Old Catalan Morphosyntax: Developing an Annotated Corpus

Figures & Tables

Incremental build-up and time investment.

TRAINING ROUND	HOURS INVESTED	WORDS/HOUR	WORD/HOUR INCREASE
0–4,500	32	140,6	0%
4,500–10,000	32	312,5	122,26%
10,000–20,000	18	555,6	295,16%
20,000–40,000	23.5	851,6	505.59%
40,000–60,000	12	1666.7	1085.42%

TARGER results.

TOKENS	GLOBAL ACCURACY
10,000	85.5
60,000	91.4

MBT results per training round.

TOKENS	GLOBAL ACCURACY	KNOWN WORDS	UNKNOWN WORDS
4,500	86.1	94.6	46.8
10,000	91.3	96.4	56.4
20,000	93.8	97.6	58.4
40,000	94.7	97.2	64.3
60,000	95.8	97.6	68.1

MBT results for highest and lowest frequency unseen tokens.

POS TAG	PRECISION	RECALL	F-SCORE	N
N	0.74	0.75	0.74	856
NPR	0.9	0.94	0.92	585
VNI	0.66	0.67	0.67	219
VBDI^{^}3^{^}PL	0.73	0.75	0.74	188
VBDI^{^}3^{^}SG	0.65	0.68	0.66	173
PRO^{^}3^{^}PL	0	0	0	1
PRO^{^}A^{^}3^{^}SG	0	0	0	1
PRO^{^}D^{^}2^{^}PL	0	0	0	1
PRO^{^}D^{^}3^{^}SG	0	0	0	0
OLB	0	0	0	0

MBT results for highest and lowest frequency tokens.

POS TAG	PRECISION	RECALL	F-SCORE	N
P	1	1	1	6191
N	0.98	0.98	0.98	6185
CONJ	1	1	1	4919
C	1	1	1	4139
COMMA	1	1	1	3910
PRO^{^}RFL^{^}2^{^}SG	1	1	1	2
ADJ^{^}POS	0	0	0	2
VBI^{^}2^{^}SG	0	0	0	2
VBDI^{^}2^{^}PL	0	0	0	1
”	0	0	0	0