Minimalistic Approach to Coreference Resolution in Lithuanian Medical Records

Table 2

Comparison of coreference resolution methods for Balto-Slavic languages.

Method	Foundation	Precision	Recall	F1
LVCoref [45]	Rule based, Hobbs’ algorithm	0.69–0.88	0.66–0.80	0.68–0.84
Ruler [46]	Rule based	0.59–0.65	0.50–0.75	0.55–0.69
BARTEK [47]	Machine learning	0.58	0.65	0.61
Mixed [48]	Deep learning, sieve based	0.70	0.68	0.69
RU-sys1 [49]	Rule based, ontology	0.82	0.70	0.76
RU-sys2 [49]	Rule based	0.71	0.58	0.64
RU-sys3 [49]	Rule based	0.63	0.50	0.55
RU-sys4 [49]	Statistical, ontology	0.54	0.51	0.53
RU-sys5 [49]	Machine learning, semantics	0.58	0.42	0.49
RU-sys6 [49]	Decision tree	0.36	0.15	0.21
Khadzhiiskaia and Sysoev [50]	Machine learning	0.84	0.77	0.80
Kučová and Žabokrtský [51]	Rule-based filters	0.60	na	na
CZ classifier [52]	Classifier-based machine learning	0.70–0.76	0.70–0.76	0.70–0.76
CZ ranker [52]	Ranker-based machine learning	0.79	0.79	0.79
Treex CR (Czech, English) [53]	Machine learning	na	na	0.61–0.68
Treex CR (Russian, German) [54]	Machine learning, projection	0.50–0.64	0.15–0.24	0.25–0.34