales to a near-ceiling 90%. Feeney and his colleagues found that manipulation of the relevance of an implicature also enhances objection towards under-informativeness, finding higher rejection rates in 7-year-olds than Noveck (2001) did (Feeney et al., 2004). Guasti and her colleagues studied children’s pragmatic competence in a binary judgment setting further by adapting Noveck’s (2001) experiments with Papafragou and Musolino’s training and explicit instructions (Guasti, Chierchia, Crain, Foppolo, Gualmini, & Meroni, 2005). They found that the performance of 7-year-olds did increase, but this effect did not persist over a longer period of time. Katsos and Bishop (under review) studied both generalised and particularized implicatures in 5-year-old children and found that in a binary judgment setting under-informative utterances of both types were often accepted.

(12)

a graded judgment paradigm, which showed that 5-year-old children are sensitive to under-informativeness. The children did not unanimously opt for the highest gradation as you expect children that are unable to detect under-informativeness would do.

However, as Katsos and Bishop (under review) critically observe, how can you tell the ability to generate implicatures apart from detecting under-informativeness? If children reject the under-informative ‘in the basket there is an apple’ when there are in fact an apple and a pear in the basket, is that because they generate the implicature that there is an apple and not a pear in the basket, which is untrue, or because they simply notice the pear is left unmentioned? Veenstra (2010a) found that many children rejected under-informative sentences, saying ‘no, he forgot to mention the X!’ indicating they might not have needed to generate the implicature. Detecting under-informativeness is required to be able to generate implicatures; however, you do not need to be able to generate implicatures in order to detect informativeness. In order not to mix up the two reasons for rejecting under-informativeness, in this study we will focus on the acquisition of under-informativeness, rather than the acquisition of scalar implicatures.

2.4 Pragmatic Tolerance

(13)

7-year-old as earlier studies (e.g. Noveck, 2001; Papafragou & Musolino, 2003; Feeney et al., 2004; Guasti et al., 2005) claimed. Children unable to reject under-informativeness fully understand informativeness, but are unable to act according to the pragmatic interpretation of under-informative utterances.

Adults however, have also shown to accept under-informative utterances. In addition to an at ceiling performance in the acceptance of semantically correct sentences, there was also an acceptance rate of over 40% for under-informative sentences in the studies by Noveck (2001), Feeney et al. (2004), Guasti et al. (2005), and Pouscoulous et al. (2007). In Katsos and Bishop (under review) this percentage was 28%, and these adults all made remarks along the lines of ‘yes, but…’ indicating it was not a straightforward acceptance. For Pragmatic Tolerant children we said they are unable to behave pragmatic, but it feels counterintuitive to say the same of adults who display Pragmatic Tolerance. Feeney et al. suggest the adults deliberately choose to behave logically instead of pragmatic, since adult reaction times were higher for under-informative sentences than semantically correct ones. Perhaps we should then distinguish between involuntary Pragmatic Tolerance (expected in children) and voluntary Pragmatic Tolerance (in adults).

(14)

3. Theory of Mind

3.1 First- and Second-Order Theory of Mind

Theory of Mind is used nowadays to refer to “children’s understanding of people as mental beings who have beliefs, desires, emotions, and intention and whose actions and interactions can be explained by taking account of these mental states” (Astington & Baird, 2005, p. 3). Although researchers do not share a mutual agreement on what exactly Theory of Mind comprises, many agree that false belief tasks measure Theory of Mind (Dennett, 1978; Wimmer & Perner, 1983; Perner & Wimmer, 1985). False belief refers to situations in which someone else can have a belief that is different from yours, or reality. Wimmer and Perner’s (1983) false-belief task makes sure that a participant and someone else have different information about a situation. The participant is then asked to predict what the other will do in that situation. If the participant predicts that the other will act on the basis of the other’s knowledge of the situation, not based on his or her own knowledge, s/he is believed to have a Theory of Mind.

(15)

3.2 Language before Theory of Mind

Although Theory of Mind is a cognitive skill, many studies have shown that a certain level of linguistic knowledge is needed in order to gain a Theory of Mind. Jill De Villiers (2005) promotes the Linguistic Determinism Theory, which basically says that “language is causally involved in the development of false-belief understanding, not just in task performance” (p. 187). Acquisition of the syntax and semantics of mental-state verbs and complement-taking verbs provides children with the knowledge needed for a Theory of Mind. In other words, when children know the appropriate usage (conceptual, lexical, syntactic and semantic) for verbs as say or think, they will also be able to understand the false belief reasoning behind them.

More specifically for second-order Theory of Mind, Hollebrandse and his colleagues argue that acquisition of second-order Theory of Mind is dependent on the acquisition of second-order language (Hollebrandse et al., 2008; Hollebrandse & Van Hout, 2008). Second-order language refers to the multiple (recursive) embedding of syntactic complements. Consider example (5) (from Hollebrandse & Van Hout), in which the that-clause in (5a) is a first-order embedding and in (5b) a second-order:

(5) a Cookie Monster thinks that it is raining.

b Big Bird thinks that Cookie Monster thinks that it is raining.

(16)

relation between second-order language production and second-order Theory of Mind reasoning.

Even in deaf children knowledge of syntax embedding and especially lexical knowledge of attitude verbs are predictors of Theory of Mind-behaviour, indicated by strong correlations as Peter De Villiers (2005) found. Since deaf children display a general delay in language acquisition, the finding that also false-belief reasoning was delayed in comparison to hearing children is in keeping with the claim that language matters for a Theory of Mind.

3.3 Theory of Mind before Pragmatic Competence

Having a Theory of Mind includes a long list of abilities: False-belief understanding, meta-representation, intentional stance, belief-desire reasoning, mental-state attribution, mentalizing, mind-reading, perspective-taking, social intelligence, social understanding, social intuition, social perception, person perception, and intersubjectivity (Astington & Baird, 2005). It can be argued that for generating implicatures you need to know about perspective-taking and other’s intentions: What are the speaker’s options, and which one did s/he choose to express the intended meaning.

(17)

Wubs & Hoeks, 2008; Wubs, Hendriks, Hoeks & Koster, 2009; Veenstra, 2010b). Wubs (2008) even alludes to a possible relation between bi-directional optimization and Theory of Mind.

In the Optimality Theory approach to child language acquisition, it is agreed that children first learn the constraint ranking that is appropriate for their native language, and later learn how to optimize bi-directionally (Blutner, De Hoop & Hendriks, 2006). This uni-directional optimization they initially use only prescribes the optimal form and meaning needed for one perspective, either as a speaker or as a hearer. In the example of subject anaphora, for instance, this leads to children producing only pronouns in a conversation. When later they learn to evaluate the other options available, children will also start using noun phrases. A similar process applies to scalar implicatures and under-informativeness. Children first use the logical interpretation of ‘some’, even in cases where a stronger term ‘all’ would be appropriate. When they later realize that the speaker would indeed have used the stronger term ‘all’ if appropriate, they pragmatically interpret this weaker term meaning ‘some, but not all.’ Since the allocation of intentions to a speaker is associated with a Theory of Mind, it is conceivable that this ability has to be acquired in order to be able to generate implicatures.

De Villiers, De Villiers, Coles-White and Carpenter (2009) tested (typically developing and autistic) children’s performance on relevance implicatures and Theory of Mind, arguing that having a Theory of Mind would facilitate interpreting conversational implicatures. These implicatures differ from scalar implicatures, consider example (6):

(6) Speaker 1: “Do you want some coffee?”

(18)

Reading other people’s intentions, one of the abilities associated with a Theory of Mind is crucial in these kinds of implicatures. They found indeed a very strong positive correlation between ToM-scores and successfulness with the implicatures in the high-functioning autistic children, which suggests this might also be found in younger typically developing children. However, they were not able to separate the effect of age from the effect of Theory of Mind.

In addition to implicatures, we argue that Theory of Mind is also needed to detect informativeness. Conform Grice’s maxims of quantity, one should not be either under-informative or over-under-informative. When describing a basket with an apple and a pear in it, the description ‘in the basket there is an apple’ can only be considered under-informative when you have the cognitive ability to compare this utterance with the more informative ‘apple and pear.’ This does not necessarily require insight in the speaker’s intentions, as with implicatures, but the hearer does have to take into account that the speaker has sufficient knowledge about the basket to have been able to use the more informative expression, see (7)

(7) Situation: A basket with an apple and a pear

Speaker’s options: a) In the basket there is an apple

b) In the basket there is a pear

c) In the basket there is an apple and a pear Speaker: “In the basket there is an apple”

Child without a Theory of Mind: That is true

Child with a Theory of Mind: He could have used the more informative utterance c.

(19)

again indicates that Theory of Mind, which the children in their study are assumed to possess, is involved in the interpretation of under-informativeness.

(20)

4. Hypotheses

From the literature on implicatures, we may conclude that different experimental methods yield different results. Children who seem to be unable to compute scalar implicatures in a binary judgment task may show sensitivity to under-informativeness in a graded judgment task. Giving children both tasks will distinguish between those who are not sensitive to under-informativeness and those who are aware that something is under-informative, but display Pragmatic Tolerance. We expect a linear development in the acquisition of implicatures: In the first stage, children are unable to compute implicatures or detect under-informativeness; in the second stage they detect under-informativeness; and in the third stage they are able to compute implicatures, see Figure 1:

Figure 1. The predicted development of implicatures in children.

However, since it is very difficult to distinguish between the ability to generate implicatures and the ability to detect under-informativeness, we will only focus on the development from insensitivity to under-informativeness to sensitivity of under-informativeness. Sensitivity to under-informativeness can be present in different forms, which the combination of a binary and graded judgment task with registered reaction times will show.

We predict that children who are unaware of informativeness, will accept under-informative sentences in a binary judgment task, as well as reward them maximally in a graded judgment task. Their reaction times will not differ between semantically correct sentences and under-informative sentences. Children who display Pragmatic Tolerance will

No implicatures or informativeness

(21)

accept under-informative sentences and rate them not maximally in the graded judgment task. Reaction times will show that accepting under-informative sentences take longer than accepting semantically correct ones. Children who do act according to pragmatic interpretations will reject under-informativeness in the binary setting, and also rate them lowest in a graded setting. They will have longer reaction times for rejecting under-informativeness than for straightforward semantic rejections.

Pragmatic Tolerance and even purely logical reasoning have also been found in adults. Nevertheless, adults are believed to have acquired implicatures. Whereas the combination of a binary and graded judgment task would not differentiate purely logical adults from children who do not detect under-informativeness, reaction times should show that the processing of an under-informative sentence takes longer than the processing of a straightforward correct or incorrect sentence (Feeney et al., 2004; Noveck & Posada, 2003). We therefore argue that adults who seem to reason purely logical are also Pragmatically Tolerant.

Thus, there are three approaches we can distinguish with regard to dealing with under-informativeness: there are children oblivious to under-informativeness, children and adults displaying Pragmatic Tolerance, and the adult-like rejection of under-informativeness, see Figure 2:

Competence:

Behaviour:

Figure 2. Approaches to under-informativeness

(22)

Theory of Mind reasoning requires a certain processing capacity, similar to perspective-taking (or, in OT terms, bi-directional optimization). It is therefore conceivable that the development of a Theory of Mind coincides with that of the perspective-taking needed for computing implicatures. A positive correlation is expected. Whereas other studies found certain linguistic abilities to be required for the successfulness with Theory of Mind (J. de Villiers, 2005), we argue that having a Theory of Mind is the factor enabling children to distinguish and, more importantly, take into account the speaker’s and hearer’s perspectives. Since computing implicatures is based on doing the math between the options a speaker has to describe something and the option he or she did actually use, the hearer has to take into account the perspective of the speaker. This cognitive ability is shown to be related to working memory in the use and interpretation of subject anaphora (Hendriks et al., 2008; Wubs et al., 2009; Veenstra, 2010b), but also hypothesized to be related to Theory of Mind. In sum, the following hypotheses will be tested in this study:

Hypothesis 1: Longer reaction times for under-informative sentences than semantically correct ones will reveal that both children and adults are sensitive towards under-informativeness, even though they accept it in binary settings, thus supporting the Pragmatic Tolerance account.

Hypothesis 2: Theory of Mind is necessary to be able to detect under-informativeness and generate implicatures. Scores on a Theory of Mind task will positively correlate with rejections of under-informativeness.

(23)

5. Methodology

In this chapter, the actual study will be described. The particulars of the subjects who participated are described in section 5.1, the materials used for the four experiments are described in section 5.2, and the test procedure in 5.3. In section 5.4, the design of the study and how the results will be analysed is presented.

5.1 Participants

Two groups of native Dutch participants took part in this study: 75 children, and 23 adult control participants. The 75 typically developing children tested were aged 4;0 to 9;8 (µ = 75.3 months, SD = 19.7 months), to be sure there are children without a Theory of Mind, children with only a first-order, and children with a second-order Theory of Mind. There were 32 boys and 43 girls. The children were recruited from primary school De Sprankel in Stiens and among children of friends and acquaintances.

The adult control group consisted of 23 participants, ages 22 to 68 (µ = 40.1, SD = 15.9), 9 men and 14 women. Their educational background was varied, as we made sure not only to test university students. None of them reported language difficulties.

5.2 Materials

In this section the four tasks of the study are described: The Binary Judgment Task, the Graded Judgment Task, the Theory of Mind Task, and the Working Memory Task.

The Binary Judgment Task

(24)

Figure 3. Marieke from the Binary Judgment Task.

She explains that the participants are going to see pictures and that she will try to describe the pictures in the best possible way. The pictures show a basket with several objects inside and outside of it, see Figure 4 for the overview picture:

(25)

Except for the teddy bear and the cat used in the practice items, the objects are all monosyllabic and inanimate. The participants are asked to indicate on the response box whether or not they think Marieke describes the picture correctly. They have to press the green button (correct) or the red button (incorrect) as quickly as possible. Responses and reaction times are registered, the reaction times are measured from the point where the pre-recorded sentence ends until one of the buttons is pushed. While the sentence is played, an empty basket is on the screen. The corresponding picture is presented immediately after the sentence is played, so that the reaction times measure the exact time it takes to look at the objects and make a decision about the sentence. Pictures and sentences are divided into 5 conditions, see Table 1:

Table 1.

Conditions in the Binary Judgment Task

Condition Picture Voice Correct Response Description Semantically

correct-1

Item A Item A Green Correct, one object

Semantically incorrect-1

Item B Item C Red Incorrect, one object

Items B + C Items A + C Red Incorrect, two objects

Semantically correct-2

Items A + C Items A + C Green Correct, two objects

Under-informative

Items A + B Item B Red/Green Under-Informative,

two objects

(26)

there is one object B in the basket and the voice describes the picture as “in the basket, there is a C.” This is a one-object-incorrect item, and participants are expected to press the red button. The semantically incorrect-2 condition consists of a picture showing objects B + C, whereas the voice mentions A + C. This is a two-object-incorrect item, and participants are expected to press the red button. The semantically correct-2 condition consists of a picture showing objects A + C, which are also both mentioned by the voice. This is a two-object-correct item, and a press on the green button is expected. The under-informative condition is the critical condition. The picture shows objects A + B, but the voice only mentions B: The sentence is informative. Participants generating the implicature or detecting the under-informativeness are expected to press the red button, but children who do not generate implicatures or do not detect the under-informativeness and participants who are Pragmatic Tolerant might press the green button. There are eight items in each condition, which are presented randomly. See also section 10.1 in the Appendix for the items. At the end of the task, Marieke thanks the participant for playing.

The Graded Judgment Task

(27)

Figure 5. The Likert scale from the Graded Judgment Task.

The small strawberry corresponds with a score of 1 point, the medium strawberry with 3 points, and the large one with 5 points. The pictures show a sofa on which various animals will be sitting, see Figure 6 for the overview picture:

(28)

Except for the rabbit and the two deer used in the practice items, all animals are monosyllabic. The participants have to indicate the size of the strawberries on the response box, left for the small strawberry, middle for the medium one, and right for the large strawberry. Only the responses and not reaction times are registered. The items are divided into five conditions, similar to the Binary Judgment Task. However, the two tasks differ only in response options, as shown in Table 2:

Table 2.

Conditions in the Graded Judgment Task

Condition Picture Voice Expected Response Description Semantically

correct-1

Item A Item A Large (5) Correct, one object

Item B Item C Small (1) Incorrect, one object

Items B + C Items A + C Small (1) Incorrect, two objects

Semantically correct-2

Items A + C Items A + C Large (5) Correct, two objects

Under-informative

Items A + B Item B Medium (3) Under-Informative,

Two objects

See section 10.2 in the Appendix for the items. There are again eight items in each condition, which are presented at random. At the end of the task, Marieke thanks the participant for playing.

The Theory of Mind Task

(29)

Hollebrandse and Van Hout (2008). This task investigates the ability for first-order false belief reasoning, and also second-order false belief reasoning in children. The items are presented in short movies, in which for example there is a person behind a window and a box, see Figure 7:

Figure 7. First-order Theory of Mind-item, the box.

During the movie, objects are placed in and removed from the box, sometimes person sees it, but sometimes she cannot because the curtains are closed. At the end of the movie, the child, who is the player of the game, has to tell the experimenter what the person thinks there is in the box. Possible answers are a) the apple that went in there first, b) the basket the person saw last, or c) the turtle that is in the box now, but not seen by the person. Theory of Mind failers often opt for the reality answer, c, whereas Theory of Mind passers should answer b.

(30)

Figure 8. Second-order Theory of Mind-item, the box.

This time, the same objects are placed in and removed from the box, while sometimes the left person and sometimes the right person cannot see. At the end of the movie, the child is asked what the right person would say that the left person thinks is in the box. Possible answers are a) the apple both persons saw was in the box, b) the basket only the left person saw, or c) the turtle only the right person saw, and which is still in the box. A second-order Theory of Mind passer would choose answer a.

In the original experiment there were four items for both the first-order and the second-order part. In this study, for reasons of time and space, we only used three movies from each part: Those featuring the box, the duck, and the treasure.

The Working Memory Task

(31)

seconds in between the words. It starts with a single word, and increases by one word after three items. In each series of a given number of words, the first item consists of words from one semantic category; the second item contains words partially from the same semantic category, whereas the third item has words from different semantic categories. See for example the series of three words in (8):

(8) a aap koe hond

monkey cow dog

b zon dak maan

sun roof moon

c boek haar ei

book hair egg

The largest of the series in the original Schlichting test consists of five words (Schlichting et al., 1995b). The test is designed to test children up to the age of seven. Young children are predicted to be able to recall up to four words in a row, whereas adults should be able to repeat six consecutive words. Since we need a similar measure for older children’s and adults’ working memory capacity as well, we will use the adaption made by Wubs (2008). She included additional items (from the Lexilijst (Schlichting, Van Eldik, Spelberg, Van der Meulen & Van der Meulen, 1995a), which indicates the words should be familiar to children from the age of two), leading up to series of nine consecutive words, which should be fairly impossible to correctly recollect.

(32)

5.3 Procedure

(33)

5.4 Design and Analysis

Based on the answers in the Binary Judgment Task, participants will be labelled as acting logically (press the green button for all under-informative sentences), or pragmatically (press mostly the red button for under-informative sentences). The group of participants acting logical will be divided further into those who are Pragmatically Tolerant, and those who do not compute implicatures. This division will be based on the performance in the Graded Judgment Task: Of participants who grade under-informative sentences around 3, it can be said that they are Pragmatically Tolerant. They accept under-informative sentences when given only two options (correct or incorrect), but show sensitivity to under-informativeness when given three options (small, medium, or large strawberry). The group of participants that does not seem to compute implicatures can also be further categorized, to make a distinction between participants who behave purely logical because they are not able to compute implicatures, and participants who deliberately choose to act logical. Veenstra (2010a) found that adults accepted under-informative sentences in a binary task, and graded those 5 in a graded task. Because it is unlikely that these adults are unable to compute implicatures, the reaction times will distinguish between inability and choice. We argue that participants behaving logically and who are not able to compute implicatures will show similar reaction times in the semantic conditions to those in the under-informative condition. Participants who behave logically because they choose to do so will have longer reaction times for the under-informative condition than for the semantic conditions, since computing an implicature and deciding to act against it will take time in addition to only looking at the objects and registering which are there. We will then categorize these adults as Pragmatically Tolerant as well.

(34)

(35)

6. Results

6.1 Binary Judgment Task

In the Binary Judgment Task, the participants had to press either a green or a red button as quickly as possible when they thought the sentence was right or wrong. Participants were at ceiling in accepting semantically correct items, as well as rejecting the semantically incorrect-1 items. Performance on the incorrect-2 condition was varied; we will discuss this later. In the under-informative condition, the children accepted on average 4 out of 8 under-informative sentences; whereas the adults accepted on average 6 out of 8 under-informative sentences. The numbers of acceptances by the children are more spread than those by the adults. Figure 9 shows the number of times the participants accepted an under-informative sentence:

Acceptances of Under-informativeness 0 5 10 15 20 25 30 35 40 45 50 0 1 2 3 4 5 6 7 8 Number P e rc e n ta g e Children Adults

(36)

For the children the numbers of acceptances are rather varied, yet, we see two main strategies: The majority of the children either rejects all under-informative sentences or accepts all of them. The adults’ behaviour also has a tendency toward these two strategies, but we see that most of them (48%) accept all under-informative sentences.

Also, the reaction times were measured; these are presented in Figure 10. Reaction times exceeding 10.000 milliseconds are considered outliers, they mostly occurred during interruptions from outside and when participants asked irrelevant questions (for example about the number of stickers they would get or about the performance of other children) during the experiment. These reaction times are left out of the analysis. We expected for the adults longer reaction times in the under-informative condition than the other conditions, these expectations were partly met, see Figure 10.

Reaction Times for All Participants

0 500 1000 1500 2000 2500 3000 3500

Corr-1 Incorr-1 Corr-2 Incorr-2 Under-inf

Condition R T i n m il li s e c o n d s Children Adults

Figure 10. The reaction times on the binary judgment task.

(37)

longer to press a button when there are two objects in the picture rather than only one (or two noun phrases are mentioned by the voice, rather than one), and it takes longer to disagree with a sentence rather than to agree with it. The reaction times in the semantically incorrect-2 condition are highest, followed by the under-informative condition. Paired t tests show that for the children, in both the semantically correct and the one-object-incorrect conditions they took significantly less time to make a decision than in the under-informative condition. The incorrect-2 condition took significantly longer, see Table 3:

Table 3

T test Results for all Children

Difference T value (74) P value Semantically correct-1 vs. under-informative -799.83 -6.17 .000 Semantically incorrect-1 vs. under-informative -494.11 -4.09 .000 Semantically correct-2 vs. under-informative -519.60 -4.26 .000 Semantically incorrect-2 vs. under-informative 333.65 2.83 .006

Although this pattern was not hypothesized for the children’s group, one should bear in mind that these results are for all children together, from age 4 to 9, whether or not they are aware of under-informativeness.

(38)

Table 4

T test Results for all Adults

We also looked at the different reaction times for purely logical children and adults, as presented in Figure 11. Reaction times should indicate whether participants are aware of the under-informativeness, which is predicted for the adults, but not the children. Participants are considered purely logical when they press the green button for all eight under-informative sentences.

Reaction Times for Logical Participants

0 500 1000 1500 2000 2500 3000 3500

(39)

The general pattern for the logical children (n = 22) is now different than that for the logical adults (n = 11). Reaction times for children in the under-informative condition are not much higher than the semantic conditions (leaving the incorrect-2 condition out of consideration) possibly indicating insensitivity to under-informativeness. And indeed t tests show that for the children, the reaction times in the semantic conditions do not differ significantly anymore from the under-informative condition, except for the incorrect-2 condition, which now does. The results are presented in Table 5:

Table 5

T test Results for Logical Children

Difference T value (21) P value Semantically correct-1 vs. under-informative -300.39 -1.59 .129 Semantically incorrect-1 vs. under-informative -104.81 -.86 .401 Semantically correct-2 vs. under-informative -111.69 -.61 .551 Semantically incorrect-2 vs. under-informative 774.75 2.60 .018

Please be reminded that this analysis still includes children who are possibly Pragmatically Tolerant. Results for the Graded Judgment Task, presented in the next section, will filter those children out of this group.

(40)

Table 6

T test Results for Logical Adults

Difference T value (10) P value Semantically correct-1 vs. under-informative -370.01 -5.024 .001 Semantically incorrect-1 vs. under-informative -226.67 -2.717 .022 Semantically correct-2 vs. under-informative -95.56 -1.071 .310 Semantically incorrect-2 vs. under-informative 60.30 .780 .453

These adults still make quicker decisions about the one-object semantic conditions, but the two-object conditions are no longer significantly different from the under-informative condition.

Now let’s have a look at the results for the pragmatic participants. Participants are considered pragmatic when they press the red button at least once in the under-informative condition. The results are presented in Figure 12:

Reaction Times for Pragmatic Participants

0 500 1000 1500 2000 2500 3000 3500

(41)

In this graph we can see that, except again for the incorrect-2 condition, both the children and the adults take longer to decide about the under-informative condition. And indeed, most of these differences proved to be significant, see Table 7 for the pragmatic children (n = 55) and Table 8 for the pragmatic adults (n = 12):

Table 7

T test Results for Pragmatic Children

Difference T value (54) P value Semantically correct-1 vs. under-informative _-981.44 _-6.26 _.000 Semantically incorrect-1 vs. under-informative _-635.68 _-4.10 _.000 Semantically correct-2 vs. under-informative -667.93 -4.50 .000 Semantically incorrect-2 vs. under-informative _173.24 _1.53 _.131

Table 8

T test Results for Pragmatic Adults

(42)

Logical vs. Pragmatic Children 0 500 1000 1500 2000 2500 3000 3500

Condition R T i n m il li s e c o n d s Logical Pragmatic

Figure 13. Reaction times for logical versus pragmatic children in the binary judgment task.

We made the same comparison between logical and pragmatic adults, and here we also did not find any significant differences within the conditions, see Figure 14:

Logical vs. Pragmatical Adults

0 200 400 600 800 1000 1200 1400

Condition R T i n m il li s e c o n d s Logical Pragmatic

(43)

The children’s patterns nevertheless point out the difference between the logical and pragmatic approach. On the other hand, the logical adults show the same pattern as the pragmatic adults, indicating that they are not logical, but rather Pragmatically Tolerant. A paired t test revealed pragmatic adults responded 96 milliseconds slower than Pragmatic Tolerant adults: t (4) = -5.56; p <.01.

6.2 Graded Judgment Task

In the graded judgment task participants had an extra option to judge the sentences; they could choose a small, medium, or large strawberry to indicate the degree of ‘correctness,’ corresponding with 1, 3, or 5 points. Results of this task should make a distinction among children who behaved logically in the Binary Judgment Task between those who do not detect under-informativeness and those who are Pragmatically Tolerant. The mean number of points all participants awarded in the several conditions is presented in Figure 15:

Graded Judgment for all Participants

0 1 2 3 4 5

Corr-1 Corr-2 Incorr-1 Incorr-2 Under-inf

Condition N u m b e r o f p o in ts Children Adults

(44)

We can see that the conditions in which the picture exactly matches the sentence, the semantically correct-1 and -2, score around 5 (the large strawberry) for both the children and the adults. The incorrect condition with one object scores for both groups of participants around 1 (the small strawberry), whereas the two-object incorrect items were awarded an average of 2 points. The critical under-informative condition was mostly awarded 3 points (the medium strawberry). This is what one would expect. An additional one sample t test on the under-informative condition shows that the children did not significantly differ from the expected 3 points, but the adults did. They awarded on average 0.9 points more with a t (23) = 3.065; p < .001.

When we look at the awarded strawberries for the logical participants separately (participants who were categorized as logical because of the Binary Judgment Task), we see roughly the same pattern, except for the higher valuation of the under-informative condition in both groups, see Figure 16:

Graded Judgment for Logical Participants

0 1 2 3 4 5

(45)

A t test revealed that the logical children awarded on average 0.7 points more than three to the under-informative condition (t (19) = 2.818; p <.05. The adults awarded even 0.9 points on average more than three (t (12) = 5.159; p <.01. This indicates that the logical and Pragmatic Tolerant participants behave rather logical in the Graded Judgment Task as well as in the Binary Judgment Task.

The results for the participants that were pragmatic in their approach to under-informativeness are presented in Figure 17:

Graded Judgment for Pragmatic Participants

0 1 2 3 4 5

Figure 17. The graded judgment of the pragmatic participants.

(46)

We compared the logical children with the pragmatic children, see Figure 18:

Logical vs. Pragmatic Children

0 1 2 3 4 5

Condition N u m b e r o f p o in ts Logical Pragmatic

Figure 18. The graded judgment for the logical and pragmatic children.

The difference of 1.28 points in grading by the children in the critical condition is significant: t (53.52) = 4.724; p <.00.

(47)

Logical vs. Pragmatic Adults 0 1 2 3 4 5

Condition N u m b e r o f p o in ts Logical Pragmatic

Figure 19. The graded judgment for the logical and pragmatic adults.

Since there is a difference between acting logically because you have not acquired implicatures yet, and because you choose to, we will also analyze the combined results of the Binary and Graded Judgment Task on a more individual level. This subject analysis is presented in the next section.

6.3 Binary vs. Graded Judgment Task

(48)

Table 9

Logical Approach in Children Age (Months) Reaction Time Correct-1 Reaction Time Under-informative Grading Under-informative Mean 69.6 2897.488 2856.659 4.875 StDev. 17.25 3598.256 2680.25 0.21

A paired samples t test reveals that there is no significant difference between the reaction times in the semantically correct-1 and the under-informative condition: T (9) = 1.28; p >.9. Neither was there between the other semantic conditions and the under-informative, but for this group we take the one-object correct condition as a baseline. Logical children only need to check whether the object mentioned is there, and do not have to worry about any other objects that might be there to accept the sentences.

Some adults would also fall in the category of the logical approach because of their performance on the Binary and Graded Judgment Tasks. That is, they also accepted under-informativeness to the max just like the children. However, they are assumed to have acquired implicatures and might consciously choose to accept under-informative sentences in the Binary Judgment Task and grade them towards 5 points in the Graded Judgment Task while they are aware of the under-informativeness. Seven adults fall into this category, one man and six women. See Table 10 for the results:

Table 10

(49)

A paired samples t test reveals that there is a difference between reaction times for the semantically correct-1 condition and the under-informative condition: The decision about the a semantically correct item takes 383 milliseconds less than the decision about an under-informative one, t (6) = -3.271; p <.05. We therefore conclude that these adults should be categorized as Pragmatic Tolerant.

The Pragmatic Tolerant group consists of participants that accept under-informative items in the Binary Judgment Task, but show sensitivity either in the Graded Judgment Task by grading items the under-informative condition towards 3 points or below (or around 5 for adults). There are ten children and thirteen adults that fall into this category. Amongst the children there are four boys and six girls, and amongst the adults (including the adults first described as logical) there are two men and nine women. See Table 11 for the results:

Table 11

Pragmatic Tolerant Approach of Children, Adults and Total Children Age (Months) Reaction Time Correct-1 Reaction Time Under-informative Grading Under-informative Mean 77.3 1631.013 2272,619 3.275 StDev 17.99 637.877 957.426 0.40 Adults Age (Years) Mean 31 657.227 1027.239 4.340 StDev. 12.569 138.408 325.231 0.868 Total Mean 1120.935 1620.277 3.833 Total StDev 664.101 933.592 0.864

(50)

The remainder of the participants does detect under-informativeness and acts accordingly and therefore falls in the category of pragmatic behaviour. They mostly reject under-informative items in the Binary Judgment Task, and rate them around 3 points or below in the Graded Judgment Task. There are 55 children (23 boys and 32 girls) in this category, and 12 adults (seven men and five women). These results are presented in Table 12:

Table 12:

Pragmatic Approach to Under-informativeness

Children Age (Months) Reaction Time Correct-1 Reaction Time Under-informative Grading Under-informative Mean 76 2014.773 2996.214 2.795 StDev 20.26 1049.221 1616.94 1.38 Adults Age (Years) Mean 45.5 763.521 1118.561 3.167 StDev 15.26 212.75 301.41 1.07 Total Mean 1790.668 2659,918 2.862 Total StDev 1068.61 1643,811 1.33

A paired samples t test shows that the difference of 869 milliseconds in which participants respond quicker to semantically correct items than to under-informative items appears to be significant: t (66) = -6.565; p <.00.

6.4 Theory of Mind Task

(51)

Of them, only 7 children also passed the second-order part (with a score of 3 out of 3). The results are presented in Table 13:

Table 13

Results of Theory of Mind Task

Number of participants Mean Age Stdev Age

No ToM ToM1 ≠ 3 20 4;9 0;7

ToM1 ToM1 = 3, ToM2 ≠ 3 49 6;8 1;5

ToM2 ToM1 = 3, ToM2 = 3 7 8;2 0;11

In this data there is a clear trend towards a positive correlation between ToM-score and age, and indeed, the correlation between the total of both ToM-scores and Age is significant: r = .691; p <.01. See the scatter plot of Figure 20:

Spread of ToM scores against Age

0 1 2 3 4 5 6 45 55 65 75 85 95 105 115 Age in months T o M 1 + T o M 2

(52)

6.5 Working Memory Task

Both the children and the adult participants took the Working Memory Task. The adults had a larger working memory capacity than the children, as can be seen in Figure 21:

Figure 21. The working memory scores.

The working memory scores for the children had a mean of 8.43 and a standard deviation of 1.41, whereas the adults on average scored 13.1 with a standard deviation of 2.56. Also working memory was strongly correlated with age: r = .504; p <.00.

6.6 Correlations

(53)

(54)

7. Discussion

7.1 Pragmatic Tolerance

From the data of the Binary and Graded Judgment Task it is evident that there is indeed a phenomenon described by Katsos and Bishop (under review) as Pragmatic Tolerance. Certain children accept under-informative sentences in the Binary Judgment Task, but do not choose the largest strawberry in the Graded Judgment Task. If they had been tested in a binary paradigm only, they would have been falsely categorized as not sensitive to under-informativeness. In this study, 10 out of 75 children and 11 out of 23 adults would have been mislabelled in a binary judgment paradigm only. The percentage of mislabelled children, 13.3% is much lower than the percentage of mislabelled adults, 47.8%. This possibly indicates that there is a difference between involuntary Pragmatic Tolerance in children and voluntary Pragmatic Tolerance in adults, as we observed in section 2.4. The adults might try to see through the purpose of the experiment, thinking that there is a hiding snag and overruling their initial rejection of under-informativeness.

(55)

This result emphasizes the need for a sound methodology: The true competence of participants might be clouded by the experimental set-up. Reaction times are a purer measure than sentence evaluation, because accepting an under-informative sentence is never totally wrong even if you know it is under-informative. The longer reaction times indicate there is more going on than a simple finding and checking the mentioned object. Although we argued earlier it is difficult to distinguish between detecting under-informativeness and generating an implicature, whichever of the two it is the sensitive participants do exactly, it is more costly than accepting a straightforward correct sentence, similar to findings in earlier studies (Noveck & Posada, 2003; Bott & Noveck, 2004; Feeney et al., 2004; Pouscoulous et al., 2007).

Katsos and Bishop (under review) proposed an action based task, in which participants have to make the contents of a container match an under-informative utterance. If the container contains an apple and a pear, and the voice mentions only the apple, removing the pear would mean the participant actually generated the implicature. Using this paradigm, perhaps with a reaction time element could show the difference between detecting under-informativeness and generating implicatures, and might distinguish Pragmatic Tolerant participants into those who are overruling the detection of under-informativeness and those who overrule generating an implicature.

(56)

under-informativeness responded significantly slower than the Pragmatic Tolerant participants (t (60.192) = 3.641; p <.01). This large difference might also be due to the large proportion of children in the Pragmatic Tolerant group, who respond slower than adults in general, but when we look at adults alone, the pragmatic adults still respond slower than Pragmatic Tolerant adults.

Furthermore, it would be interesting to investigate the development of informativeness in a eye-tracking set-up. Such data might reveal more about the strategies participants adopt, for instance whether logical participants in the under-informative condition do or do not consider the unmentioned object in the picture. This would strengthen our decision to take the one-object condition as a base-line to compare the under-informative condition to.

From our results it is difficult to conclude the age at which children start to acquire sensitivity to under-informativeness. It is clear, however, that this is a linear development, young children initially interpret under-informative sentences logically (in our study, these children have a mean age of 5;9) and then become either Pragmatically Tolerant (mean age 6;5), or pragmatic (6;4). This supports the schematic representation suggested in chapter 4, see Figure 22:

Competence:

Behaviour:

Figure 22. Approaches to under-informativeness

(57)

In previous studies on the acquisition of implicatures, adult control groups were never reported to have an acceptance rate as high as in this present study. The majority of these studies however, investigated generalized scalar implicatures (of the <some, most, all> -type) (Guasti et al., 2005). It is possible that the different nature of the two types of implicature plays a role: The Default Hypothesis (Levinson, 2000; Chierchia, 2004) argues that generalized implicatures are lexicalized, and therefore easier to compute than particularized implicatures that are generated based on context only. This would explain the variety of approaches especially in the adult control group; the particularized implicatures are not set as default and are more subject to individual differences. And as Pragmatic Tolerance might be a least-demanding and time-saving strategy to deal with under-informativeness, it is conceivable that participants more often resort to Pragmatic Tolerance for particularized implicatures than for generalized implicatures. The Relevance Hypothesis (Sperber & Wilson, 1986/1995) on the other hand, that claims both types of implicatures are computed equally according to the relevance of the implicature does not provide an explanation for the high rate of Pragmatic Tolerance in this study. The relevance of the computation of the implicature is made explicit by the instructions of both tasks, stating that the girl will try to give the best description possible, and that the participant has to indicate whether they think she has succeeded. However, this did not yield many pragmatic responses.

(58)

of them was mentioned correctly with the addition of another object. That other object was in the picture, but not in the basket, see Figure 23:

Figure 23. A semantically incorrect item with two objects.

(59)

7.2 Theory of Mind

The children in this study showed to be at different stages of their development of a Theory of Mind. We predicted to find a relation between informativeness and ToM-scores, since perspective-taking and knowledge about others’ intentions are required to evaluate a speaker’s options for describing a situation. We found that the ToM-scores correlate with reaction times in the Binary Judgment Task; a higher ToM-score is related to a shorter reaction time in the under-informative condition (with the effects of age and working memory left out of the equation). Adults, who are assumed to have both a first- and a second-order Theory of Mind, are quicker than children in deciding about under-informativeness. This result pleads in favour of hypothesis 2, which claimed that Theory of Mind is needed for the detection of under-informativeness and generation of implicatures.

(60)

to 20 who did not have any Theory of Mind, and 49 who had only a first-order Theory of Mind. A more balanced age range might include more second-order ToM passers, and influence the correlation.

Yet the other conclusion one might draw is that Theory of Mind is simply not related to the approach towards informativeness. The adult control participants did not behave unequivocally in their strategies towards under-informativeness; some behaved purely logical, some Pragmatically Tolerant, and some pragmatic. Based on our study, we can conclude that there is not one approach that can be labelled as typically ‘adult-like’ in contrast to earlier studies that believed the rejection of under-informativeness was the ultimate end-state in the development of informativeness. There is not one single preferred approach for adults; it is therefore not justified to expect children to develop this one alleged adult-like approach and relations between ToM-scores and rejections of under-informativeness should not be predicted.

Pragmatic Tolerance and Theory of Mind in Children’s Comprehension of Under-Informativeness