MIT Press

Table 11

Token-Level probing task results for all languages. Bold represents the best score, and italics represents the second best.

Finnish
Task	baseline	MUSE	word2vec	GloVe-BPE	fastText	D-ELMo	C-ELMo
Case	35.9	79.7	69.1	85.0	97.3	97.9	98
Mood	89.7	94.9	94.3	96.7	97.4	97.3	98.3
Number	82.2	92.8	90.4	94.4	97.8	98.3	98.7
POS	29.5	69.0	72.0	68.9	71.3	74.8	87.5
Person	64.5	92.8	89.7	95.1	97.3	96.8	96.4
Tense	62.9	95.2	92.8	97.5	98.4	98.4	98.1
Voice	86.2	96.2	92.7	96.5	97.7	98.0	97.5
German
Task	baseline	MUSE	word2vec	GloVe-BPE	fastText	D-ELMo	C-ELMo
Case	32.7	56.9	52.9	53.0	56.8	53.4	80.8
Gender	38.6	72.6	69.3	66.5	73.5	72.8	78.7
Mood	96.4	98.3	98.4	98.4	98.9	98.7	98.9
Number	79.5	88.5	88.1	86.1	89.5	89.2	94.7
POS	20.6	76.5	82.6	74.5	76.4	77.1	91.2
Person	72.8	94.6	94.1	93.8	95.1	95.1	97.9
Tense	51.6	98.7	97.8	98.7	98.6	99.2	98.8
Spanish
Task	baseline	MUSE	word2vec	GloVe-BPE	fastText	D-ELMo	C-ELMo
Gender	58.0	98.6	95.3	97.9	99.2	99.3	99.4
Mood	92.3	97.7	97.2	95.9	99.1	98.2	97.8
Number	75.2	98.7	96.4	98.6	99.6	99.6	99.6
POS	19.3	78.1	83.2	78.0	79.6	78.9	92
Person	68.4	97.4	95.3	98.1	99.2	99.0	99.0
Tense	55.4	98.2	97.1	98.7	99.3	99.2	99.1
Russian
Task	baseline	MUSE	word2vec	GloVe-BPE	fastText	D-ELMo	C-ELMo
Case	31.7	75.3	47.6	68.2	78.6	78.7	89.7
Gender	45.4	86.0	64.7	87.0	89.9	89.8	89.7
Number	75.3	93.9	78.2	92.8	96.8	97.0	97.4
POS	30.3	76.4	64.5	73.3	76.5	76.9	95.4
Person	51.1	91.8	84.3	98.1	98.6	98.6	98.8
Tense	53.0	90.6	78.4	97.6	98.7	97.2	96.9
Voice	69.5	88.9	79.0	95.6	93.7	95.4	94.5
Turkish
Task	baseline	MUSE	word2vec	GloVe-BPE	fastText	D-ELMo	C-ELMo
Case	47.7	84.3	67.7	89.0	91.4	95.4	96.6
POS	35.4	71.7	68.0	75.4	78.3	77.4	83.2
Person	78.3	91.0	84.3	95.0	97.2	96.8	96.5
Polarity	89.0	95.4	92.3	98.3	99.2	99.0	98.8
Tense	55.2	87.7	76.3	94.6	96.6	95.9	96.3

Finnish
Task	baseline	MUSE	word2vec	GloVe-BPE	fastText	D-ELMo	C-ELMo
Case	35.9	79.7	69.1	85.0	97.3	97.9	98
Mood	89.7	94.9	94.3	96.7	97.4	97.3	98.3
Number	82.2	92.8	90.4	94.4	97.8	98.3	98.7
POS	29.5	69.0	72.0	68.9	71.3	74.8	87.5
Person	64.5	92.8	89.7	95.1	97.3	96.8	96.4
Tense	62.9	95.2	92.8	97.5	98.4	98.4	98.1
Voice	86.2	96.2	92.7	96.5	97.7	98.0	97.5
German
Task	baseline	MUSE	word2vec	GloVe-BPE	fastText	D-ELMo	C-ELMo
Case	32.7	56.9	52.9	53.0	56.8	53.4	80.8
Gender	38.6	72.6	69.3	66.5	73.5	72.8	78.7
Mood	96.4	98.3	98.4	98.4	98.9	98.7	98.9
Number	79.5	88.5	88.1	86.1	89.5	89.2	94.7
POS	20.6	76.5	82.6	74.5	76.4	77.1	91.2
Person	72.8	94.6	94.1	93.8	95.1	95.1	97.9
Tense	51.6	98.7	97.8	98.7	98.6	99.2	98.8
Spanish
Task	baseline	MUSE	word2vec	GloVe-BPE	fastText	D-ELMo	C-ELMo
Gender	58.0	98.6	95.3	97.9	99.2	99.3	99.4
Mood	92.3	97.7	97.2	95.9	99.1	98.2	97.8
Number	75.2	98.7	96.4	98.6	99.6	99.6	99.6
POS	19.3	78.1	83.2	78.0	79.6	78.9	92
Person	68.4	97.4	95.3	98.1	99.2	99.0	99.0
Tense	55.4	98.2	97.1	98.7	99.3	99.2	99.1
Russian
Task	baseline	MUSE	word2vec	GloVe-BPE	fastText	D-ELMo	C-ELMo
Case	31.7	75.3	47.6	68.2	78.6	78.7	89.7
Gender	45.4	86.0	64.7	87.0	89.9	89.8	89.7
Number	75.3	93.9	78.2	92.8	96.8	97.0	97.4
POS	30.3	76.4	64.5	73.3	76.5	76.9	95.4
Person	51.1	91.8	84.3	98.1	98.6	98.6	98.8
Tense	53.0	90.6	78.4	97.6	98.7	97.2	96.9
Voice	69.5	88.9	79.0	95.6	93.7	95.4	94.5
Turkish
Task	baseline	MUSE	word2vec	GloVe-BPE	fastText	D-ELMo	C-ELMo
Case	47.7	84.3	67.7	89.0	91.4	95.4	96.6
POS	35.4	71.7	68.0	75.4	78.3	77.4	83.2
Person	78.3	91.0	84.3	95.0	97.2	96.8	96.5
Polarity	89.0	95.4	92.3	98.3	99.2	99.0	98.8
Tense	55.2	87.7	76.3	94.6	96.6	95.9	96.3

View Large

This Feature Is Available To Subscribers Only