English Light Verb Construction Identification Using Lexical Knowledge

Authors: Wei-Te Chen, Claire Bonial, Martha Palmer

AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Evaluation shows that this system achieves an 89% F1 score (four points above the state of the art) on the BNC test set used by Tu & Roth (2011), and an F1 score of 80.68 on the Onto Notes test set, which is significantly more challenging.
Researcher Affiliation Academia Wei-Te Chen Department of Computer Science University of Colorado at Boulder Weite.Chen@colorado.edu Claire Bonial Department of Linguistics University of Colorado at Boulder Claire.Bonial@colorado.edu Martha Palmer Department of Linguistics University of Colorado at Boulder Martha.Palmer@colorado.edu
Pseudocode No The paper describes the system and its features, but no pseudocode or algorithm blocks are provided.
Open Source Code No The paper does not provide a statement about releasing open-source code for the described methodology or a link to a repository.
Open Datasets Yes This research uses several resources: Prop Bank (PB) (Palmer, Guildea, and Kingsbury 2005), the Onto Notes (ON) sense groupings (Pradhan et al. 2007), Word Net (WN) (Fellbaum, Grabowski, and Landes 1998) and the British National Corpus (BNC). 1http://www.natcorp.ox.ac.uk/XMLedition/ 3http://wordnet.princeton.edu/wordnet/
Dataset Splits No The BNC data is a balanced data set, including 1,039 positive LVC examples and 1,123 negative examples. We randomly sample 90% of the instances for training and the rest for testing. ... In all of the ON data, 1,768 LVCs are annotated (in Table 3). Among all these LVCs in ON, 1,588 LVCs are listed in the training data set, and 180 LVCs are in the testing data set.
Hardware Specification No The paper does not specify the hardware used for running the experiments.
Software Dependencies No This supervised classifier is trained with the Lib Linear (Fan et al. 2008) algorithm. ... We first train and evaluate our model with the BNC data using automatic parsers produced by Clear NLP (Choi and Mccallum 2013).
Experiment Setup No For the target word, we select 3 words as the window size, while we adopt the same feature list that was used in Lee (2002). It lacks other specific experimental setup details like learning rates, epochs, or optimizer settings.