|
Word embeddings |
POS, head POS, dependency relation, gender, case, number, tense |
Classification |
|
Word embeddings |
POS, dependency relations, morphological features, emotions |
Classification |
|
Word embeddings |
Referential attributes |
Classification |
|
RNN states/gates |
POS, syntactic role, gender, case, definiteness, verb form, mood |
Classification, correlation |
|
RNN states |
POS, top syntactic sequence, smallest constituent, tense, voice |
Classification |
|
RNN states/gates |
Acoustic features |
Correlation |
|
RNN gates |
Phoneme boundaries |
Change in activation signal |
|
RNN states |
POS, ancestor label prediction, dependency relation prediction |
Classification |
|
Sentence embeddings |
Sentence length, word presence, word order |
Classification |
|
Sentence embeddings |
Sentence length, word presence, word order; orthography; social tasks |
Classification |
|
Sentence embeddings |
Semantic role, word presence |
Classification |
|
Sentence embeddings |
Synthetic syntactic patterns |
Clustering |
|
Sentence embeddings |
Sentence length, word presence, word order; tree depth, top constituent; main tense, subject/object number, semantic odd man out, coordinate inversion |
Classification |
|
Sentence embeddings |
Sentence length, word presence, word order; POS, word sense disambiguation; sentence order |
Classification |
|
Attention weights |
POS, word alignment |
Distribution measures, match with alignments |
|
Attention weights |
Anaphora |
Attention score |
|
Attention weights |
Word sense disambiguation |
Distribution measures |
|
Hidden activations in feed-forward acoustic model |
Phonemes, phonetic features, gender |
Clustering, average activations by group/label |
|
Hidden activations in feed-forward acoustic model |
Phonemes, phonetic features |
Classification, clustering measures |
|
CNN/RNN activations |
Phonetic units |
Classification |
|
Speaker embeddings |
Speaker, speech content, word order, utterance length, channel, gender, speaking rate |
Classification |
|
CNN activations |
Style, accent, broadcast program |
Classification |
|
Audio-visual RNN activations |
Utterance length, word presence, homonym disambiguation |
Classification, regression, similarity measures |
|
Audio-visual CNN embeddings |
Word classes |
Clustering |
|
RNN states in audio-visual model |
Phonemes, synonyms |
Classification, clustering, discrimination |
|
Audio-visual CNN activations |
Phonemes, speakers, word identity |
Clustering, discrimination |
|
Word embeddings, vision CNN |
Concepts |
Similarity measures |
|
RNN states |
Word presence |
Direct classification |
|
NMT encoder neurons |
Sentence length |
Regression |
|
NMT word embeddings |
synonyms, morphological features |
Nearest neighbors |
|
NMT states |
POS, morphology |
Classification |
|
NMT states |
POS, lexical semantics |
Classification |
|
NMT states |
Morphology |
Classification |
|
NMT and LM neurons |
POS, morphology, lexical semantics |
Classification |
|
RNN states |
Subject-verb agreement |
Likelihood comparison, direct classification |
|
RNN / self-attention states |
Subject-verb agreement |
Likelihood comparison, direct classification |
|
RNN states |
Number agreement |
Likelihood comparison |
|
Parser word embeddings |
Word features (shape, etc.) |
Classification; also other methods |
|
RNN sentence embedding |
Hierarchical structure |
Classification |
|
biLM representations (RNN, Transformer, gated CNN) |
POS, constituency parsing, coreference |
Classification; similarity scores |
|
RNN states in language-vision model |
Word boundary, word similarity |
Classification |
|
Hidden activations in feed-forward audio-visual model |
Phonetic features |
Discrimination |