CTeTex ↔ mova — converter draft

The rule language is The Mova dialect: a UD metadialect, treebanks, and converters. Conventions beyond the specification are in English UD 2.18 treebanks — comparative seed. Matches were counted by en expert-check: ctetex → mova on CTeTex, mova → ctetex on EWT.

CTeTex is test only, so the main direction is mova → ctetex: the en output must be brought to CTeTex customs so that quality is measured rather than the difference between dialects. ctetex → mova is needed for the round trip and for checking the two roads to the goal.

CTeTex FEATS are incomplete (CTeTex — FEATS reduced to Number and Tense). On the way there the converter does not fill them in: a test does not need training. On the way back it reduces the full mova FEATS to the CTeTex set.

ctetex → mova

rule: ctetex.nn-compound
what: noun before a noun as nmod without a preposition → compound
from: ctetex
to: mova
match: h[upos=NOUN|PROPN]; n[upos=NOUN|PROPN, rel=nmod, head=h, before=h]
require: none c[rel=case, head=n]
set: n[rel=compound]
source: UD en compound; seeds/nmod-compound.md

rule: ctetex.prt
what: particle on a verb as compound → compound:prt
from: ctetex
to: mova
match: v[upos=VERB]; p[upos=ADP|ADV, rel=compound, head=v, after=v]
set: p[rel=compound:prt]
source: UD en compound:prt; seeds/old-conventions.md

rule: ctetex.poss
what: possessive DET nmod → PRON nmod:poss, lemma = form
from: ctetex
to: mova
match: p[form=my|your|his|its|our|their, upos=DET, rel=nmod]
set: p[upos=PRON]; p[rel=nmod:poss]; p[lemma=@form]
source: UD en nmod:poss, pos/PRON; EWT README v2.11

rule: ctetex.wh-advmod
what: subordinate when/where as SCONJ mark → ADV advmod
from: ctetex
to: mova
match: w[form=when|where|how|why|whenever, rel=mark]
set: w[upos=ADV]; w[rel=advmod]
source: EWT README v2.11 (#88)

rule: ctetex.an
what: lemma an → a
from: ctetex
to: mova
match: d[form=an, upos=DET]
set: d[lemma=a]
source: UD en pos/DET

rule: ctetex.enum
what: list item number as nummod on the predicate → discourse
from: ctetex
to: mova
match: h[upos=VERB|ADJ|ADV|AUX]; n[upos=NUM|X, rel=nummod, head=h, before=h]
set: n[rel=discourse]
source: UD docs/changes.md No. 14; EWT README v2.14 (#518)

Matches: nn-compound 472 (of 487 nmod nouns before a head noun, 15 have a preposition), prt 14, poss 16, wh-advmod 36, an 41, enum 25.

nn-compound takes not only adjacent pairs (307, CTeTex — a noun before a noun: half nmod, half compound) but also chains: in Speed Dump Lag Frames several nmod attach to Frames. In EWT such chains are also compound.

mova → ctetex

rule: mova.ctetex.prt
what: compound:prt → compound
from: mova
to: ctetex
match: p[rel=compound:prt]
set: p[rel=compound]
source: seeds/old-conventions.md

rule: mova.ctetex.poss
what: possessive nmod:poss → DET nmod
from: mova
to: ctetex
match: p[form=my|your|his|its|our|their, upos=PRON, rel=nmod:poss]
set: p[upos=DET]; p[rel=nmod]
source: seeds/old-conventions.md

rule: mova.ctetex.when
what: subordinate when/where on advcl → SCONJ mark
from: mova
to: ctetex
match: v[rel=advcl]; w[form=when|where, rel=advmod, head=v, before=v]
set: w[upos=SCONJ]; w[rel=mark]
source: seeds/old-conventions.md

rule: mova.ctetex.an
what: lemma an
from: mova
to: ctetex
match: d[form=an, upos=DET]
set: d[lemma=an]
source: seeds/old-conventions.md

rule: mova.ctetex.enum
what: discourse numbering marker → nummod
from: mova
to: ctetex
match: h[upos=VERB|ADJ|ADV|AUX]; n[upos=NUM|X, rel=discourse, head=h, before=h]
set: n[rel=nummod]
source: seeds/lists.md

rule: mova.ctetex.feats
what: CTeTex FEATS — only Number, Tense, Typo, ExtPos
from: mova
to: ctetex
match: t[]
set: t[feats-=Person]; t[feats-=VerbForm]; t[feats-=Mood]; t[feats-=Voice]; t[feats-=Case]; t[feats-=Gender]; t[feats-=PronType]; t[feats-=Poss]; t[feats-=Reflex]; t[feats-=Definite]; t[feats-=Degree]; t[feats-=NumType]; t[feats-=NumForm]; t[feats-=Polarity]; t[feats-=Foreign]; t[feats-=Abbr]; t[feats-=Style]
source: seeds/feats-minimal.md

rule: mova.ctetex.number-nonoun
what: Number only on NOUN, AUX and VERB
from: mova
to: ctetex
match: t[upos=PROPN|PRON|DET|ADJ|NUM|ADV|ADP|X|SYM]
set: t[feats-=Number]
source: seeds/feats-minimal.md (PROPN without Number — 293 of 293)

rule: mova.ctetex.verb-pres
what: VERB in the present — without Tense; Number only in 3sg
from: mova
to: ctetex
match: v[upos=VERB, feats.Tense=Pres]
set: v[feats-=Tense]
source: seeds/feats-minimal.md (VERB Tense=Pres — 0)

rule: mova.ctetex.verb-past
what: VERB in the past — without Number
from: mova
to: ctetex
match: v[upos=VERB, feats.Tense=Past]
set: v[feats-=Number]
source: seeds/feats-minimal.md (VERB Tense=Past — 286, without Number)

rule: mova.ctetex.verb-plur
what: VERB with Number=Plur — without Number (VBP)
from: mova
to: ctetex
match: v[upos=VERB, feats.Number=Plur]
set: v[feats-=Number]
source: seeds/feats-minimal.md

Needs string XPOS: CTeTex has no XPOS. A xpos=_ action is needed — clear XPOS. If xpos= accepts only PTB tags, this is convert-todo. The rule itself is t[] → t[xpos=_], as in ATIS ↔ mova — converter draft.

Matches on EWT: prt 907, poss 3467, when 414, an 619, enum 75, feats 254 820, number-nonoun 117 378, verb-pres 8030, verb-past 8571, verb-plur 2398.

Round trip ctetex → mova → ctetex

Lossless:

compound:prt (14);
possessives (16);
an (41);
numbering markers (25);
when on advcl.

FEATS do not change on the way there, and the reduction on the way back does not touch them. The exception is 3 VERB with Number=Plur, which verb-plur removes.

Lossy:

A noun before a noun. On the way there 472 nmod → compound. On the way back the CTeTex custom cannot be predicted: CTeTex itself splits adjacent pairs in half, 324 versus 307. So 472 DEPREL divergences, 5.1% of words.
when not on advcl. 36 mark become advmod on the way there, and on the way back only the 32 that are on advcl and before it get mark. 4 UPOS and DEPREL divergences.
3 VERB Number=Plur in FEATS.

Total: DEPREL ≈ 94.9% (the limit is noun pairs), UPOS ≥ 99.9%, FEATS ≥ 99.9%, lemmas 100%.

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/treebanks/en/ctetex/convert/