apertium/apertium-anaphora

First Person Pronouns

Closed this issue · 8 comments

khannatanmai commented

While evaluating, the current method is changing my/our to his/their.
We need a method to only do Anaphora Resolution for third person pronouns as we decided earlier, because first and second are usually resolved in the real world.

However, the tagger isn't adding the person in the possessive determiners.
Is this a tagger issue to be fixed?
Or should we make a method of providing exceptions on which we don't want to apply the transfer rule?
Should this be done in my module or can this be written in t1x somehow?

@unhammer @ftyers

unhammer commented
khannatanmai commented

@unhammer Yes look at the examples in the original comparison report I made, with the basic anaphora system:
https://github.com/khannatanmai/apertium-anaphora/blob/master/Evaluation/Comparison_WordChanges.txt

unhammer commented
khannatanmai commented

Original Spanish: ¿Por qué no se han mejorado las escaleras desde mi accidente?
English Translation: Why they have not improved the stairs from my accident?

Original Spanish: Si su decisión es que no podemos explicar nuestro voto, la acataré, pero no sin reservas.
English Translation: If his decision is that we can not explain our vote, will observe it, but no wholehearted.

Biltrans Output for mi and nuestro: (First Person)
^mío<det><pos><mf><sg>/my<det><pos><mf><sg>$
^nuestro<det><pos><m><sg>/our<det><pos><m><sg>$

Biltrans Output for su: (Third Person)
^suyo<det><pos><mf><sg>/his<det><pos><mf><sg>$

So they aren't the same as the third person pronouns, but the Biltrans output doesn't describe the number, i.e. doesn't distinguish between mi and su.

So we can add exceptions for some words or we can somehow modify the biltrans to add number to the analysis and then detect that. @unhammer @ftyers

khannatanmai commented

@unhammer @ftyers should I add the capability of adding exceptions to the rule in the ref file or can that be done in the transfer t1x file as well?

For eg., match "det.pos" except when it's "our" or "my".

Or, should I assume the tagger should tell me the person?

unhammer commented

Can the error propagate?

If it's a completely local error, then it's easy to solve in the transfer rule – e.g. if you have a transfer rule that checks the ref clip and would change "my" to "his", I'd be fine with just having the transfer rule check for that possibility first (transfer files already need to check such things other places). But if this my→his error can lead to later words also getting the wrong number, then the anaphora module needs to rule them out.

khannatanmai commented

Nope, it won't propagate. I'll modify the transfer rule and see the results.

khannatanmai commented

New Macro (Modified):
<def-macro n="ref" npar="1"> <!-- anaphora -->
<choose>
<when>
<test><not> <!-- ignore if it's first person, i.e. my or our -->
<or>
<equal><clip pos="1" side="tl" part="lem"/><lit v="my"/></equal>
<equal><clip pos="1" side="tl" part="lem"/><lit v="our"/></equal>
</or>
</not></test>
<choose>
EARLIER TRANSFER RULE
</choose>
</when>
</choose>
</def-macro>

Input Examples:

¿Por qué no se han mejorado las escaleras desde mi accidente?
Si su decisión es que no podemos explicar nuestro voto, la acataré, pero no sin reservas.

Output:

Why they have not improved the stairs from my accident?
If their decision is that we can not explain our vote, will observe it, but no wholehearted.

So it doesn't touch first person pronouns now. If there's a better solution, we can discuss it. @unhammer @ftyers