Domain Adaptation for Semantic Parsing

Authors: Zechang Li, Yuxuan Lai, Yansong Feng, Dongyan Zhao

IJCAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on a benchmark dataset show that our method consistently outperforms several popular domain adaptation strategies. We conduct experiments on OVERNIGHT [Wang et al., 2015], a semantic parsing dataset with paired utterances and logical forms (in function calling styled Lambda DCS) in 8 domains. We summarize all the model performance in Table 3.
Researcher Affiliation Academia 1Wangxuan Institute of Computer Technology, Peking University, Beijing, China 2Center for Data Science, Peking University, Beijing, China 3The MOE Key Laboratory of Computational Linguistics, Peking University, China
Pseudocode No The paper describes the model architecture and training procedures using natural language and mathematical equations, but does not include any structured pseudocode or algorithm blocks.
Open Source Code Yes The implemention is open source.1 1https://github.com/zechagl/DAMP
Open Datasets Yes We conduct experiments on OVERNIGHT [Wang et al., 2015], a semantic parsing dataset with paired utterances and logical forms (in function calling styled Lambda DCS) in 8 domains
Dataset Splits Yes Only 10% of the target domain training data is employed, to stimulate a harsher domain adaptation scenario. We randomly sample 20% of training instances for validation. Detailed statistics are shown in Table 2.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU models, CPU models, memory) used to run the experiments.
Software Dependencies No The paper mentions using Glove for word embeddings, Bi-LSTM, LSTM, and RMSProp optimizer, but does not provide specific version numbers for any software dependencies or libraries.
Experiment Setup Yes For implementation details, word embeddings are initialized with Glove [Pennington et al., 2014]. And our encoders are Bi-LSTM with 300 hidden size. For domain relevance attention, rc = 60 and rf = 2. As for balance parameters, λc is 0.4 and λf is 0.2. All activation functions are tanh. The dropout rate and L2 regularization rate are 0.6 and 1e 5, respectively, with batch size of 64. We use the RMSProp optimizer [Tieleman and Hinton, 2012] with learning rate= 1e 3 and decay factor = 0.9.