2 |
Language Models Use Monotonicity to Assess NPI Licensing ...
|
|
|
|
BASE
|
|
Show details
|
|
5 |
Causal Transformers Perform Below Chance on Recursive Nested Constructions, Unlike Humans ...
|
|
|
|
BASE
|
|
Show details
|
|
6 |
Sparse Interventions in Language Models with Differentiable Masking ...
|
|
|
|
Abstract:
There has been a lot of interest in understanding what information is captured by hidden representations of language models (LMs). Typically, interpretation methods i) do not guarantee that the model actually uses the encoded information, and ii) do not discover small subsets of neurons responsible for a considered phenomenon. Inspired by causal mediation analysis, we propose a method that discovers within a neural LM a small subset of neurons responsible for a particular linguistic phenomenon, i.e., subsets causing a change in the corresponding token emission probabilities. We use a differentiable relaxation to approximately search through the combinatorial space. An $L_0$ regularization term ensures that the search converges to discrete and sparse solutions. We apply our method to analyze subject-verb number agreement and gender bias detection in LSTMs. We observe that it is fast and finds better solutions than the alternative (REINFORCE). Our experiments confirm that each of these phenomenons is mediated ... : 12 pages, 4 figures, 6 tables ...
|
|
Keyword:
Computation and Language cs.CL; FOS Computer and information sciences; Machine Learning cs.LG
|
|
URL: https://arxiv.org/abs/2112.06837 https://dx.doi.org/10.48550/arxiv.2112.06837
|
|
BASE
|
|
Hide details
|
|
7 |
Mechanisms for Handling Nested Dependencies in Neural-Network Language Models and Humans ...
|
|
|
|
BASE
|
|
Show details
|
|
9 |
Compositionality decomposed: how do neural networks generalise? ...
|
|
|
|
BASE
|
|
Show details
|
|
10 |
Under the Hood: Using Diagnostic Classifiers to Investigate and Improve how Language Models Track Agreement Information ...
|
|
|
|
BASE
|
|
Show details
|
|
11 |
Do Language Models Understand Anything? On the Ability of LSTMs to Understand Negative Polarity Items ...
|
|
|
|
BASE
|
|
Show details
|
|
|
|