Lexical entry filtered out? How to understand why?

grammars SRG
#1

My newest problem debugging the updated SRG :).

In the SRG updated with new Freeling tags and token mapping and lexical filtering, I lose the gold parse for La ventana se abrió. (‘The window opened.’)

In the old reference grammar, I can see the gold parse:

Screen Shot 2023-01-10 at 2.28.47 PM

With the new grammar, I believe what I am missing is the desired lexical entry for se, SE_PR. The only one I get in the new chart is SE_PR_I, which means impersonal rather than reflexive. The verb abrió does not want that, it wants the reflexive one.

Running ACE with --vv, I can confirm that the lexeme is found:

lexical lookup found lexeme 'se_pr'
lexical lookup found lexeme 'se_pr-i'
...
generated #520 1/1 from #232, #9 for [2-3] with rule p03cn00: se
generated #521 1/1 from #232, #10 for [2-3] with rule p03cn00: se
...
lexical edge #520 le 0x7f907e470100 vtx [2-3] lexeme 'se_pr'
lexical edge #521 le 0x7f907e470df8 vtx [2-3] lexeme 'se_pr-i'
...
finished lexical parsing
finished lexical filtering
added lexeme `el_d' = #516  [0-1]
added lexeme `ventana_n' = #517  [1-2]
added lexeme `se_pr-i' = #521  [2-3]
added lexeme `abrir_vprn-pp_e_cp_p' = #523  [3-4]
...

So, the desired lexical entry disappears after lexical filtering, looks like? Or could it be the “lexical parsing” stage as well?..

I checked the lexical item for SE_PR, and I do not think it underspecifies TRAITS or anything like that; it ultimately inherits from lex-item which is properly amended in tmt.tdl.

Anyone could help me understand what is going on? Many thanks in advance.

#2

I am assuming the description of the problem is not good enough in order to trigger any suggestions?

I am wondering what I could do to make any progress here, including how to phrase the issue differently… Hmm.

#3

Based on the log it does look like lexical filtering may be removing edge #520. In the most recent SRG I have at my end, se_pr has a TRAITS list that looks native, as it should. The next question is what about the p03cn00 rule – is it copying TRAITS up properly now? In my copy it is not, but I believe you made an across the board fix for that problem since my copy’s vintage. If it is still not copied up, that would result in an undefined choice between se_pr and se_pr-i being killed by the native-generic filter rule.

1 Like
#4

Ah! Thank you, @sweaglesw. Really not sure why I wasn’t able to debug this myself, as this is indeed the same issue as before: the lexical rule supertype ultimately inherits from basic-lex-rule. The “across the board” fix hasn’t yet happened at this stage, because if I make basic-lex-rule propagate TRAITS, then other things break. So I need to fish out all such instances separately. I still don’t understand the grammar well enough. For example, I don’t know why some lexical rules inherit from lex-rule while others from basic-lex-rule.