The approach of not connecting the ltop of the determiner and the label of its relation does not translate into RMRS-RASP because we need access to the label to connect up the Nbar. Hence, for now, pretend that the NP is headed by RULE, so we can disconnect the label at that point ... RASP doesn't distinguish between main verb `have' and the auxiliary, which probably means we can't make `have' semantically empty. Presumably we have the same issue with `do' (ERG has a contentful `do' for `do the dishes' etc). There is also a contentful `be' in the ERG for `Abrams is a dog' etc. Test suite examples. The lack of comments indicates a perfect match (up to the level being considered). I haven't systematically noted cases where the ERG-RMRS has a more specific pred name than RASP-RMRS. I also haven't noted cases where the mismatch arises from something already noted - e.g. the `it rained' always mismatches. 1. It rained RASP has no chance of distinguishing expletive `it' 2. Abrams barked CARG treatment - case is an issue. Lexical entries in the ERG have nearly all got completely lower case CARG values (it appears that something may be downcasing systematically anyway, but I can't easily check this). For the time being, all RASP CARG values are also downcased. In principle, RMRS allows case to be preserved in CARGs (and only in CARGs), so this issue may be revisited if it's useful for disambiguation or something. 3. The window opened open_v_1 in ERG vs open_v (note that `Abrams opened the window' gives `open_v_cause' (ARG1 ARG2) in the ERG) 4. Abrams chased Browne 5. Abrams handed Browne the cigarette 6. Abrams handed the cigarette to Browne ARG1, ARG2, ARG3 in ERG (with to_p_sel) but ARG1 ARG2 with `real' prep in RASP. Can't do anything about this, since it might be a `real' `to' in this rule, although it would be nice to figure out a way to make the parses look more similar. 7. Abrams bet Browne a cigarette that it rained expl it as with `hand' - ARG2 is a cigarette, ARG3 is Browne `that' supplies the prpstn rel in the ERG, but the prpstn is coming from the construction in RASP (and `that' is semantically empty). Hence mismatch in character position, but OK otherwise. 8. Abrams knew that it rained prpstn - as 7 9. Abrams intended to bark RASP is missing the ARG1 for `bark' here. To get this to work would require having two anchors and retrieving the second as well as the first when we get to the subject. May be possible, but difficult. Would require checking that V1/v_inf didn't turn up in contexts where this isn't the right behaviour. 10. Abrams intended Browne to bark At the moment, these differ because the subject for Browne is missing. It would need further investigation to see whether the V1/v_np_inf rule could make the np the subject of the inf or not. 11. Every cat barked 12. Every cat chased some dog 13. My cat barked This is quite horrible. The only way the RASP version can work properly is for the code to recognise that we've got a RSTR from the DEF_EXPLICIT_Q_REL already with a qeq, and when we get another one for the N, it'd have to merge them ... Leave it like this for now, but see if a better idea emerges! 14. It barked 15. The cat chased it 16. The cat chased itself 17. The cat chased one This is the pronoun `one' - in RASP it has the same tag as somebody and everybody - which means we have to treat it as very underspecified 18. Mine barked 19. That opened In the ERG there's a generic_nom_rel which can't be replicated in RASP. The RASP output is ill-formed because `that' is treated as a quantifier but there's nothing to act as a RSTR. This is not likely to be fixable - RASP treats DD1 as an NP here - this is just the ordinary np_vp rule. 20. Cats bark 21. Tobacco arrived. This one works because of the S/n1_vp rule - its application is a mixed bag - seem to be a lot of errors. But overall, putting the quantifier on here looks like the right thing to do. 22. Some bark. RASP doesn't give the correct parse here. But cf 19 anyway. 23. Some of the cats bark. NP/np-pro_pp-of rule - maybe OK to fix by putting construction semantics on it, but need to come back to this. Jan 1 - sampling NP/np-pro_pp-of occurs with `some, most, one, much, many, any, several, none' which have this semantics in ERG (also `those' - generic-nom_rel (see above) `who' and `that' - don't parse in ERG) Looks like it's worth adding the semantics to the construction, but the ARG1 of part_of binds the index of the NP within the PP. So getting a proper match to the ERG here would either require a special treatment of `of' PPs (which requires features, but seems a plausible thing to do since `of' is nearly always treated differently from ordinary PPs), or a fudge, where the repetition of the ARG2 causes X1 to be bound to the NP index (cf 13). If we allow this, it makes the semantics much less constrained. 24. No cat barked. 25. Did the dog bark? Wrong parse from RASP 26. Which dog barked? The ERG gives both an int_m_rel and a prpstn_rel here. S/np-wh_vp is set to give an underspecified message relation prop_ques_m_rel: this would subsume either of those relations, and is thus OK for the relative clause case too. 27. Whose dog barked? whose = `which person' in the ERG - fix RASP later via SEM-I 28. Chase Browne! 29. Abrams wondered which dog barked. see 26 30. Abrams wondered whether Browne barked. Subordinating conjunctions like `whether' and `if' end up looking like PPs in RASP, which makes it difficult to produce any sort of sensible semantics. Revisit this. 31. The dog that Browne chased barked. N1/n_s is used both for relative clauses starting with that and for Ns like belief which subcat for a sentence. There's no way currently of underspecifying the attachment, so the S is currently not attached at all in RASP. Revisit this. Actually the comments in the tsg14.1 file suggest that the N1/n_s rule should only apply to complements - ask ejb? 32. The dog to chase is barking. Same as above but with N1/n_inf - N could subcat for the `to-' VP or be intersective, as here. Comments in tsg14.1 confirm intention that this covers both cases. 33. The dog was chased by Browne. Right now, this really doesn't work. RASP-RMRS treats `the dog' as the ARG1 and the only way of fixing this would be to make the semantics sensitive to the class of V (either by making the semantics conditional on some syntactic feature, or perhaps more easily, making it conditional on an aspect marker on the event). 34. The dog chased by Browne barked. 35. The dog is barking. There are two issues with auxiliaries: the first is that we haven't agreed a set of values for tense etc yet ... The second is that the auxiliary will contribute a rel in RASP because all have a contentful equivalent which isn't straightforward to distinguish. 36. The dog has barked. 37. The dog has been barking. 38. The dog had been barking. 39. The dog will bark. 40. The dog is going to bark. all these concern tense - not sorted out yet 41. The dog could bark. 42. The dog couldn't bark. a) characterization differences RASP/ERG - RASP treats as could not b) could scopes over not in RASP - inverse in ERG c) BUG - ARG1 in neg_rel - return to this when we've decided proper approach 43. The old dog barked. mostly OK but RESTR doesn't match because qeq hasn't had ING var replaced. FIX compare.lisp End of comparison notes! 44. The dog barked softly. 45. The dog probably barked. 46. The dog barked in the garden. 47. The dog barks now. 48. The garden dog barked. 49. The tobacco garden dog barked. 50. The cat is old. 51. The cat is in the garden. ; 51 52. The barking dog chased Browne. 53. Chased dogs bark. 54. Chasing the cat is old. 55. That the cat chases Browne is old. 56. Dogs chase whatever barks. 57. The dog barked every day. 58. When did the dog bark? 59. Three of the dogs bark. 60. Three bark. 61. Browne's dog barks. 62. Browne's barks. 63. Twenty three dogs bark. 64. Two hundred twenty dogs bark. 65. Abrams arrived by car. 66. Abrams kept barking. 67. Browne squeezed the cat in. 68. Browne squeezed in the cat. 69. The picture of Abrams arrived. 70. Abrams wiped the table clean. 71. Abrams put Browne in the garden. 72. The dog will bark if Browne arrives. 73. Abrams and Browne arrived. 74.Abrams, Browne and the dog arrived. 75. The dog arrived and barked. 76. The dog arrived and Browne barked. 77. The dog barked, didn't it? 78. It is obvious that the dog barked. 79. Abrams promised Browne to bark. 80. Abrams seems to bark. 81. Abrams believes Browne to be barking. 82. It bothered Abrams that Browne barked. 83. It took Abrams ten minutes to arrive. 84. Abrams left it to Browne to bark. 85. Abrams strikes Browne as old. 86. Browne considers Abrams old. 87. Abrams liked the idea that Browne could bark. 88. Abrams barked from ten to three. 89. Abrams was very old. 90. Nearly every dog barked. 91. Abrams barked very softly. 92. Browne's chasing of cats bothered Abrams. 93. It bothered Browne that Abrams chased cats. 94. June third arrived. 95. Abrams arrived at three twenty. 96. Browne arrived on Tuesday morning. 97. The cats found a way to bark. 98. The happier dog chased Browne. 99. There were cats in the garden. 100. That dog chased Browne. 101. Somebody chased Abrams. 102. How happy was Abrams? 103. The number five bothers Browne. 104. Abrams could. 105. Browne tried to. 106. Don't bark! 107. The dog arrived barking.