The Gap of Semantic Parsing: A Survey on Automatic Math Word Problem Solvers

Area of Science:

• Artificial Intelligence (AI)
• Natural Language Processing (NLP)
• Computational Linguistics
• Educational Technology

Background:

• Automatic solving of mathematical word problems (MWPs) is a long-standing challenge, dating back to the 1960s.
• Recent advancements in AI have renewed interest in MWPs, viewing successful solutions as a milestone towards Artificial General Intelligence (AGI).
• Existing MWP solvers demonstrate limited precision on large, diverse datasets, indicating a gap between current capabilities and desired performance.

Purpose of the Study:

• To present a comprehensive survey of automatic math problem solvers.
• To provide a clear and complete picture of the state-of-the-art in MWP solving.
• To identify emerging research directions in the field.

Main Methods:

• Focus on algebraic word problems, summarizing extracted features and techniques used to bridge the semantic gap.
• Compare the performance of various MWP solving methods on publicly accessible datasets.
• Include coverage of automatic solvers for other problem types, such as geometric problems requiring diagram understanding.

Main Results:

• Current MWP solvers exhibit insufficient precision when applied to large and diverse datasets.
• Significant challenges remain in bridging the semantic gap between natural language and machine-understandable logical representations.
• A comparative analysis of techniques and their performance on benchmark datasets is provided.

Conclusions:

• Despite progress, current automatic math problem solvers require substantial improvement for real-world applicability.
• The survey highlights the need for more robust methods to handle the complexity and diversity of MWPs.
• Several promising research avenues are identified to advance the field of automatic MWP solving.