Verb Class Induction with Partial Supervision
Authors: Daniel Peterson, Susan Brown, Martha Palmer8616-8623
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate that this partial supervision guides the resulting clusters effectively, improving the recovery of both labeled and unlabeled classes by 16%, for a joint 12% absolute improvement in F1 score compared to clustering without supervision. |
| Researcher Affiliation | Collaboration | Daniel Peterson Oracle Labs Burlington, MA 01803 daniel.peterson@oracle.com Susan W. Brown, Martha Palmer University of Colorado Boulder, CO 80309 {susan.brown, martha.palmer}@colorado.edu |
| Pseudocode | No | The paper does not contain any sections or figures explicitly labeled as 'Pseudocode' or 'Algorithm', nor are there structured algorithmic steps formatted like code. |
| Open Source Code | No | The paper mentions future work on building a tool: 'The most obvious next step for this research is to build that tool, and use it to expand and improve Verb Net in as many languages as possible.' However, there is no explicit statement or link indicating that the code for the current methodology is open-source or publicly available. |
| Open Datasets | Yes | Semlink (Bonial, Stowe, and Palmer 2013b) provides labels of Verb Net class for each sentence in the Penn Treebank s Wall Street Journal corpus (Marcus et al. 1994). Our primary sources of data are Gigaword (Parker et al. 2011) and the Wall Street Journal sections of the Penn Treebank (Marcus et al. 1994), both licensed through the Linguistic Data Consortium. |
| Dataset Splits | Yes | To test whether a small number of labels can improve the senses learned from LDA, we split this annotation into a training portion and a test portion... we first split the data by Verb Net class, using 2/3 of the classes as training (hereafter, C1 denotes the set of classes in the training portion of the split) and 1/3 for testing (C2). We then split by verb, keeping 2/3 for training and 1/3 for testing... This training/test split produced 6400 sentences with known labels for training and 6500 for testing. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware used to run the experiments, such as CPU or GPU models, memory, or cloud instance types. |
| Software Dependencies | No | The paper discusses the use of 'latent Dirichlet allocation (LDA)' and 'MCMC algorithm' but does not specify any software names with version numbers for these or any other dependencies. |
| Experiment Setup | No | The paper describes conceptual aspects of the experimental setup, such as initializing topic models, observing labeled sentences, and including a hyperparameter for weighting labeled instances. However, it does not provide specific numerical values for hyperparameters (e.g., the weight of labeled instances, learning rates, batch sizes, number of epochs) or other detailed training configurations. |