... task1.1
A task in which the vector components are discrete values rather than continuous values. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... fnTBL2.1
fnTBL is to be pronounced as funtible < $ f\Lambda nt:bl $> 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... stop2.2
$ \theta $ is a threshold that needs be set in advance; it can be $ 0 $, in which case the algorithm learns until there are no more rules to be learned. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...3.1
In general, there should be $ \left\vert F\right\vert +2\cdot \left\vert C\right\vert $ features, where $ F $ is the set of features associated with the sample and $ C $ is the set of classifications. For POS tagging $ F=\left\{ word\right\} $ and $ C=\left\{ pos\right\} $. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... word3.2
The gold-standard POS, to be more precise. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...3.3
The presented class is the Wordnet label of one of the senses of the word use. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... rule4.1
A rule whose output classification is part of the decision - such as the example above - the classified class - NN - is part of the predicate. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...5.1
The argument to the -t parameter of the most_likely_tag.prl script is capital o (O) and not the number 0. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... constraints5.2
One might want to not impose constraints on low occurrence words, because the statistics on them might not be reliable. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... distribution5.3
The file is presented here in double column, to save some space, but it's single column in the actual file. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... types5.4
We are describing here the basic predicate types - the rules' predicates are made out of a conjunction of these predicate types. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... time5.5
One might want to do this, as the distribution on the first part of the training data might be different than the distribution on the second part. However, one should exclude the words that appear only on the second part, as it will create conditions similar to the testing time situation, i.e. do not use the true tag for the unknown words, but rather the guessed tag. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... memory5.6
Everything is relative nowadays, though. It is debatable whether is better to invest time in making the program run in smaller amounts of memory or to wait until computers with a large amount of memory become standard. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... properties5.7
But it should be mentioned that the fact they applied only one on the data also says that they did not apply on the rest of the samples, which means that they should not make a big negative impact on the test data, as they should not apply too often. However, if the training and test data are not drawn from the same distribution, they can hurt the performance. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... classifications5.8
This is true if the instances are not part of the same document, where the ``one sense per discourse'' [11]; as the task was presented in Senseval2 [2], the samples are independent. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... necessary5.9
This character was selected because it still keeps the sample line readable, but you can replace it with any character you like. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... time5.10
Actually, we observed a significant decrease in running time when using this feature; its use is highly recommended in cases such as this one. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...strike%2:35:03::|strike_a_chord%2:31:00::5.11
The values presented here represent the wordnet 1.7 tags [3]associated with the senses. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... count5.12
In this case defined as would convert the value to correct, if applied, regardless of the actual value of the classification - usually, rules are not considered to apply if the classification is already correct. 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... tiesA.1
Tie = a case where 2 rules have the same score 
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.