|
JAIST Repository >
i. 北陸先端科学技術大学院大学(JAIST) >
i20. 学位論文 >
M-IS. 修士（情報科学） >
M-IS. 2012年度(H24) >
このアイテムの引用には次の識別子を使用してください:
http://hdl.handle.net/10119/10781
|
| タイトル: | Modifying the Learning Algorithm on the Kirby's Model for Simultaneous Bilingual Case |
| 著者: | Namchisawatwong, Teerat |
| キーワード: | Iterated Learning Model
Grammar acquisition |
| 発行日: | Sep-2012 |
| 抄録: | To study the language evolution, there are many frameworks for inves-
tigating the cultural evolution of linguistic structure. One of them is a
famous Iterated Learning Model (ILM) that named Kirby's model.
Kirby's model can work well in various experiments. But it cannot sim-
ulate some of case studies such as case studies of Latin language evolution.
I am interested in this problem and set the objective of the research that is
to design and implement language separator in order to make the Kirby's
model which supports bilingual environments.
Language separator is a feature helping agent in Kirby's model differ-
entiate two languages by observing the frequency of the co-occurrence of
vocabularies and sentence structure. And Language separator will be suc-
cessful, when languages in former generation can be transmitted to the
next generation.
For this objective, we purposed the method. First of all, we modified
the Kirby's model according to bilingual education in the real world. We
give more definition about language in modified Kirby's model. We change
the start generation from agent with blank grammar to agent with ideal
grammar. We set a new characteristic of speaker agent and learner agent.
We modify the speak process and invention to make agent can speak two
language without mixing them. We modify the learning process by adding
scoring system to evaluate the relation score of each rule named front end
process. The mechanism of front end process is observing co-occurrence of
using rules. And we implemented the separator to help agent differentiate
the languages by using relation score table that got from front end process.
The separator will be applied when learner agent change role to be speaker
agent.
After applying our proposed method, languages in former generation
can be transmitted to the next generation. It shows the best performance
when using the grammar that has no common area between two languages
as grammar of n-th generation. And it shows the good performance when
using the grammar that the number of rules in common area between two
languages is grammar of n-th generation less than four. |
| 記述: | Supervisor:Satoshi Tojo
情報科学研究科
修士 |
| タイトル（英語）: | Modifying the Learning Algorithm on the Kirby's Model for Simultaneous Bilingual Case |
| 著者（英語）: | Namchisawatwong, Teerat |
| 言語: | eng |
| URI: | http://hdl.handle.net/10119/10781 |
| 出現コレクション: | M-IS. 2012年度(H24) (Jun.2012 - Mar.2013)
|
このアイテムのファイル:
| ファイル |
記述 |
サイズ | 形式 |
|---|
| abstract.pdf | | 66Kb | Adobe PDF | 見る/開く | | paper.pdf | | 644Kb | Adobe PDF | 見る/開く |
|
当システムに保管されているアイテムはすべて著作権により保護されています。
|
