Experiment 1 preliminar evaluation. The first experiment was done at the beginning of annotation process. The objec- tive was to know the level of agreement between annotators without the annotation guide and without any training process. The results obtained in this experiment will be compared with the results obtained in the experiment 2. In this first experiment we have compared the annotation of four specific words with high frequency in the corpus: one verb (“decir”: to say), one adjective (“pri- mero”: first), one abstract noun (“vida”: life) and one concrete noun (“hombre”: man). All occurrences of these words in the corpus were annotated by two annotators. The results are shown in Table 1. The results show that the agreement between annotators at the beginning of the annotation process is very low. Indeed, in the annotation of the verb and the adjective, the agreement according to kappa coefficient (see later) is 0, that is, the agreement between annotators is the agreement expected by chance. On other hand, as is shown in [21], the level of ambiguity is different in each part of speech. Effectively, according to this preliminar experiment, adjectives are more ambiguous than verbs and nouns; and abstract nouns are more ambiguous than concrete nouns.
Experiment 1. Simple static allocation This section describes the results we obtained in Experiment 1. Figure 5 shows the throughput of the system during the experiment. After load level 30, we do not observe any growth in the system’s throughput because one or both of the tiers have reached their saturation points. Although the load level increases with time, the system is unable to serve all requests, and it either rejects or queues the remaining requests.
Experiment 1. Static resource allocation to Web application In this experiment, I established an experimental setup in which only one virtual machine (VM1) hosts the Web application. Figure 4.1 shows the experimental setup established for this experiment. The Ngnix based Web farm consists of only one virtual machine (VM1). VLBManager is only used to obtain the average response time by actively monitoring the Ngnix logs.
Experiment 1. In our first experiment, we manipulated the overt-case marking of the subject head.
Experiment 1
