!!!XML-structure for questions and answers

This document contains thoughts about xml-structure for questions for
qa-drill and also for the dialogue game. The suggestions are based on
two documents: Morphdrill and qaoutline. 

We should first think what kind of information we need for
qa-system. There is a general form for questions and answers that is
described in qaoutline-document. Here are couple of examples of proto-questions:

{{{
   Maid (SUBJ) TV-QPN (ADVL) 
   Goas IV-QPN (ADVL)
   TV-QPN go (SUBJ) OBJ (ADVL) 
   LEAT-QPN go (SUBJ) TV-PRTPTC OBJ (ADVL)
}}}

The grammatical form for basic questions is thus quite nicely
determined. But how do we select the elements to these sentences?
When you are learning a language, you should try to use the word
combinations that are familiar for the user and sentences from
real-life situations. New words or word combinations
may confuse the user. For example, if the user concentrates on
inflecting nouns, the questions could be quite fixed so that the meaning of
the sentence is clear and the learner may better concentrate to the task.
There are couple of things we can do:  

1. Form set that limit the set of words where the elements are picked
to the sentences.

This may not be enough, since it may be customary to
use a certain sentence for asking certain things. Some questions are
more idiomatic than others, and since this is a pedagocial program, we
should take those uses into account. 

In addition, the user can at the same time learn sentences that are not
akward if used in a conversation. For example, practising conditional
and potential requires different kind of setting thatn practising
e.g. present tense. However, more advanced student may have more fun with
random sentences.

2. Design the questions more detailly. 

Morphdrill-document contains examples of more detailled
questions. Another thing that has to be taken into account is the
expected answer. For example, the same question may be used for both
verb and noun inflection:
{{
Menittekö kouluun?
Ei, me __ __ kouluun. (emme menneet)
Ei, me __ uimaan. (menimme)
Ei, me menimme __ . (uimaan)
}}

How is this information presented? Do we have different sets of
questions for testing verbs and nouns or can we use the same set of
questions?

There are two kinds of grammatical information that has to be taken
into account in the program:

* Dependencies inside a sentence: agreement
* Answer is dependent on the forms that are used in the question.

So for example:
{{
M: Maid Pron+Pers+QPN+Nom bargat+Ind+Prs+QPN odne?
V (ACTIVITY Inf)
U: (Pron+Pers+QPN+Nom) V+Ind+Prs+QPN

M: Maid mun barggan odne?  
'stoahkat' (ACTIVITY Inf > Prs Sg2)
U: (Don) stoagat    
}}

* Agreement
** Subject-verb agreement inside the question (N+QPN=V+QPN)
** case and number inflection for adjectives and other elements inside an NP.
** Time adverbials selected according to the tense (not actually agreement)
** Other?

* Question-answer dependencies
** QPN = APN
** Preserve tense and mood in the main verb
** Other?

Then there should be possibility to select the properties of sentences
according to the settings. Perhaps we should decide first what kind of
settings we have, and tag the questions accordingly? Lene has listed
the following in the Morphdrill-docu:

* Adjectives: attributive or predicative drill
* Nouns: which case
* Verbs: 
** indicative:
*** present tense or past tense
** affirmative or negative
** conditional (only present tense)
** potential (only present tense)
** imperative
* Numerals: 
** as attribute to nouns, in all cases. Also plural numbers
** collective numerals, in all cases

These sets all require a bit different kind of question setting.

If we think about the general question type, there are the following
variables, which have a grammatical status, i.e. they agree with each
other etc. So they have to be explicitely marked in the sentence.

{{{
SUBJ, MAINV, LEAT, PRTPTC?, NEG, ADJ, other?
}}}

In addition to these, any element may be a variable. The idea of the
xml-structure that I suggest is that sentences contain this
kind of variables, which have the grammatical information specified.
For example, we may have questions where everything else is given, but the answer is
selected from a set. I have marked the variable as "ANSWER" here, but
it could have any name:

{{{
M: Gosa don manat dál?	(Where do you go?)
'Suopma'  (PLACE Sg Nom > Sg Ill)			
S: Supmii

<q>
  <question>
    <text>Gosa don manat dál</text>
  </question>
  <answer> 
    <text>Mun manan ANSWER</text>
    <element pos="N">
      <grammar tag="N+Sg+Ill"/>
      <sem class="PLACE"/>
     </element>
  </answer>
</q>
}}}

The information for the word 'Suopma', Sg+Nom is assumed in the
program, so that the suggested answer is in N+Sg+Nom or for
verbs, V+Inf. (or it can be explicitely specified if needed)

Information for the variables in the sentences are given inside
<element> -tags. The order of the elements should be the same as the
order of the variables inside the sentence. (Or the elements should be
marked otherwise, we have to see if the order is too error-prone). Let
us generalize the example above.

{{{
<q>
  <question>
    <text>Gosa SUBJ MAINV dál</text>
    <element pos="Pron">
      <grammar tag="Pron+Pers+Person-Number+Nom"/>
     </element>
    <element pos="V">
      <grammar tag="V+Ind+Prs+Person-Number"/>
      <lemma>manat</lemma>
     </element>
  </question>
  <answer> 
    <text>(SUBJ) MAINV N-ILL</text>
    <element/>  <!-- inherited from the question -->
    <element/>  <!-- inherited from the question -->
    <element pos="N">
      <grammar tag="N+Sg+Ill"/>
      <sem class="PLACE"/>
     </element>
  </answer>
</q>
}}}

Here the variables SUBJ and MAINV in the answer depend fully on the
SUBJ and MAINV in the question, thus they do not require any other
information. The information for the ANSWER-variable is given. This
question could be used also in contexts:


{{{
Gosa don manat dál?
Mun __ Supmii.
}}}

Now how do we present this information that this question can be used
in different drills? The question could be tagged according to which
type of drill they can be used in.

Thus, I think that we do not have try to find the most general
format for the questions. Instead we should write the questions quite
explicitely so that they have a natural feeling. This means use of
small words and word combinations that are familiar to the user. In
addition, we need quite specified questions in the dialogue game, so
there should be a way to represent them. However, some generality can be
achieved by using variables, when part of the grammatical information
e.g. subject-verb agreement is handled by the program.