Chapter 3. Experimental Designs and Experimental Tasks
3.1 Experimental Design
3.1.1 Variables
Variables in this study include the semantic specificity of words to be learned
(general versus specific) and the test condition (pre-conditioning-training test versus
post- experimental-training test). For visual stimuli, two sets of novel actions involving
breaking and carrying events were designed. These two types of actions were selected
because previous studies have shown that many languages differed in specificity or
semantic categories of carrying verbs and breaking verbs (e.g., Bowerman, 2005;
Brown, 2001; Tardif, 2006a). Also, the two types of actions differ in some factors such
as punctuality, affectedness, and kinesis. The differences might help to balance the
unknown effect of perceptual factors on learning outcome.
Children’s performance in both the pre-test session and the post-test session for
each condition was compared so as to explore the effect of conditioning training. In the
production task, through comparison between the pre-test session and the post-test
session, we could explore if children can use an additional label to encode another
action in the Specific Condition or if children learning a general label performed better
after exposure of visual stimuli with more variation and higher frequency for a
particular label. As for the comprehension task, we examined if children could properly
identify the subtle semantic boundary as encoded in the novel word(s) presented in the
reinforcing conditioning training.
We also made comparisons between the General Condition and Specific Condition
in their performance in production and performance on extension tasks to probe into the
effect of specificity. Specifically, this study examined whether children under the
Specific Condition in which more words were presented to describe the same set of
visual stimuli were more likely to learn an additional word than the General Condition.
Also, we explored whether children under the Specific Condition would be more likely
to develop a one-to-one mapping strategy and avoid extending the novel words to other
contexts when compared to the General Condition.
3.1.2 Counterbalancing
In the reinforcing conditioning training, in the Specific Condition, two similar
actions with slightly differences would be mapped onto two different novel verbs
whereas in the General Condition they would be mapped onto one novel verb. A
participant who was assigned to the Specific Condition when presented with the
carrying actions would be assigned to the General Condition when shown the breaking
actions (i.e., Order 2 and Order 3 in Appendix 1), and vice versa. Therefore, the total
number of novel words to be learned throughout the experiment for each participant was
controlled. Each participant would hear three novel words, two of which were baseline
verbs, and the other of which was an additional word in the Specific Condition. The
novel word for the baseline carrying action was mū and that for baseline breaking action
was dē. In addition, the word for the additional word for carrying actions or breaking
actions was fō. All of these words are gap words, consisting of non-occurring syllables,
in Mandarin.
3.1.3 Confounding factor: Input frequency
In our study, we manipulated the existence of the contrast provided by novel labels
but controlled exemplars of actions across conditions. However, conducting an
experiment on mapping between actions and different number of label, we should make
a compromise between controlling the total number of exemplars for actions as well as
the total input frequency and controlling the input frequency for each label. On the one
hand, if we control the total number of exemplars of actions and the total number of
exposures of labels as shown in Table 3.2 (a), this would result in the discrepancy in the
frequency for each label across conditions (e.g., Label 1 in the General Condition was
presented for 18 times whereas Label 1 in the Specific Condition for 9 times). On the
other hand, if we control the input frequency for each label as shown in Table 3.2 (b),
discrepancy in the total number for actions and labels across the two conditions would
occur: The participants under the Specific Condition would be presented with a larger
amount of visual stimuli as well as linguistic stimuli when compared to those under the
General Condition.
On the basis of the assumption that language conventions vary in semantic
boundaries while perceptual capacity is constant (Bowerman, 2005; Majid, Bowerman,
van Staden, & Boster, 2007) and the observation that frequency of a semantically
general verb is usually higher since it can applied to a wider context11 (Casenhiser &
Goldberg, 2005; Gordon & Dell, 2003), this study controlled the total number of actions
and total number of labels shown and left frequency of each word different: The input
frequency of an individual label in the General Condition would be higher than that in
the Specific Condition. Therefore, when comparing the performance on Label 1 in one
condition to that in the other condition, input frequency would inevitably be a
confounding factor though we aim to examine the effect of specificity. In this sense,
children under the Specific Condition would encounter a more difficult task since they
11 Crosslinguistically, the example of carry used in Tardif’s (2006) study can serve to illustrate the relationship between specificity and frequency. Since Mandarin has more types of carrying verb roots than English, the frequency of carry in English should be relatively high when compared to specific verbs in Mandarin when speakers of both languages are asked to describe a variety of carrying events. Also, within a language, there are usually some general words used frequently since they can be used in a variety of contexts (Casenhiser & Goldberg, 2005; Gordon & Dell, 2003).
were presented with fewer exposures for each individual label and they had to
distinguish between two similar verbs. Better performance observed in the Specific
Condition would be strong evidence for Tardif’s account that specificity facilitates word
learning; nevertheless, two possible accounts would be aroused if better performance
was found in the General Condition: (i) frequency facilitates word learning and (ii)
specificity impedes word learning.
Table 3.2: Different designs for mapping in the conditioning training session
(a) Controlling total number of exposure of actions and total number of labels
General Specific
(b) Controlling the number of frequency for each verb
General Specific
As shown in Table 3.3, the production test session consists of five trials. For both
conditions, two trials were administered before the conditioning training (pre-test) and
three trials after the conditioning training (post-test). Pre-test trials in the two conditions
were identical whereas some differences lied in post-test trials across the conditions.
One of pre-test trials and two of post-trials were designed to test the participants’
responses to the actions they had seen in the training sessions; on the other hand, in
training sessions, children had not been shown the action appeared in one of pre-test
trials (Trial 2) and one of post-test trials (Trial 5) since these trials were designed to
examine extending uses of the novel words. As shown in Table 3.3, the video clips in
Trial 2 and Trial 5 showed Action 3, which was similar to Action 1 and Action 2 but
differed from them in the body part involved. The characters in the video clips of the
production task were different from the puppet used in the demonstration for the
training sessions yet the same across trials of the production task.
Table 3.3: Trials of the production task
General Specific
Test condition
ActionTarget Label
VideoTarget Label
(1) a Pre-test Action 1 Label 1 Action 1 Label 1(2) Pre-test Action 3 b - Action 3 b - (3) Post-test Action 1 Label 1 Action 1 Label 1 (4) Post-test Action 2 Label 1 Action 2 Label 2 (5) Post-test Action 3 b - Action 3 b -
a The number in this table is listed for discussion but does not mean the order of stimuli.
b Action 3 was similar to Action 1 and 2 but not displayed in the training session. This trial was designed to explore children’s extending uses of the novel words.
3.1.5 Comprehension task
As shown in Table 3.4, the comprehension task consists of five trials. For both
conditions, one trial was administered before the conditioning training session (i.e., in
the pre-test session) and four trials after the training (post-test). Pre-test trials in the two
conditions were identical while there were some differences in post-test trials between
the two conditions. In the Specific condition, it was also examined whether children
could make the distinction between Action 1 and Action 2, which were mapped onto
different verbs. To put it more specifically, in the Specific Condition, Trial 2 and Trial 4
were designed to test whether children could distinguish the target action from others
they had never been shown in the training session. These two trials could thus serve for
the comparison with performance in the General Condition. On the other hand, Trial 3
and Trial 5 were designed to test whether children could make distinction between the
two actions, which were matched to different novel verbs in that Action 1 and Action 2
served as choices in each of the two trials. The reason that the materials in Trial 3 and
Trial 5 in the Specific Condition were not adopted for the General Condition is that the
two actions serving as choices in these two trials in the Specific Condition could be
referred to by the same label under the General Condition, and thus it would be
inappropriate to ask children which of the two actions this label referred to. The
characters in the video clips of the test session were different from that in the
demonstration of the training session. In addition, the characters in the video clips of
post-test trials (from Trial 2 to Trial 5) differed from one another. This difference in the
characters among trials was designed to avoid strategies that might be developed
because of the repetition of the same question. On the other hand, the characters in Trial
1 and Trial 2 were identical since they were designed for comparison between the
pre-test and the post-test.
Table 3.4: Trials of the comprehension task
General Specific
Test condition
Target ActionLabel Target Action Target Label
(1) Pre-test Action 1 Label 1 Action 1 Label 1 (2) Post-test Action 1 Label 1 Action 1 Label 1(3) Post-test Action 1 Label 1 Action 1a Label 1 (4) Post-test Action 2 Label 1 Action 2 Label 2
(5) Post-test Action 2 Label 1 Action 2 a Label 2
a In these two trials, participants should distinguish the target action from the other that was also labeled so as to make the correct choice.
3.1.6 Levels of analyses
To explore the effect of the conditioning training, three levels of analyses were
conducted: (1) comparison of the performance in the post-test session between two
conditions (general versus specific), (2) comparison between the performance on
pre-tests and post-tests for each condition, and (3) comparison between the performance
level and the chance level. Comparison of the performance in post-test between two
conditions would allow us to examine whether specificity facilitated word learning. For
each condition, the performance in pre-tests was employed as a baseline level for
comparison with the performance in post-test. In the General Condition, the comparison
would allow us to demonstrate whether a different action coming with the same label
confused children about the meaning of label, or whether the accumulated frequency of
the label and less variation among visual exemplars facilitated word learning. On the
other hand, the conditioning training in the Specific Condition was designed to explore
whether one-word-to-one-action mapping was facilitative or whether subtle semantic
distinctions confused children. Additionally, there were some differences in post-test
trials between the two conditions since the conditional training in the two conditions
provided children with labels encoded different information. Therefore, the comparison
between the performance level and the chance level would be helpful if the differences
in the test trials make it inappropriate to directly compare results from two conditions.