Parse Tree Fragmentation of Ungrammatical Sentences

Authors: Homa B. Hashemi, Rebecca Hwa

IJCAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show that the proposed fragmentation strategies are competitive with existing methods for making fluency judgments; they also suggest that the overall framework is a promising way to handle syntactically unusual sentences.
Researcher Affiliation Academia Homa B. Hashemi, Rebecca Hwa Intelligent Systems Program, Computer Science Department University of Pittsburgh hashemi@cs.pitt.edu, hwa@cs.pitt.edu
Pseudocode No The paper describes algorithms in prose but does not include any pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper does not provide any explicit statement about releasing open-source code for the described methodology, nor does it include a link to a code repository.
Open Datasets Yes English as a Second Language corpus (ESL) The First Certificate in English (FCE) dataset [Yannakoudakis et al., 2011] is a learner s corpus that contains ungrammatical sentences and their corresponding error corrections. Machine Translation corpus (MT) The LIG corpus [Potet et al., 2012] contains 10,881 French-English machine translation outputs and their human post-editions.
Dataset Splits Yes From this corpus, we create two datasets for the experiments. First, we randomly select 5000 sentences with at least one error; this dataset is for training the CLF fragmentation method as well as for the intrinsic evaluation of different fragmentation methods. Then, we create a second, non-overlapping dataset for the extrinsic evaluation. It consists of 7000 sentences and is representative of the corpus s error distribution; there are 2895 sentences with no error; 2103 with one error; 1092 with two errors; and 910 with 3+ errors. We then build a dataset of 4000 sentences with HTER score more than 0.1 for training CLF and intrinsic evaluation, and a dataset of 6000 sentences with real distribution of HTER scores for extrinsic evaluation. We performed a 10-fold cross validation for the two domains: ESL and MT.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies Yes The Stanford parser version 3.2.0 [Klein and Manning, 2003] is used to generate parses for all sentences. For all binary classification or regression tasks (which will be discussed in Section 5.2), we run a 10-fold cross validation with the standard Gradient Boosting Classifier or Regressor [Friedman, 2001] in the scikit-learn toolkit [Pedregosa et al., 2011].
Experiment Setup Yes We tune Gradient Boosting parameters with a 3-fold cross validation on the training data: learning rate over the range {0.0001 . . . 100} by multiples of 10 and max depth over the range {1 . . . 5}.