Morning to all.
For my PhD, I am developing an MT system which will test the
relationship between the quantity of semantics, and the level of
translation performance the system obtains. To achieve this, I am
building a functional (but very small coverage) MT system of my own
which can perform the same translation with different quantities of
semantics so the results may be compared. If you're interested, feel
free to read my web site (the URL is below), but the reason for
mailing the MT group is to find answers to the following question:
What "surface roles" should I identify during parsing? The roles
will then be used as a start for deeper semantic analysis (i.e.
subjects tend to become agents, and object 1's become patients).
Subject:
*I* like Mary; *Mary* me pla�t.
Object 1:
I ate *chicken*; j'ai mang� du *poulet*
Object 2 (& 3 & 4):
I gave John *the book*; j'ai donn� le livre � *John*
Compound Head:
The park *keeper*; le *gardien* du parc
Non-Compound Head:
*apple* pie; tourte aux *pommes*
PP:
I went *to the school*; je suis all� *� l'�cole*
Adverb:
I drove *quickly*; j'ai roul� *vite*
Modifying adjectives etc.:
I like *big* smiles; j'aime les *gros* sourires
the man *identified by the guard* disappeared; l'homme *identifi� par
le garde* a disparu
What other such roles do you identify in your MT system?
I'll post the results when I get them.
David
HTTP://www.ccl.umist.ac.uk/ra/dmowatt/
--
For MT-List info, see http://www.eamt.org/mt-list.html
