Open-Vocabulary Semantic Parsing with both Distributional Statistics and Formal Knowledge

Authors: Matt Gardner, Jayant Krishnamurthy

AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our open-vocabulary semantic parser on a fill-in-the-blank natural language query task. By giving open vocabulary semantic parsers direct access to KB information, we improve mean average precision on this task by over 120%.
Researcher Affiliation Industry Matt Gardner, Jayant Krishnamurthy Allen Institute for Artificial Intelligence Seattle, Washington, USA {mattg,jayantk}@allenai.org
Pseudocode No The paper describes the model components and equations but does not present structured pseudocode or algorithm blocks.
Open Source Code Yes All of the data and code used in these experiments is available at http://github.com/allenai/open vocab semparse.
Open Datasets Yes We thus use the dataset introduced by Krishnamurthy and Mitchell (2015), which consists of the ClueWeb09 web corpus3 along with Google s FACC entity linking of that corpus to Freebase (Gabrilovich, Ringgaard, and Subramanya 2013).
Dataset Splits Yes We also used the test set created by Krishnamurthy and Mitchell, which contains 220 queries generated in the same fashion as the training data from a separate section of Clue Web. However, as they did not release a development set with their data, we used this set as a development set. For a final evaluation, we generated another, similar test set from a different held out section of Clue Web, in the same fashion as done by Krishnamurthy and Mitchell. This final test set contains 307 queries.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU, GPU models, memory, or cloud instance types) used for running the experiments.
Software Dependencies No The paper mentions tools like a "CCG parser" and uses "Freebase", but it does not specify version numbers for any software dependencies or libraries used in their implementation.
Experiment Setup Yes In each of these models, we used vectors of size 300 for all embeddings.