[Parley-devel] Fwd: Features in next version of Parley

[Andreas Xavier]

Hello Ansa,

Thanks for your well considered answers.  

Inge has responded to most of your email, I am only going to address the 
"automatically generated" content, because I think the idea has been expanded 
considerably from my intent, because of my poor articulation
of the two underlying ideas.  And you probably didn't even see the 
initial explanations.

>> The idea is that the sentences will give you the structure of the language
>> rather than individual words. You will learn the vocabulary anyway since
>> many of the sentences use the same word in different contexts.
>>
>> It's an interesting idea and I am looking into it if we can incorporate it
>> into Parley at some point in the future.
>
>I always copy the whole text of the lessons from textbooks/courses/phrasebooks
>into parley, separate it into items and use it for practice. If it is
>possible, I use the sound recordings that come directly with the
>original materials - I am an aural learner.
>
>I am not sure if Parley could really do more for me in this respect,
>because automatically generated sentences would have to be equipped
>with automatically generated sound, and it is unlikely that this
>automatically generated sound would be satisfactory. In English, for
>example, word stress changes depending on the sentence context. This
>is important, because learners actually remember the "melody" of words
>and phrases before they fully learn them. (I read just today that it
>was shown already in the 1960's or 1970's that learners could
>correctly answer questions about the number of syllables and accent
>patterns of vocabulary items, even though they were not able to recall
>them correctly.)
>
>My feeling is that users should be encouraged to fill their
>collections with sentences that occur in whatever materials they are
>working with. They also should be encouraged to use definitions and
>synonyms in the target language instead of translation equivalents in
>their mother tongue (unless they intend to be translating between the
>two languages a lot, in which case it makes more sense to strengthen
>the associations between the two languages). All this provides natural
>structure and context.
>

The two ideas, both revolve around generalizing inflected content.  
I will explain what Parley currently does and then explain the two ideas.  

Parley currently handles the following types of inflection: comparisons, conjugation and 
declension.  It does this with a hard-coded set of productions.

For comparisons, Parley handles the comparative and superlative by asking the lesson creator
to fill in a box.  It does not handle the positive if it is different from the root.  It does not handle
the elative.  It does not handle declension of adjectives.  

For conjugation, Parley uses a set of pronouns and plurality, with some options all hard-coded in the C++ code.  
Adding honorifics, formality, proximity etc. would require C++ changes and add complexity to the UI.
Tenses can be arbitrary, just click new and add a tense.  Filling in conjugations/declensions
requires the user choosing a tense and filling in a value for each pronoun.

Reading through the wishlists I saw the following issues:
1. Students want a larger set of pronouns, specifically honorifics and formality.
2. Students want arbitrary gender names, if their language doesn't name/have male/female/neuter genders.
3. Students want declined adjectives and articles.
4. Students want articles other than indefinite/definite.
5. Students want conjugation of "non-personal tenses" in conjugations.
6. Students want to use inflected forms in example sentences.
7. One user deleted pronouns and then created a new word type which did not work correctly
    , because pronouns are a special hard coded type
8. Programmers want fewer special cases.
9. Students want to generalize regular conjugations.

I called the solution to all but the last wish constructions.

Constructions are the collection items identified with a tuple from a set of sets of words.  The 
lesson creator still enters each item in a box individually.  There is no labor saving.  Because the lists
are arbitrary, there are no limits on the the number/type of pronoun/gender/plurality.  All sets of sets
that the lesson creator can imagine are valid.  The UI is almost unchanged.  The UI complexity is 
unchanged.

For example, given the lists [first person, second person, third person], [single, plural],[to be, to have]
you would get boxes to enter the following items (exactly the same as current conjugation):
(first person, single, to be) -> I am
(, , ) -> you are
(, , ) -> he/she is
(, , ) -> we are
(, , ) -> you are
(, , ) -> they are
(, , to have) -> I have
(, , ) -> you have
(, , ) -> he/she has
(, , ) -> we have
(, , ) -> you have
(, , ) -> they have

But if the lesson creator picked these two lists [single, plural], [mice, rice, goose, moose, foot]
, then the boxes to be filled in would be.
(single , mice) -> mouse
( plural, mice) -> mice
(single, rice) -> a grain of rice
(plural, rice) -> rice
(single, goose) -> goose 
(plural, goose) -> geese
(single, moose) -> moose
(plural, moose) -> moose
(single, foot) -> foot
(plural, foot) -> feet

The whole set can be trained with the inflection game.  Individual items can be trained with
multiple choice/flash cards.  

All of the special code for pronouns, articles, verbs, nouns, comparison forms, gender gone 
and replaced with generic code to handle sets of words and projections.  Parley can keep
the same grammar creation dialog, but you can add any word to gender.  Furthermore,
any  set of words in word types is equally valid as a set for the inflection tool.

The only change to the editor UI is to add to the inflection tool the option to select a list of word types.
Once one/more word types is selected there will be one box per tuple for an answer.

The second suggestion is to solve the regular conjugation problem.  
The solution is to facilitate some form of regular expression use to generate constructed
answers.  It provides that added bonus of generalizing all constructions not just conjugations.

I emphasize that this part of the UI would be considered an advanced option, disabled by 
default in the editor.  It is completely transparent in the training portion of Parley.  
With some examples I think that the simple case of prefixes/suffixes will be copy pastable
by many users. Perhaps simple cases could also have their own UI item similar to conjugation currently.

First I'll provide a motivating example.  Then I'll present the solution.

The motivating example is regular English plurals.  For the lists 
[single, plural], [cat, dog, bear, house, homes, shack, car, automobile, truck]
, all of the plurals are obtained by just adding an 's'.  The typical lesson creator will just proceed 
through the list and add an s to every plural.  The advanced user should have the 
option to specify
(plural, [cat, dog, bear, house, homes, shack, car, automobile, truck]) -> grab the root and tack on an s

I proposed passing the tuple to a regular expression search and replace provided 
by the lesson creator.  I would separate the individual items of the tuple with the
secondary separator. So for this specific example the search pattern for single case would be
single;(*)
The replacement pattern for single case would be
\1
The search pattern for plural would be
plural;(*)
The replacement pattern for plural case would be
\1s

The UI would need: a toggle for each list so the lesson creator can specify which 
list are fully expanded and which are dealt with by the regular expression,  a popup explaining the
secondary separator convention, and a box showing the results of the regular expression editing.
This is exactly as complicated as the various regular expression helpers in existence.

Again constructions can be tested collectively in the inflection tool, or individually in multiple
choice/flash card game.  Training will be unchanged by this feature.

This is the full extent of the automatically generated content that I proposed.
I do envision the advanced student using it to create short sentences. 
I will use it to create sets of prepositional phrases, which are otherwise missing in Parley.
Parley is not going to have a deep understanding of grammar and 
make meaningful paragraphs.  

This solution is intentionally, open ended so that a motivated lesson 
creator can add considerable sophistication to the lesson plan. 


Andreas