Comparisons (and equatives, which we treat similarly) are some of the trickiest structures to annotate. The correct structure depends on the size of the standard of comparison (clause vs. smaller phrase) and type of comparee (a specific NP vs. a clause or vague comparison). Correlative comparisons like je...desto are different still.
Non-clausal standard of comparison
If only an NP or something similarly small is the standard of comparison, tag so/als/dan/wie (and possibly gleich)as P, with the standard of comparison as its complement. (Here we follow the older Penn corpora, rather than HeliPaD practice of treating these as clauses with much elision.) The complement of this P is often NP, but can also be ADVP, IP-INF, or CP-THT (but not CP(X)-CMP, see below). The complement NP doesn’t get a grammatical function tag (as the object of a P) but any nominal head is tagged for case, using the same case as the comparee:
… (ADJP-PRD
(ADV so)
(ADJ^N^PL groß) <- the comparee
(PP (P wie) <- change 'C-comparison' to 'P'
(NP (N^N^PL hering)))) <- NP standard of comparison
…
(ID 1557_Staden_Historia_Hesse.,141))… (NP-OB2
(QP (QR^D^PL me) <- comparee here is 'mehr gulden'
(PP (P dan)
(NP (NUM^D^PL 6000)))) <- NP standard of comparison
(N^D^PL gulden))
…
(ID 1360_Hauwe_NeuesBuch_Cologne.,71))
'... more guilders than 6000 [guilders] ...'… (IP-MAT
(CONJ Ind)
(NP-SBJ *con*)
(VBDI^3^PL baden)
(NP-OB1 (PRO sij))
(ADVP (ADV guotlichen) <- comparee is an ADV
(PP (P as)
(ADVP (ADV vuor)))) <- standard of comparison is ADV
…
(ID 1360_Hauwe_NeuesBuch_Cologne.,37)) The PP standard of comparison should be the sister of the comparee, as in the three examples above. (If the comparative PP and the comparee are not adjacent, the comparative PP is annotated as being moved out of the comparee's phrase.) If comparison is to a verb or the clause, the PP is daughter of IP as in the example below.
The word gleich is typically treated as an adverb preceding the comparative C or P als/wie/dan. If gleich itself is the sole comparison word, it may be tagged as C or P:
( (IP-MAT (NP-SBJ (D^N^SG das))
(VBDS^3^SG schine)
(ADVP (ADV gleich)) <- 'gleich' with a comparative word
(PP (P als)
(NP (ONE^N^SG eyn) (N^N^SG creutz)))
...
(ID 1557_Historia_Travel_Hesse.,349))(IP-MAT-CONJ (CONJ dan)
(NP-SBJ (Q^N^SG solches))
(BEPI^3^SG ist)
(, /)
(PP (P gleich) <- 'gleich' as the sole comparative
(PP (P in) <- standard of comp is itself PP!
(NP (Q^D^PL anderen)
(N Sprachen))))
(, /)
(ADVP (ADV also))
(ADVP (ADV auch))
(PP (P in)
(NP (D^D^SG der)
(ADJ Teutschen)))
(NP-PRD (ONE^N^SG ein)
(ADJP (ADV sehr)
(ADJ gemeine))
(N Regel))
...
(ID 1609_Quad_Teutscher_Cologne.,71))Clausal standard of comparison
Comparison to a specific constituent
When the standard of comparison is a clause, there are three possible structures (CP-CMP for specific comparison, CPX-CMP for vague comparison, and CP-THT for dann dass), each treated in turn below. CP-CMP and CPX-CMP are introduced by a C (so, als, dan, wie, or possibly gleich as described above). As with CP-ADV, CP-CMP may have more than one C (so wie, als wie, als dan,etc, but see below for dann dass); however, als ob (als wenn) is considered conditional rather than comparative, and should thus be CP-ADV.
If the comparison is to a specific constituent, use CP-CMP inside (or moved out of) the comparee's phrase. E.g. with comparison to a N, CP-CMP is inside (or moved out of) the NP that it modifies. CP-CMP has null wh-operator in its spec, coindexed to a trace in the IP-SUB; this serves to link the elided constituent in the comparison clause with the comparee. IP-SUB indicates that nothing is elided except the argument linked to the comparee; IPX-SUB means that additional material is elided.
… (NP-POS (Q alsulgs)
(N briefs) <- comparee is a N; CP-CMP is its sister
(CP-CMP
(WNP-OB1-1 0) <- null op links the missing argument from
(C as) … the comparison clause to the comparee
(IP-SUB (NP-OB1 *T*-1)
(NP-SBJ (PRO sij))
(PP (P vuor) (N zijden))
(PP (P mit)
(NP (NP-POS (D der) (N gemeynden))
(N guoede)))
(PP (P an) (NP (D deme) (N Rijche)))
(VBN (VBN geworuen) (CONJ vnd) (VBN behalden))
(HVDS^3^PL hedden))))
…
(ID 1360_Hauwe_NeuesBuch_Cologne.,68))
Note that especially equatives may occur with no comparison word. These should have a null C (and a null wh-phrase if the comparison is to a particular constituent):
... (CP-REL (WADVP-1 (ADV+P dahin))
(C 0)
(IP-SUB (ADVP *T*-1)
(NP-SBJ (D^N^SG der)
(N Satan))
(VBPI^3^SG komt)
(, /)
(ADVP (ADV so)
(ADV offt) <- the comparee is an adv
(CP-CMP (WADVP-2 0) <- null operator
(C 0) <- null subordinator
(IP-SUB (ADVP *T*-2)
(NP-SBJ (PRO^N^PL sie))
(NP-OB1 (PRO^A^SG es))
(VBPI^3^PL begeren)))))))
(. .))
(ID 1624_Schiffarten_Travel_Switzerland.,177))Vague comparison, or comparison to a clause
If the standard of comparison is a clause and the comparee is also a clause (or the comparee is vague), the comparsion is a CPX-CMP, as no null operator can be posited. Any CPX-CMP is a daughter of IP.
(IP-MAT …
(CPX-CMP (C as)
(IP-SUB (NP-SBJ *pro*)
(ADVP (ADV herna))
(PTP (VBN geschreuen))
(VBPI^3^SG steit)))
…
(ID 1360_Hauwe_NeuesBuch_Cologne.,77))
'as stands written hereafter'
If material (other than the null operator) has been elided, CP(X)-CMP dominates IPX-SUB:
( (IP-MAT (ADVP (ADV so))
(BEPI^3^SG ist)
(NP-SBJ (PRO er))
(ADJP-PRD (NP-COM (D^G^PL der)
(N sünden))
(ADJ frey))
(, /)
(CPX-CMP (C wie) <- comparison is to more than one constituent
(IPX-SUB <- missing verb and 'frey'
(NP-SBJ (ONE^N^SG ein)
(N hund))
(ADJP-PRD (NP-COM (D^G^PL der)
(N floech)))))
(, .))
(ID 1534_Franck_Weltbuch_Swabia.,582))
'thus he is free from sins, as a dog [is] [free] from fleas'Comparison with dann dass
If the comparison clause begins with dann dass, treat dann as a P that takes the that-clause as its complement: ... (NP-OB1 …
(Q anders)
(, /)
(PP (P dan)
(CP-THT (C das)
(IP-SUB (NP-SBJ (PRO$ sein)
(N will))
(VBPS^3^SG beschee))))))
(, /)
(ID 1533_Rodler_Fierrabras_Rhine.,228))Correlative comparatives
Correlative comparitives are like German je...je or je...desto. Correlative equatives may be so... so...
Both correlates are daughters of the same IP. One correlateis an ADVP, ADJP, or NP-ADT with embedded CP-CMP (a clause compared to a specific constituent), and the other correlate is an ADVP, ADJP, or argument NP:
( (IP-MAT (ADVP (ADV ye)
(ADVR weitter) <-1st correlate is the comparee
(CP-CMP (WADVP-1 0) <- standard of comparison clause
(C 0)
(IP-SUB (ADVP *T*-1)
(NP-SBJ (PRO wir))
(ADVP (ADV hinnein))
(VBDI^1^PL fuohren))))
(, /)
(ADJP-PRD (ADV ye) (ADJR lenger)) <- 2nd correlate
(NP-SBJ (D die)
(N refier))
(BEDI^3^SG war)
(, .)
(ID 1557_Staden_Historia_Hesse.,336))( (IP-MAT (CONJ Dann)
(NP-SBJ (PRO^N^SG er))
(HVPI^3^SG hat)
(NP-OB1 (QP (ADV so) <- 1st corr. is the actual object
(Q^A^PL manch))
(N^A^PL Muenche)
(N^A^PL Closter))
(VBN gebawet)
(, /)
(NP-ADT (ADV so) <- 2nd correlate (comparee) is an adjunct
(Q^A^PL viel)
(CP-CMP (WQP-1 0) <- with the standard of comparison
(C 0)
(IP-SUB (NP-SBJ (QP *T*-1)
(N^N^PL Buchstaben))
(PP (P i@)
(NP (D^D^SG @m)
(N^D^SG Alphabet)))
(BEPI^3^PL sind))))
...
(ID 1599_Bange_ThurChron_Thuringia.,666))