Experiment 2 minimum level of agreement. The second experiment was developed at the end of the annotation process, once the annotation guide was developed and annotators have experience in semantic annotation. The objective of this second experiment was to know the minimum level of agreement between annotators; that is, to know the agreement achieved in complex words: words with high level of ambiguity. Word Part of Speech Senses Frequency Historia (history) Noun 9 33 Xxxxxxx (race) Noun 11 27 Ley (law) Noun 6 22 Tierra (earth) Noun 11 18 Papel (paper) Noun 7 18 Ganar (to win) Verb 8 33 Suponer (to suppose) Verb 10 33 Pensar (to think) Verb 8 38 Trabajar (to work) Verb 8 33 Jugar (to play) Verb 7 26 Abierto (open) Adjective 28 17 Table 2: Words and frequency Word Average Agreement by chance (P(e)) Kappa Historia (history) 45% 0.23 k = .28 Carrera (race) 89% 0.43 k = .8 Ley (law) 75% 0.266 k = .66 Tierra (earth) 56% 0.17 k = .46 Papel (paper) 78% 0.42 k = .61 AVERAGE 68% - k = .56 Table 3: Minimum inter-annotation agreement in nouns. Experiment 2. The inter-annotation agreement has been calculated following an evaluation method of “lexical sample” corpus. The annotation of 13 ambiguous words (5 nouns, 5 verbs and 3 adjectives -see table 2-) has been compared. The results are shown in table 3, 4 and 5. The average of agreement between the three part of speech are 68%. Similar to the first experiment, the part of speech with less agreement is the adjective (63%). However, verbs are the part of speech with higher level of agreement (72%). Together with the agreement average, we have calculate the kappa agreement, following [17] 6. This measure calculates and removes from the agreement rate the amount of agreement that is expected by chance. Therefore, the result are more exact than a simple agreement average [2] [22] [6]. Kappa measure is calculated according to the formula: 6There are two method for calculate kappa [7]: one in [6] and the other one in [17]. We have used both formula. The results are quite similar, so we show here only the results obtained with Xxxxxx and Xxxxxxxxx formula [17]. Word Average Agreement by chance (P(e)) Kappa Ganar (to win) 87% 0.66 k = .61 Suponer (to suppose) 28% 0.25 k = .15 Pensar (to think) 89% 0.45 k = .8 Trabajar (to work) 71% 0.54 k = .36 Jugar (to play) 76% 0.3 k = .65 AVERAGE 72% - k = .51 Table 4: Minimum inter-annotation agreement in verbs. Experiment 2. Word Average Agreement by chance (P(e)) Kappa Nacional (national) 62% 0.45 k...
Experiment 2. In Experiment 2, we changed the comprehension question from “what happened first?” to “what happened last?”. This modification changed the sentence types for which the correct response was facilitated by the main clause. Hence, in Experi- ment 2 the correct answer was located in the main clause for sentences with con- nective after, instead of the connective before as in Experiment 1. If comprehension is facilitated by the effect of salience of information in the main clause rather than the type of connective, one would expect that the sentence type for which most comprehension errors are made should shift from sentences with the sentence-initial connective after in Experiment 1 (see Table 4.1a), to sentences with the sentence-initial connective before in Experiment 2 (see Table 4.1b). If comprehension is facilitated by a recency effect, one would expect that perfor- xxxxx would be better for sentences in which the correct answer corresponds to the most recently read event (before-medial and after-initial sentences).
Experiment 2. Adaptive resource allocation to Web application In this experiment, I used our proposed system to prevent response time increases and rejec- tion of requests by the Web server. Figure 4.6 shows the experimental setup I established for this experiment. Initially, the Web farm is deployed with one virtual machine (VM1), while VM2 and VM3 are cached to the physical system using EUCALYPTUS. In this experiment, I try to satisfy a Service Level Agreement (SLA) that enforces a two-second maximum average response time requirement for the sample Web application on an undefined load level. I use VLBManager to monitor the Ngnix logs and detect the violations of the SLA. I use VLB- Coordinator to adaptively invoke additional virtual machines as required to satisfy the SLA. The same workload from Experiments 1 is used in this experiment, as explained in Figure 4.2.
Experiment 2. Children’s Knowledge about the Number Agreement
Experiment 2. TJ If you aim the remote control at the floor a little bit in front of the robot and hold down a button, what does it do? TJ If you aim the remote control directly at the robot’s eyes and hold down a button, what does TJ do? From how far away can you still affect the robot? Experiment 3 Go to your robot teachers to reprogram your robot. Plug the 6-wire cable into TJ’s head. Switch the robot to Download mode, and turn the power on. Watch the lights on the communications board. One of them should flash every time you push the RESET button. If this does not happen, then the cable is plugged in backwards on the robot. Pull it out, turn it around and plug it in, then try again. When the robot is plugged in correctly, type the commands listed below: mscc11 code\race9 Push <ENTER> and follow the directions on the screen. What happens to the lights on the communications board while the TJ is being programmed? What do these lights mean? (HINT: Ask your robot teachers.)
Experiment 2
