Commit 4c89efdf authored by Bruno Guillaume's avatar Bruno Guillaume

abstract Djamé

parent c06d14dd
......@@ -105,10 +105,25 @@
\begin{document}
\maketitle
\begin{abstract}
The 2.0 release of the Universal Dependency treebanks show the effectiveness of the UD scheme
to cope with very diverse languages. The next move is to get more of syntactic analysis,
and the ``enhanced dependencies'' sketched at in the UD 2.0 guidelines is an attempt in that direction. In this work
we propose to enrich the enhanced dependencies along two axis: extending the cases of recovered arguments of non finite verbs, and neutralizing syntactic alternations. In order to test its feasibility, we implemented the proposal for two French UD treebanks. Evaluation on a 200 reference sample shows that deep syntactic graphs can be obtained with good accuracy. Further, in such graphs, the number of additional argumental edges is substantial ($17\%$ of arguments of verbs were not in the input surface tree), and a significant amount benefit from syntactic alternation normalization (between $10$ and $14\%$ of the argumental edges, depending on the corpus).
The 2.0 release of the Universal Dependency treebanks demonstrates the
effectiveness of the UD scheme to cope with very diverse languages.
The next \draftreplace{move}{step} \draftreplace{is}{would be} to
get more of syntactic analysis\draftnote{j'aime pas du tout to get
more, un truc de ouf -ds}, and the ``enhanced dependencies'' sketched \draftremove{at} in the UD
2.0 guidelines is a \draftadd{promising}, attempt in that direction. In this work we
\draftadd{propose to} go further and enrich the enhanced dependency scheme along two axis:
extending the cases of recovered arguments of non finite verbs, and
neutralizing syntactic alternations. Doing so leads to more stable
analysis with respect to surface syntax variations, and
thus, entails more useful representations, deep syntactic graphs, for downstream
processing.
We implemented this proposal in two UD treebanks of French. An
evaluation \draftreplace{on a 200 reference sample}{carried out on a representative gold
standard} shows
that deep syntactic graphs can be obtained \draftadd{from surface
syntax annotations} with a high accuracy and that additional
argumental edges account now for 17\% of all verb
arguments, thus confirming the interest of our proposal.
\end{abstract}
\input{intro}
......@@ -132,7 +147,7 @@ compatible with all choices already made by the current specifications
\cite{nivre2016universal,schuster:2016:lrec}. Moreover, our de-facto adhesion
to the {\sc ConLL-U} representation format allows
for a straight-forward use by current data-driven graph
parsers. We leave this promising path of study to further work.
parsers. We leave this promising path of study to further work.
%% FOR FINAL VERSION \section*{Acknowledgments}
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment