Page 62 - My FlipBook
P. 62
工
智
慧
計
畫
Arti cial Intelligence Projects A primary objective of our project is to construct a apples is worth 30 dollars." By extension, the question could
primary school MWP system that can not only solve a be posed "How many kilos of apples would 400 dollars
word problem, but also explain the process by which a buy?" or "How much would 8 kilos of apples cost?" This
solution was derived. We used RPBA to transform a given scenario implies that every sentence related to a problem
sentence into its corresponding incidence map and added plays a role that can be associated with another sentence.
its mathematical entailments. For example, the sentence Through such sentence compatibility, it is possible to derive
"Ten kilos of apples are worth 300 dollars." represents "the associated sentences from the question sentence.
unit price". Its mathematical entailment is "Every kilo of
In order to describe the solution process, we have created 2. Find out which two persons represent "both of
natural language scripts (NL script) in solution ow charts. them".
To illustrate this, consider the following problem:
3. Find out the number of stickers each one has.
A sticker sells for $5. John has two stickers, and Josh 4. Add the numbers in (3) together.
bought $20 worth of stickers. How many do both of
them have? Each instruction in these NL solution scripts can yield
finer scripts that are executed by the model to identify
From the sentence, "A sticker sells for $5," we can obtain answers and, ultimately, to compile an explanation. The
the unit price of a sticker. From "Josh bought $20 worth of
stickers," we derive that "Josh bought four stickers" or "Josh rst NL solution scripts are written manually. Later, scripts
has four stickers." From the question "How many do both for similar sentences can be written automatically via the
of them have?" we can create the following NL solution same framework by substitution. Similarity is measured by
scripts: matching the incidence maps. Furthermore, appropriate
word generalization can be adopted to expand the
1. Find out how many items of "what". application range of these NL solution scripts.
The advantage of using NL solution scripts in ow charts is errors without having to rely on expert programmers. Thus,
that domain experts lacking a programming background more versatile NL solution scripts can be designed to meet
(such as Project Manager) would also be able to understand the speci c needs of customers.
and write these instructions. Moreover, they could correct
Summary /
This project has two major contributions. First, we have developed a simple and unique RPBA algorithm,
that couples reduction with our SPBA approach, which can be used to parse and understand the Chinese
language. Second, we are creating NL solution scripts to describe solution ow charts and explanations.
Such scripts separate domain knowledge and programs, allowing non-programmers to write instructions
as NL solution scripts and to construct more user-friendly applications.
60
智
慧
計
畫
Arti cial Intelligence Projects A primary objective of our project is to construct a apples is worth 30 dollars." By extension, the question could
primary school MWP system that can not only solve a be posed "How many kilos of apples would 400 dollars
word problem, but also explain the process by which a buy?" or "How much would 8 kilos of apples cost?" This
solution was derived. We used RPBA to transform a given scenario implies that every sentence related to a problem
sentence into its corresponding incidence map and added plays a role that can be associated with another sentence.
its mathematical entailments. For example, the sentence Through such sentence compatibility, it is possible to derive
"Ten kilos of apples are worth 300 dollars." represents "the associated sentences from the question sentence.
unit price". Its mathematical entailment is "Every kilo of
In order to describe the solution process, we have created 2. Find out which two persons represent "both of
natural language scripts (NL script) in solution ow charts. them".
To illustrate this, consider the following problem:
3. Find out the number of stickers each one has.
A sticker sells for $5. John has two stickers, and Josh 4. Add the numbers in (3) together.
bought $20 worth of stickers. How many do both of
them have? Each instruction in these NL solution scripts can yield
finer scripts that are executed by the model to identify
From the sentence, "A sticker sells for $5," we can obtain answers and, ultimately, to compile an explanation. The
the unit price of a sticker. From "Josh bought $20 worth of
stickers," we derive that "Josh bought four stickers" or "Josh rst NL solution scripts are written manually. Later, scripts
has four stickers." From the question "How many do both for similar sentences can be written automatically via the
of them have?" we can create the following NL solution same framework by substitution. Similarity is measured by
scripts: matching the incidence maps. Furthermore, appropriate
word generalization can be adopted to expand the
1. Find out how many items of "what". application range of these NL solution scripts.
The advantage of using NL solution scripts in ow charts is errors without having to rely on expert programmers. Thus,
that domain experts lacking a programming background more versatile NL solution scripts can be designed to meet
(such as Project Manager) would also be able to understand the speci c needs of customers.
and write these instructions. Moreover, they could correct
Summary /
This project has two major contributions. First, we have developed a simple and unique RPBA algorithm,
that couples reduction with our SPBA approach, which can be used to parse and understand the Chinese
language. Second, we are creating NL solution scripts to describe solution ow charts and explanations.
Such scripts separate domain knowledge and programs, allowing non-programmers to write instructions
as NL solution scripts and to construct more user-friendly applications.
60
