Should categories really fallback to /sl if no /tl?

[unhammer]

If we have something like

number = (GD sg) sg pl GD ;
a_adj = sint ord ;
adj: _.a_adj.number;
A -> %adj { %1 };

and bidix output is ^good<adj><sint>/OK<adj>$ then I don't want the <sint> to be output, that's an "inherent" tag (the word OK is can't do synthetic comparative/superlative, so it doesn't get a <sint>). But to ensure that, I have to explicitly say that the attribute should have the tl value, for each such where bidix can remove it:

A -> %adj { %1[a_adj=1.a_adj/tl]  };

I know I can always be explicit like this, but I would think the default should be to not fallback here, that if bidix removes a tag, it's for a reason. Does anyone actively use sl fallbacks?

@jonorthwash @ftyers @marcriera and anyone else using apertium-recursive

[jonorthwash]

Yes, I've run into this before and agree. Subcategory sl or tl tags (or even grammatical tags) included in bidix should somehow be treated together by default.

I've talked with @mr-martian about it before, and it seems like there are two issues involved in its implementation:

• What is the right way to handle all this, at the level of the syntax?
• How to implement it technically?

My impression is that the first issue is the blocking one.

[mr-martian]

A couple of options that come to mind that might address this issue and definitely wouldn't break backwards compatibility:

adj: _.a_adj/tl.number so that a_adj will never get clipped from sl but number retains the current default behavior

A compiler option that makes all unspecified clips be tl clips rather than the current tl > ref > sl fallback approach

Either of those would be pretty easy to implement, but there could very well be an option I haven't thought of that would be more correct.

[unhammer]

I would prefer an all-or-nothing option, but would it be possible to do it as a file pragma? Like the CG subreadings option, we could have SIDE_SOURCES = tl ; or something (default value being = tl ref sl ;).

Since the rules in the files necessarily will depend on the value of that option, I prefer having it inside the rtx, instead of having to look it up in Makefile.am. (Also, having to remember compile switches while doing rtx-comp manually to debug means one more way I can mess things up at the time when I am the most confused.)

[marcriera]

I've just bumped into this while doing some tests with apertium-spa-cat, where some entries "lose" a tag in bidix.

I agree with @unhammer regarding including the option directly in the file. Using a compiler option will probably lead to confusion. However, I think the two proposed solutions shouldn't be mutually exclusive. Currently I don't really need fallbacks and disabling them entirely would solve my issues, but I can imagine myself using them in the future. In that case, it would be good to have the ability to do something like adj: _.a_adj/tl.number and keep fallbacks enabled globally. Otherwise, we would need to add conditionals and rules could get too complicated.

[mr-martian]

I have implemented @unhammer's SIDE_SOURCES = tl ref sl ; syntax. It can be gotten by depending on version 1.1.2.

I'm leaving this issue open because I like the adj: _.a_adj/tl.number syntax but I need to think a little more about how to implement it.