Differences

This shows you the differences between two versions of the page.

--- courses:rg:2011-report-parser [2012/09/26 13:46]
ufal vytvořeno
+++ courses:rg:2011-report-parser [2012/09/27 10:39]
ufal
@@ Line 3: / Line 3: @@
 written by Stephen Tratz and Eduard Hovy (Information Sciences Institute, University of Southern Carolina)
-spoken by Martin Popel
+presented by Martin Popel
 reported by Michal Novák
 ===== Introduction =====
+The paper describes a high-quality conversion of Penn Treebank to dependency trees. The authors introduce an improved labeled dependency scheme based on the Stanford's one. In addition, they extend the non-directional easy-first first algorithm of Goldberg and Elhadad to support non-projective trees by adding "move" actions inspired by Nivre's swap-based reordering for shift-reduce parsing. Their parser is capable of producing shallow semantic annotations for prepositions, possesives and noun compounds.
 ===== Notes =====
+==== Dependency conversion structure ====
+  * in general, there are (at least) 3 possible types of dependency labels:
+    * unlabeled - is it really a set of labels?
+    * coarse labels of the CoNLL tasks
+        * 10-20 labels
+        * for example NMOD is always under a noun - it's an easy task and the result is not quite useful
+    * their scheme is based on the Stanford's dependency labels
+==== Conversion process ====
+  * converting phrase trees of Penn Treebank to dependency ones
+  * it consists of 3 steps:
+      - add structure to flat NPs
+      - constituent-to-dependency converter with some head-finding rule modifications
+          * a list of rules in Figure 2 is hardly understandable without reading a paper their conversion method is related to
+          * they reduced the number of generic "dep/DEP" relation
+              * Stanford tags are hierarchical and "dep/DEP" is the top-most one
+          * 1.3% of arcs are non-projective (out of 8.1% of all non-projective arcs) because of the following conversion (agreement can be a motivation for this, i.e. in Czech):
+            {{:courses:rg:dependency-conversion.png|}}
+==== Parser ====
+  * we illustrated a step of the parser:
+{{:courses:rg:ndef_parsing.png|}}
+  * we compared time complexity of this system with other commonly used ones

[ Back to the navigation ] [ Back to the content ]

Institute of Formal and Applied Linguistics Wiki

Differences