Structured Reordering for Modeling Latent Alignments in Sequence Transduction
Authors: bailin wang, Mirella Lapata, Ivan Titov
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on synthetic problems and NLP tasks (semantic parsing and machine translation) showing that modeling segment alignments is beneficial for systematic generalization. |
| Researcher Affiliation | Academia | 1University of Edinburgh 2University of Amsterdam 3Innopolis University |
| Pseudocode | Yes | Algorithm 1 Dynamic programming for computing marginals and differentiable sampling of permutation matrix wrt. a parameterized grammar |
| Open Source Code | Yes | 1Our code and data are available at https://github.com/berlino/tensor2struct-public. |
| Open Datasets | Yes | Our code and data are available at https://github.com/berlino/tensor2struct-public. ... We use the SCAN dataset [27] ... We use Geo Query dataset [56] ... We use the small en-ja dataset extracted from TANKA Corpus. ... 8https://github.com/odashi/small_parallel_enja ... We extract a subset from FBIS corpus (LDC2003E14) |
| Dataset Splits | Yes | For standard splits (IID), we randomly sample 20k infix-postfix pairs whose nesting depth is set to be between 1 and 6; 10k, 5k, 5k of these pairs are used as train, dev and test sets, respectively. ... The original split (IID) has 50k/500/500 examples for train/dev/test ... The LEN split has 50k/538/538 examples for train/dev/test, respectively. ... The IID split which has 141k/3k/3k examples for train/dev/test, respectively. ... The LEN split has 140k/4k/4k examples as train/dev/test sets respectively. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running experiments, such as GPU models, CPU types, or memory specifications. |
| Software Dependencies | No | The paper does not list specific software dependencies with their version numbers, such as Python versions, deep learning frameworks (e.g., PyTorch, TensorFlow) with their versions, or specific library versions used for reproducibility. |
| Experiment Setup | No | The paper describes the datasets used and the evaluation metrics, but it does not provide specific experimental setup details such as concrete hyperparameter values (e.g., learning rate, batch size, number of epochs), optimizer settings, or model initialization procedures in the main text. |