3. Methodology
3.2 Experiment 2
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
54
which one is the cake. For the comprehension task, children had to choose one answer from three objects, in which one was the target answer, and the other two were the distracters. And all the target objects in this task were the test objects which were presented to children in the previous experiment. All of the objects shown to children in word comprehension task are presented in the appendix.
3.2 Experiment 2
In the experiment 2, the effects of previously retrieved words and perceptual similarity on children‟s naming were compared between 2- and 4-year-old children.
The design of experiment 2 followed again the study in Gershkoff-Stowe et al. (2006), but we modified the priming pictures. There were three groups in this experiment.
One was pictures with high similarity to test objects, another was pictures with low similarity, and the other one was unrelated pictures to the test objects. The third group acted as a control group to compare and contrast with the first and second groups to see the effects on children‟s naming when there is no perceptual similarity between test objects and priming. We expect that children who participated in this condition will produce more extraneous answers since they are not primed by related concepts.
And high and low similarity priming were used to examine how the priming and similarity affect children‟s naming behaviors.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
55
In this experiment, novel objects were used. The novel objects which we used in this experiment were from the study of Gershkoff-Stowe et al. (2006). And for the target names in high-similarity and low-similarity conditions were decided from adults‟ imagination and judgments, which would be illustrated in the section 3.2.3.
The purpose of using novel objects in this experiment was to test the differences of word choosing between 2- and 4-year-old children. We expected 4-year-old children would be less susceptible to previously retrieved words since it‟s reasonable that 4-year-old children have more mature vocabulary and concept base than younger children, they may have better performances toward naming.
3.2.1 Procedure
Children at each age level were evenly distributed to three sets. All the procedures in this experiment were the same as the experiment 1. The only difference between these two experiments was that the experiment 2 had eight trials rather than six. And after the end of the experiment, there was no comprehension test for objects since there is no real name for the novel objects in this experiment.
3.2.2 Subjects
Twenty two 2-year-old (mean=2;5) and 25 4-year-old ranged from 3;11 to 4;11 (mean=4;4) children participated in this experiment. They were all normal Chinese-speaking children and had no brain damage before. Two-year-old children
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
56
were the same as the experiment 1. And 4-year-children were recruited from six kindergartens in Taipei.
3.2.3 Materials
The novel test objects followed the study of Gershkoff-Stowe et al. (2006). And all of the priming pictures were modified. They were decided from adults‟
imagination and similarity judgments.
Ten adults participated in the imagination task. Eight novel objects were shown to them. And they were asked to write down what they thought of based on shape similarity when they saw these novel items. Each adult was asked to write down two to three items they could think of and these items must have some perceptual similarity with those novel objects presented to them. After 10 adults had completed, we collected all the answers they had provided. After that, the corresponding pictures of those objects were found. By comparing with the target object, we thought of some unrelated items by our intuition for ready to be 16 fillers and unrelated targets in the experiment 2. And all the unrelated and related items were judged by other 10 adults for their similarity to target objects. These 10 adults did not participate in the imagination task before. In the adult judgment task, adults were shown a sheet full with many pictures. They were asked to grade the similarity between the target objects and the pictures we had chosen. The scale was ranged from 1 to 10. Scale 1
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
57
represented the lowest similarity and 10 represented the highest. These 10 adults were asked to mark a point which would show the similarity between the target object and the judged item. After grading, the total points of each item were counted. The one which receives most points was the item used in the high similarity condition, the one receiving lower points was the item used in the low similarity condition, and the ones receiving least points were the items used in the unrelated control and fillers. The mean rating of high similarity items was 6.73 (S.D=2.64). The mean rating of low similarity items was 4.35 (S.D=2.28), and the mean of unrelated items was 1.09 (S.D=0.28). Finally, the mean of fillers was 1.26 (S.D=0.28). And the similarity of target primes in three conditions to the novel objects reached significant difference, F(2, 237)=156.50, p<0.001. And all the test items used in experiment 2 were shown in
appendix.
Target priming pictures were divided into three sets. One is high similarity set, another is low similarity set, and the other is unrelated control set. Every priming set included three pictures, i.e., one target prime and two unrelated fillers. And the 16 fillers are shown in the appendix.
3.2.4 Coding
In the experiment 2, we adopted the same coding as Gershkoff-Stowe et al. (2006) for data analysis. The classification was similar to experiment 1. Children‟s responses
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
58
were coded as the following four types: omission, extraneous, prime target, and filler.
The meaning of each type has explained in the coding section in experiment 1. What‟s different is that in experiment 2, prime target was divided into three kinds, i.e., high, low, and unrelated. If children in the high similarity condition gave the responses of the high target, the answers would be coded into the „high‟ kind. However, if children in the high similarity condition did not give its target prime, rather, they gave the target prime in the low or unrelated similarity conditions, the answer would be counted into „low‟ or „unrelated‟ kind. In the other two conditions, the coding of target primes was the like. Besides, another type of error would be reported separately, i.e., perseverative naming. This kind of naming refers to the naming which was perseverated from previous naming. But there was no perceptual similarity between these two objects at all.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
59
Chapter 4 Result and Discussion
In this chapter, we would display the results of our two experiments and make a discussion. 4.1 to 4.3 are the results of experiment 1. In 4.1, we compare children‟s performances of comprehension and production. In 4.2, we describe children‟s responses of object naming. And in 4.3, the priming effects in experiment 1 are examined. 4.4 to 4.6 are the results of experiment 2. 4.4 show the comparison and contrast of the novel object naming between children aged 2 and 4. In 4.5, we examine the priming effects in this experiment. And in 4.6, the results of children‟s perseverative naming are discussed. Finally, in 4.7, a brief summary for the chapter 4 is presented.
4.1 Children’s Word Comprehension and Production in Experiment 1
We have checked whether familiar test objects are better known to children in the word comprehension task. Children performed better in the familiar condition and worse in the unfamiliar one. And the comparison of correct percentage between children‟s comprehension and production performances is shown in the following figure.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
60
Figure 4. The Comparison Correct Percentage between Children‟s Comprehension and Production Performances
From this figure, we can observe that in the comprehension task, children gave correct responses up to 81.82% in the familiar condition, while only 39.4% was correct in the unfamiliar condition. As for production performances, children also performed better in the familiar condition, they had 45.4% correct responses. In the unfamiliar condition, their percentage of correct responses was lower (9%). Based on ANOVA test, children‟s correct responses had significant relations with both familiarity and comprehension. They had significant better performances in the familiar condition, F(1,31)=12.38, p=0.0014<0.05. And their comprehension was significantly better than
their production as well, F(1,31)=17.22, p=0.0002<0.05. There was no interaction effect of these two factors, F(1,30)=0.08, p=0.7856>0.05. According to this data, we could corroborate that familiar test objects are better known for children and
‧
children‟s comprehension is better than their productions. Their performance errors do not necessarily reflect their comprehension errors. The result is consistent with previous literature (Huttenlocher, 1974; Thompson & Chapman, 1977; Fremgen &
Fay, 1980; Rescorla, 1980a; Naigles & Gelman, 1995; Gelman, Croft, Fu, Clausner, &
Gottfried, 1998).
4.2 Object Naming in Experiment 1
After children were primed by three pictures, they named the target objects. And their naming results are shown in the Table 4:
Table 4. Children‟s Naming Results in the Object Naming Task
Familiar Unfamiliar
Frequency Total Frequency Total
Correct 45.4% 45.4% 9% 9% condition (45.4%) than in the unfamiliar one (9%). And for the omission answer type, children kept silence more often in the unfamiliar condition (24.2%) than in the familiar one (16.7). There is an explanation for the fact. When children confronted some objects with unknown names, they could not find a proper word to label the objects so that they did not give a response and just kept silence.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
62
Extraneous naming happens much more often in the unfamiliar condition (66.7%) than the familiar condition (37.8%) in Mandarin children. The cues for children to give extraneous naming in the object naming task were shape, texture, function and color. For example, children named bang4bang4tang2 „lollipop‟ as gua1niu2 „snail‟.
This is the naming based on shape similarity. The shape similarity between those two was that they both had one circle with spiral on it. Other overextension naming example based on shape was like that children named liang2jiao3qi4 „protractor‟ as xi1gua1 „watermelon‟. For the naming based on the texture, children named fa3dai4
„hair band‟ as nei4ku4 „underpants‟ or wei2jin1 „scarf‟. They were both made of cloth.
Notably, children‟s overextension naming based on the texture all appeared in the fa3dai4 „hair band‟ item. Since fa3dai4 „hair band‟ is not a hard and firm object, this
kind of feature may lead children‟s attention to the texture rather than shape. Children sometimes may overextend based on function as well. For example, children named lan2zi5 „basket‟ as he2zi5 „box‟ or pan2zi5 „plate‟, which may come from the same
function of these items, i.e., they all can load things. Moreover, the example of
naming based on color is that yuan2bao3 „odd-shaped gold‟ was named as yue4liang4
„moon‟. However, besides color, this overextension may be still based on shape at the
same time. This result was consistent with previous findings found in Clark‟s (1993) study. Clark (1973) proposed that children seem not to do overextension based on
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
63
color. We also did not find overextension case based on color criterially in Mandarin Chinese. If children overextend a certain object which has the same color with the target, color seems not to be the only feature on which children based. Rather, color often combines with other features at the same time. In addition to the naming with clear cues, sometimes children may give some naming which does not have apparent relations to the target objects. For example, a child named yu2wang3 „fish net‟ as bao3ling2qiu2 „bowling ball‟.
For the control naming type, we can observe from table 4 that control errors seldom appear both in the familiar and unfamiliar condition (3% in the familiar condition and 7.6% in the unfamiliar one). When children were not primed, the possibilities for children‟s naming being affected were both low in two conditions.
In the target types, the percentage in the familiar condition was lower (7.6%) than in the unfamiliar condition (21.2%). It seems that Chinese children were easier to make errors on those things that they felt unfamiliar.
Finally, children gave filler answers 4.5% of the time in the familiar condition and 1.5% in the unfamiliar condition. Both of them are not high since fillers have no perceptual similarity with the target objects.
As a whole, erroneous naming (extraneous, control, target, and filler) happens more often in unfamiliar condition than in familiar one (66.7% vs. 37.8%). This result
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
64
could imply that concept and word have strong influences on children‟s word selection. Unfamiliar words seem to be affected easier and thus have a higher frequency of errors. This may be related to the assumptions that through repetitive practice, words could gain stronger linking among object, concept, and word.
Unfamiliar words were used less than familiar ones. Thus, the naming of unfamiliar words would be interfered easier. This finding was consistent with Gershkoff-Stowe et al. (2006) as well, which may imply that this should be a universal phenomenon.
4.3 Priming Effects in Experiment 1
According to Gershkoff-Stowe et al. (2006), the design was to compare children‟s naming when they were primed (children in the set A provided A target prime answers) versus when they were not primed (children in the set A provided B target prime (control) answers). To understand whether there are the same priming effects in Mandarin Children as in English children, (2) × (2) ANOVA test were adopted to examine the effects of tasks (set A/set B) and conditions
(familiar/unfamiliar) on children‟s object erroneous naming. In the four error types,
„extraneous‟ and „filler‟ are not affected by different tasks and conditions. However,
there is a main effect of condition on children‟s „target‟ error type. The ANOVA result is shown in the table 5.
‧
Table 5. The Effects of Conditions and Tasks on Children‟s „Target‟ Error Type
Source DF Type I SS Mean Square F Value Pr > F children‟s „target‟ error naming. The shape similarity with the objects should have the same effects on children‟s naming. However, different conditions (familiar or unfamiliar) seem to have significantly different effects on children‟s „target‟ error naming (SS=3.68, MS=3.68, F(1,18)=5.88, p=0.0260<0.05), and there was an interaction effect (SS=9.22, MS=9.22, F(1,18)=14.73, p=0.0012<0.05). Children in the unfamiliar condition were susceptible to be influenced by perceptually similar target prime pictures easier. This result was not consistent with Gershkoff-Stowe et al‟s (2006) study. Their findings showed that familiarity should not cause children to have different reactions to target priming pictures. Thus, they concluded that the underlying mechanisms among category errors, pragmatic errors, and retrieval errors should be the same. Nevertheless, in Mandarin children, the situation appears to be another case. Children made less overextension errors from the effect of target prime
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
66
pictures previously shown to them in familiar condition. On the contrary, it is much easier for children to make „target‟ error naming in the unfamiliar condition. This should be explained by two reasons. The first one is because of activation strengths.
Activation strengths will be stronger when the use frequency of a word is higher (Dapretto & Bjork, 2000). It is reasonable that familiar words have higher frequency of retrieval and practice in children‟s daily life. Thus, stronger strengths may make children have fewer possibilities to be affected by previously primed perceptually similar target pictures. The second reason may come from different types of languages.
Since Mandarin Chinese and English are two different languages and thus have different morphology, this may be the reason to lead to different results. Chinese words/compounds are often composed of two syllabic morphemes, which bring phonological and semantic information at the same time while English words are composed of letters, which often bring just phonological information (Mok, 2009;
Toyoda & Scrimgeour, 2009). And studies have shown that activation strengths would be stronger when the perceived thing is semantically/perceptually similar to the image in our concept (Huttenlocher & Kubicek, 1983; Martin, Weisberg & Saffran, 1989).
Thus, concept (semantic/perceptual) relations will fortify the strengths in people‟s mental lexicon. Thus, the characteristics of Mandarin Chinese morphology may cause words/compounds to be less vulnerable to be influenced than English because words
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
67
in Mandarin mostly own information more than just phonology, which may let Chinese vocabularies have stronger strengths than English naturally. It is more difficult for Chinese words to be overextended because of its natural stronger strengths. Especially when the words are familiar, the strengths are double. Therefore,
conditions (familiar/unfamiliar) become a significant factor to affect children‟s
„target‟ error naming.
Although familiarity becomes a significant factor to influence children‟s error naming in Mandarin children, we do not want to neglect the effects of spreading activation mechanism on children‟s naming (Dell, 1986). In fact, it still could be used to explain children‟s three types of overextension errors because we could not deny the fact that even in the familiar condition, the frequency of „target‟ error naming (7.6%) was more than „filler‟ (4.5%) or „control‟ (3%). This means that the strengths of perceptually similar target prime pictures are still stronger than the other two.
Similarity indeed makes the target primes become a more powerful competitor to intervene children‟s word retrieval despite the fact that the power of this interference is not that strong as in English. Thus, we still want to agree that the underlying mechanisms among three types of overextension errors should be the same.
Regarding to the last kind of erroneous naming, i.e., „control‟, the results of ANOVA test are displayed in the table 6.
‧
Table 6. The Effects of Conditions and Tasks on Children‟s „Control‟ Error Type
Source DF Type I SS Mean Square F Value Pr > F Task (set A/ set B) 1 1.96 1.96 12.62 0.0023 Condition (Familiar/Unfamiliar) 1 0.73 0.73 4.68 0.0443 Task×Condition 1 0.87 0.87 5.61 0.0292
In this table, we could observe that both Task (SS=1.96, MS=1.96, F(1,18)=12.62, p=0.0023<0.05) and Condition (SS=0.73, MS=0.73, F(1,18)=4.68, p=0.0443<0.05)
had main effects on children‟s „control‟ naming. And there was an interaction effect of Task and Condition (SS=0.87, MS=0.87, F(1,18)=5.61, p=0.0292<0.05).
In fact, almost all the „control‟ naming data came from the same trail: trail 3 in experiment 1. Children mostly still replied qiu2 „ball‟ when they were primed by pin2guo3 „apple‟. In other trails, the „control‟ naming scarcely happened. This fact
leads to the result that Task became a significant factor to influence children‟s
„control‟ naming. Gershkoff-Stowe et al. (2006) mentioned that when labeling an
object, several related concepts will be activated at the same time, and the one which receives the strongest strength will win and be retrieved successfully. It appears that qiu2 „ball‟ is a strong competitor to affect children‟s naming when they saw a round
thing. This suggests that qiu2 „ball‟ in Mandarin should have stronger activation strengths among the competitors with round shape so that it has possibilities to overpower the target prime pictures in some cases. Besides, condition also plays a significant role in children‟s „control‟ naming as well. This is reasonable since
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
69
familiarity will make the strengths become stronger so that the possibilities of other erroneous retrieval will decrease accordingly.
In this experiment, some facts were revealed. First, shape similarity indeed will fortify the strengths of the word to intervene children‟s word retrieval even though this kind of effect was lower for the familiar objects in Mandarin Chinese children.
Second, priming is a necessary factor to make children to give „target‟ error naming especially in the unfamiliar condition because the strengths for the unfamiliar objects should be weaker in children‟s mental lexicon. Third, word retrieval errors should not be totally affected by previously accessed words. Rather, concepts and words in our
Second, priming is a necessary factor to make children to give „target‟ error naming especially in the unfamiliar condition because the strengths for the unfamiliar objects should be weaker in children‟s mental lexicon. Third, word retrieval errors should not be totally affected by previously accessed words. Rather, concepts and words in our