MRS Won't Generate

Similar to my post yesterday, I have another MRS that won’t generate, but this time it’s not ill-formed, and it seems to match exactly the MRS the ERG outputs.

The string I’m going for is “the glowing grass.”


[ TOP: h12
  INDEX: e9
  RELS: < [ unknown LBL: h11 ARG: x5 ARG0: e9 ]
          [ def_udef_a_q LBL: h8 ARG0: x5 RSTR: h6 BODY: h7 ]
          [ _glow_v_1 LBL: h4 ARG0: e2 ARG1: x5 ]
          [ _grass_n_1 LBL: h4 ARG0: x5 ] >
  HCONS: < h6 qeq h4 h12 qeq h11 > ]


[ TOP: h0
  INDEX: e2 [ e SF: prop ]
  RELS: < [ unknown<0:17> LBL: h1 ARG: x4 [ x PERS: 3 NUM: sg ] ARG0: e2 ]
          [ _the_q<0:3> LBL: h5 ARG0: x4 RSTR: h6 BODY: h7 ]
          [ _glow_v_1<4:11> LBL: h8 ARG0: e9 [ e SF: prop TENSE: untensed MOOD: indicative PROG: + PERF: - ] ARG1: x4 ]
          [ _grass_n_1<12:17> LBL: h8 ARG0: x4 ] >
  HCONS: < h0 qeq h1 h6 qeq h8 > ]

I’m wondering if there’s something about this -ing construction that makes generation difficult.

You’re right to suspect the -ing construction. The obstacle you’re hitting has to do with the PROG aspectual feature which distinguishes the progressive (-ing) form of verbs from other forms. The ERG currently has the SEM-I set up to make the default value of PROG be - if it is not already assigned (presumably for better efficiency in generation), but there is a note in the file erg/semi.vpm observing that this default value causes trouble for generation. While I try to figure out a better balance for efficiency and completeness, you can decorate your MRS for such constructions with a minimal constraint on the ARG0 event variable of that modifying verb: let that EP be

[ _glow_v_1 LBL: h4 ARG0: e2 [e PROG: bool] ARG1: x5 ]

This assignment of the neutral value “bool” (for boolean: either plus or minus) to the feature PROG still saves you from having to decide whether the grammar wants PROG + or PROG - for this construction, but it stops the SEM-I from aggressively assigning a - value to PROG, so your MRS will generate successfully.