short description of your system - little urgent

[Lucia Specia]

Dear all,

We are preparing the description paper for Task 1. Could you please send us a short description of your system(s)? 1-2 paragraphs (max 1/2 page) would be great. Please specify the method, whether you used the trial data for training, any external resource you used and anything else you judge particularly relevant. If you are not submitting a paper to SemEval describing your system and already have a publication for it, please send us the reference.

If you can send this by April 12 it will be great!

Best,

Lucia

[Cyril Grouin]

Hi Lucia,

Thank you for the results provided in the Excel file. Could it be possible to compute the statistical significance between all participants, especially between the three first one? Indeed, results are very close.

Best regards,
Cyril (on behalf of "annlor" team).

[marilisa amoia]

Dear Lucia,

here a more detailed description of our system  SB:mmsystem

best,

marilisa

%% latex start

Our approach builds on the baseline definition of simplicity based on word frequency but attempt at defining a more  

linguistically motivated notion of simplicity based on lexical semantics considerations. Therefore, in order to operationalize the notion

 of simplicity in our system we adopt different strategies depending on the syntactic complexity of the lexical form that forms the substitute.

\begin{itemize}

\item in the trivial case of one-word substitutes or common collocations we use the frequency associated by Wordnet to the lexical form as 

          metrics to rank the substitutes. For instance, the lexical item \emph{intelligent} is ranked lower that \emph{clever} as it has a lower 

          frequency in the language (as defined in Wordnet). 

\item In the case of multi-words substitutes of bigger syntactic complexity, we apply so-called "relevance" rules 

          that apply (de)compositional semantic criteria and attempt to identify a unique content word in the substitute that might better approximate the whole  lexical form. Thus we assign to the whole lexical form the 

         frequency associated to the most relevant content word and use it for ranking the whole substitute.

         For instance, relevance rules assign to multi-word substitutes such as \emph{most able} or \emph{not able} the same frequency associated

         to the content word \emph{able}.    

\end{itemize}

We implemented the following steps:

\begin{itemize}

\item A syntactic parser was used to assign Part-Of-Speech to each word of the context and of the substitutes. We used the Stanford Parser.

\item We performed WordSense Tagging so to assign a WordNet  sense (WordNet 3.1) to the target words and the substitutes.

          We used the Ted Pedersen WordNet-Similarity Package for word sense disambiguation and the JWNL package for information retrieval form WordNet.

\item We defined hand-written rules and apply them to decompose the meaning of a complex structure and identify the most 

          relevant word conveying the semantics of the whole.

\end{itemize}

%% end latex