[1909.03368] Designing and Interpreting Probes with Control Tasks

Probes, supervised models trained to predict properties (like parts-of-speech) from representations (like ELMo), have achieved high accuracy on a range of linguistic tasks. But does this mean that the representations encode linguistic structure or just that the probe has learned the linguistic task? In this paper, we propose control tasks, which associate word types with random outputs, to complement linguistic tasks. By construction, these tasks can only be learned by the probe itself. So a good probe, (one that reflects the representation), should be selective, achieving high linguistic task accuracy and low control task accuracy. The selectivity of a probe puts linguistic task accuracy in context with the probe's capacity to memorize from word types. We construct control tasks for English part-of-speech tagging and dependency edge prediction, and show that popular probes on ELMo representations are not selective. We also find that dropout, commonly used to control probe complexity,

1 mentions: @johnhewtt
Date: 2019/09/10 03:47

Referring Tweets

@johnhewtt How do we design probes that give us insight into a representation? In #emnlp2019 paper with @percyliang, our "control tasks" help us understand the capacity of a probe to make decisions unmotivated by the repr. paper: t.co/hGxcle0ar4 blog: t.co/hkl93ZGvit t.co/1NA5hoyF7t

Related Entries

Read more GitHub - Hironsan/anago: Bidirectional LSTM-CRF and ELMo for Named-Entity Recognition, Part-of-Speec...
0 users, 0 mentions 2018/04/22 03:40
Read more Improving a Sentiment Analyzer using ELMo — Word Embeddings on Steroids – Real-World Natural Languag...
0 users, 0 mentions 2018/10/30 19:17
Read more The Illustrated BERT, ELMo, and co. (How NLP Cracked Transfer Learning) – Jay Alammar – Visualizing ...
0 users, 7 mentions 2019/03/01 00:47
Read more The Illustrated BERT, ELMo, and co. (How NLP Cracked Transfer Learning) – Jay Alammar – Visualizing ...
23 users, 23 mentions 2018/12/03 19:09
Read more GitHub - Hironsan/anago: Bidirectional LSTM-CRF and ELMo for Named-Entity Recognition, Part-of-Speec...
0 users, 0 mentions 2018/06/08 04:06