El dt 05 de 07 de 2011 a les 17:42 +0200, en/na Keld Jørn Simonsen va
escriure:
> On Tue, Jul 05, 2011 at 03:24:09PM +0000, Francis Tyers wrote:
> > El dt 05 de 07 de 2011 a les 16:49 +0200, en/na Mikel Forcada va
> > escriure:
> > > Hi there,
> > > > I would like to attach attributes to lemmas. Only a few but maybe there
> > > > could be more, so a kind of introducing an attribute name would be nice,
> > > > instead of having a predefined set of attribute names..
> > > Lemmas as such aren't represented as such in Apertium dictionaries. They 
> > > are part of the lexical forms (one could say that the lemma is the 
> > > material from the beginning of the lexical form up to where the first 
> > > part-of-speech tag appears. For instance, for surface form "thought" an 
> > > English dictionary would derive the lexical forms "thought<n><sg>" and 
> > > "think<vblex>...". The lemmas would then be "thought" and "think". There 
> > > is a attribute lm="...." in some entries, but it is optional.
> > > > I believe there are already lemma attributes, such as the word class of
> > > > the lemma: noun, verb, adjective, adverb etc.
> > > Not for lemmas. Lemma information is encoded either as the content of 
> > > the element (see above). Part of speech as well as other morphological 
> > > information is encoded as attributes of the <s> (symbol element).
> > > > what I have in mind is to attach data from wordnet, such as sense,
> > > > hypernym, hyponum, holonym, meromnym, and also combine it with the
> > > > Swedish SALDO attributes of father and mother relations.
> > > >
> > > > The idea is then to choose a sense of a homonym based on the shortest
> > > > distance to maybe the previous and following five words.
> > 
> > Which language pair(s) are you working with ? Is it really necessary ? 
> 
> sv-da. I cannot get further with my work without such features - at least in a
> rudymentary form.
> 
> 
> I think it would be fun. Anyway there are of cause problems with
> homonyms in swedish and danish that could be better solved with more
> intelligent selection machinery.
> 
> I have about 40000 new swedish words that I have used quite some time on
> and they should not damage the already existing work.

In any case if you do mass addition you're probably going to damage what
is there. 

> > > > a lemma may have more than one sense. Eg 'nut' may mean several things
> > > > such as the offspring of a plant, nuts and bolts, and testicles.
> > > >
> > > > Is this easy to do? How do I do it?
> > > I think the attribute lm="...." could be stretched a bit to have any 
> > > value, which could be used to identify the lemma in another structure 
> > > which could contain all of these (for instance, giving an XPath to 
> > > another XML file containing all the desired information).
> > > 
> > > Perhaps it would be better to have some kind of new general purpose 
> > > attribute that could be used to attach *standoff* information of this 
> > > kind to any entry <e>.
> > 
> > I think that might be nice... also for, for example verb valency or
> > other features that we don't necessarily want to represent with tags.
> 
> But maybe we can just do it with tags. Is it possible to add arbitrary
> tags?

Yes it is possible to add arbitrary tags.

> > > Fran is working on lexical selection and I'm sure his opinion would be 
> > > interesting to read!
> > 
> > Could also use the attribute 'c' for comment. 
> 
> I think it would be misleading to call it a comment. Maybe <a> -
> attribute?

Sure, but to start with you can use c="" and later if you have good
results we can add a="" to the DTD. 

> > 
> > in the instance that a word has the same lemma/pos/gender and different
> > paradigms/declensions, I use a pseudo lemma, for example from Russian:
> > 
> >     <e lm="????????"><i>????????</i><par n="????????__n_m_nn"/></e>
> >     <e lm="????????"><p><l>????????</l><r>????????:1</r></p><par
> > n="??????__n_m_aa"/></e>
> 
> I would then still need a way to disambiguate and chose the right one.

Yep, but it's a tiny part of the problem. A more sensible place to start
would be adding compounding support, like nn-nb and af-nl have.

Fran


------------------------------------------------------------------------------
All of the data generated in your IT infrastructure is seriously valuable.
Why? It contains a definitive record of application performance, security 
threats, fraudulent activity, and more. Splunk takes this data and makes 
sense of it. IT sense. And common sense.
http://p.sf.net/sfu/splunk-d2d-c2
_______________________________________________
Apertium-stuff mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/apertium-stuff

Reply via email to