Established: August 15, 2015
Building a rekentuig system to automatically solve math word problems written ter natural language. SigmaDolphin is a project initiated te early 2013 at Microsoft Research Asia, with the primary objective of building a rekentuig slim system with natural language understanding and reasoning capacities. Wij concentrate on the application of automatic problem solving, i.e., automatically solving problems (especially math word problems) written ter natural language.
Rekentuig programs can finish many tasks much more effectively and efficiently than human beings, such spil calculating the product of two large numbers, or finding all occurrences of a string te a long text. However, the spectacle of computers on many slim tasks is still low. For example, te a talking script, computers often generate irrelevant or incorrect responses, and wij can lightly find ridiculous results ter automatic machine translation, and it is still a very challenging task for state-of-the-art laptop programs to solve even primary-school-level math word problems.
A well trained person can evidently outperform computers on the above slim tasks. The main reason, wij believe, lies te our amazing power of natural language understanding and reasoning (especially common sense reasoning) overheen state-of-the-art artificial slim (AI) systems.
Once wij are able to empower computers with the capability of natural language understanding and reasoning, wij may see significant spectacle improvements on text processing tasks like question answering and machine translation. Te addition, it may open doors to fresh applications which are impractical so far because of the limited slim level of current pc systems.
Present Targeting Script: Math Word Problem Solving
Wij drive our system vormgeving and implementation by the screenplay of automatically solving math word problems. It is an ideal commencing point to our research because of the following reasons:
- Math word problems typically contain concise text with clear semantics
- They fundamentally voorkant a broad range of domains (considering the fact that math word problems are often simplified versions of real problems from various domains)
- It is relatively effortless to measure spectacle: A loterijlot of problems have bot compiled for education purpose, one can often unambiguously determine whether the solutions of a math problem is onberispelijk or not.
The following research topics are very related to our project.
- Meaning representation: How to represent the semantic meaning of natural language text
- Semantic parsing: How to parse natural language text into structured semantic representation
- Reasoning: How to perform reasoning based on the structured semantic representation spil well spil common sense skill