Problem: Maritime QA system aims at enhancing human-machine communications. For converting human natural language into a machine-understandable language (e.g., SQL queries), a semantic parser is needed. It requires a very large amount of high-quality manually-annotated data for training.
Achievement: Possibility of training a semantic parser with limited manually-annotated data enabling zero-shot training.
Benefit to Defence: In many real-world Defence scenarios, it is infeasible to obtain large amount of training data, i.e. human utterances paired with complex logical forms. Our approach addresses this issue.
Next steps: Adding other modalities like images/videos.
problem addressed
What?
Converting human utterances into a machine-understandable language (e.g., SQL queries).
Non-availability of a large-scale manually-annotated data for training a semantic parser, in many real-world Defence scenarios.
How?
We propose to design compact grammars to simultaneously generate both (i) logical forms, and (ii) canonical utterances, hence synthesizing the training dataset.
outcomes
A domain-adaptive QA System.
A web-based user interface for the backend QA system is further developed and delivered to Defence.
An effective approach for generating large-scale domain specific training dataset for Question Answering (QA) systems.
We are interested in endowing the situational awareness system with a capability to assist human decision makers by allowing them to interact with the system using human natural language.
big picture for Defence
Ultimate Goal
Augment the intelligence of intelligence analysts using cutting edge AI approaches
To provide intelligence analysts with knowledge and inferences using advanced AI approaches, helping them to improve the efficiency and effectiveness.
This is achieved by a hybrid neuro-symbolic approach.
publications
Shiri, F, Zhuo, TY, Li, Z, Nguyen, V, Pan, S, Wang, W, . . . Li, Y-F 2022, ‘Paraphrasing Techniques for Maritime QA system’, 2022 25th International Conference on Information Fusion (FUSION), IEEE, pp. 1-8. DOI: 10.48550/ARXIV.2203.10854
Shiri, F, Wang, T, Pan, S, Chang, X, Li, Y-F, Haffari, R, . . . Yu, S 2021, ‘Toward the Automated Construction of Probabilistic Knowledge Graphs for the Maritime Domain’, 2021 IEEE 24th International Conference on Information Fusion (FUSION), IEEE, pp. 1-8. DOI: 10.23919/FUSION49465.2021.9626935