0.25
0.5
0.75
1.25
1.5
1.75
2
Should Model Architecture Reflect Linguistic Structure?
Published on 2016-05-277941 Views
Sequential recurrent neural networks (RNNs) over finite alphabets are remarkably effective models of natural language. RNNs now obtain language modeling results that substantially improve over long-st
Related categories
Presentation
Should Neural Network Architecture Reflect Linguistic Structure?00:00
Learning language - 100:34
Learning language - 201:48
Learning language - 302:24
Learning language - 402:34
Learning language - 502:57
Learning language - 603:38
Compositional words - 105:25
Compositional words - 206:17
Example: Dependency parsing07:16
Dependency parsing CharLSTM > Word Lookup - 108:43
Dependency parsing CharLSTM > Word Lookup - 209:52
Language modeling: Word similarities11:57
Structure-aware words - 113:54
Structure-aware words - 214:02
Open Vocabulary LMs14:37
Open Vocabulary LMs: Turkish morphology15:06
Input word representation16:50
Open Vocabulary LM17:25
Character vs. word modeling: Summary17:54
Modeling syntax18:57
Language is hierarchical20:36
One theory of hierarchy24:15
Terminals, Stack, Action26:28
Syntatic Composition - 128:31
Syntatic Composition - 229:42
Recursion - 130:11
Recursion - 230:15
Syntatic Composition - 330:28
Implementing RNNGs: Parameter Estimation30:42
Implementing RNNGs: Inference31:51
English PTB (Parsing)32:47
English PTB (LM and Chinese CTB (LM))33:29
This Talk, In a Nutshell33:59
Thanks34:23