summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorolivia.buzek <olivia.buzek@ec762483-ff6d-05da-a07a-a48fb63a330f>2010-10-18 02:35:41 +0000
committerolivia.buzek <olivia.buzek@ec762483-ff6d-05da-a07a-a48fb63a330f>2010-10-18 02:35:41 +0000
commit0c2514868f58bbfe422aa275e2905182cf2f57eb (patch)
treeb462fe337345192ffa76c4823d42eb83d6103389
parent41ace11343affe46d47b56583da3adbcd77b9227 (diff)
Updating report.
git-svn-id: https://ws10smt.googlecode.com/svn/trunk@680 ec762483-ff6d-05da-a07a-a48fb63a330f
-rw-r--r--report/backoff/backoff.tex2
1 files changed, 0 insertions, 2 deletions
diff --git a/report/backoff/backoff.tex b/report/backoff/backoff.tex
index ca786ffd..a47e6296 100644
--- a/report/backoff/backoff.tex
+++ b/report/backoff/backoff.tex
@@ -8,8 +8,6 @@ Multi-category grammars can better represent complex structure and reordering us
This portion of the project seeks to design a backoff grammar which can build on a multi-category grammar such as the one induced by the Pitman-Yor process described elsewhere in these proceedings. The backoff grammar adds the flexibility of a single category grammar without sacrificing reordering.
-In the Motivation section, the reason for designing such a grammar is explored in more detail. In the Naive Backoff section is explained the original implementation of the backoff grammar, and in the Hierarchical Backoff section, the last tested iteration of the grammar is explained. Results of both grammars on the BTEC corpus versus a single-category Hiero baseline are explained in the Results and Analysis section. Finally, we will conclude with this part's role in the larger workshop project as well as describe possible future work in the Conclusions section.
-
\subsection{Naive Backoff}
As was previously mentioned, given the sparsity of the training data, no phrase will have all of its linguistically possible contexts accounted for in the grammar. As an example, consider the production in (\ref{eq:whenwill}):