pagelj / em-for-inducing-sem-classes

Using EM to induce verb classes based on argument slots

Using EM for inducing verb and noun classes in an unsupervised manner

Using Estimation Maximization to induce verb and noun classes based on extracted verb-noun-pairs, following Rooth et al. (1999)

Contributors

• pagelj
• thefth

Variables

Variable Name	Content
p_c	Probability for each class
p_v_c	Probability for a verb being in a class
p_n_c	Probability for a noun being in a class
pairs_classes	Joint probability for a verb and noun being in a class
v_c	Hard verb classes
n_c	Hard noun classes
verb_noun_rank	Ranked nouns for a intransitive verb
verb_noun_value	Accumulated probability for all nouns belonging to an intransitive verb

Data

• gold_deps.txt: Reduced data
• all_pairs: All pairs, not preprocessed
• new_pairs: All pairs, preprocessed

Abbreviations

Abbr.	Explanation
s	intransitive verb
so	transitive verb
nsubj	nominal subject
csubj	clausal subject
nsubjpass	nominal subject in passive construction
dobj	direct object
iobj	indirect object

Verb-Noun pairs e.g.

Verb	Co-occurring noun
testify_s_nsubj	list
make_so_nsubj	teacher
lead_s_csubj	deciding
give_so_csubj	adding
leave_s_nsubjpass	vegetable
make_so_nsubjpass	project
reach_so_dobj	leader
become_so_iobj	member

Example class

Class 5: Change on a Scale - Numbers

Verbs	Prob	Nouns	Prob
pay_so_dobj	0.0674	price	0.0389
rise_s_nsubj	0.0623	number	0.0304
fell_s_nsubj	0.0459	attention	0.0300
draw_so_dobj	0.0349	profit	0.0257
fall_s_nsubj	0.0203	rate	0.0257
rise_so_nsubj	0.0193	share	0.0226
bring_so_dobj	0.0191	cost	0.0210
increase_so_dobj	0.0185	sale	0.0169
cut_so_dobj	0.0179	earnings	0.0152
increase_s_nsubj	0.0168	level	0.0107

Example intr. verb noun ranks

begin_s_nsubj

Noun	Score
work	11.3236
process	8.4428
which	6.3095
trial	6.2411
discussion	5.9485
phase	5.8081
career	5.6166
shipment	5.5813
series	5.4629
programme	5.2699

ToDo

• Achieving tracable computation of transitive verbs with ranked nouns

References

Mats Rooth, Stefan Riezler, Detlef Prescher, Glenn Carroll, and Franz Beil. 1999. Inducing a semantically annotated lexicon via EM-based clustering. In Proceedings of the 37th Annual Meeting of the Association for Computational Linguistics on Computational Linguistics. Association for Computational Linguistics, Stroudsburg, PA, USA, ACL ’99, pages 104–111. http://www.aclweb.org/anthology/P99-1014.