Problem Set (3b)

When we first posted our final Problem Set (3b) a little more than a week ago, there were three mistakes: (a) we had used the name ¡®train.mrg¡¯ in SVN, instead of ¡®wsj.mrg¡¯, as used in the exercise text; (b) the lexeme count in the grammar estimated from this file (i.e. the training part of the treebank) is around 44,000 (rather than 50,000); and, most importantly, (c) our skeleton code in ¡®chart.lsp¡¯ contained some tabulator characters (where there should rather be eight spaces).  We encourage everyone to ¡®svn update¡¯ their INF4820 directory, to fix the file naming and have those tabulators removed from ¡®chart.lsp¡¯.

In the lecture tomorrow, we will fill in the remaining theory for one-best Viterbi chart parsing with PCFGs and evaluation of statistical parsers.  Also, we will consolidate our emerging understanding of syntactic analysis for natural language with a quiz towards the end of the lecture.  The deadline for submitting our final problem set is Wednesday next week, November 19.