-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.html
128 lines (114 loc) · 5.8 KB
/
index.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
<!DOCTYPE html>
<html>
<head>
<!--
<meta http-equiv="refresh" content="600">
-->
<meta http-equiv="Content-Type" content="text/html;charset=utf-8" />
<title>Code-switching Sentence Generation</title>
<script type="text/javascript" src="js/jquery-2.1.3.js"></script>
<script type="text/javascript" src="js/bootstrap.min.js"></script>
<script type="text/javascript" src="js/bootstrap.bundle.min.js"></script>
<script type="text/javascript" src="js/index.js"></script>
<link rel="stylesheet" type="text/css" href="css/bootstrap.min.css"/>
<link rel="stylesheet" type="text/css" href="css/bootstrap-grid.min.css"/>
<link rel="stylesheet" type="text/css" href="css/index.css"/>
<style>
/* Style the header */
header {
background-color: #666;
padding: 30px;
text-align: center;
font-size: 35px;
color: white;
}
.container.custom-container-width {
max-width: 650px;
}
</style>
<!--
-->
</head>
<body>
<header>
<h1 align="center">Code-switching Sentence Generation</h1>
<h2 align="center">by Generative Adversarial Networks and its Application to Data Augmentation</h3>
</header>
<ul class="nav nav-tabs justify-content-center">
<li class="nav-item">
<a class="nav-link active" href="/Code-Switching-Sentence-Generation-by-GAN/" >
Home</a>
</li>
<li class="nav-item">
<a class="nav-link" href="/Code-Switching-Sentence-Generation-by-GAN/csp.html" >
CSP prediction</a>
</li>
<li class="nav-item">
<a class="nav-link" href="/Code-Switching-Sentence-Generation-by-GAN/gen.html" >
CS Generation</a>
</li>
</ul>
<div class="container">
<br>
<!--
<h5 align="center">Ching-Ting Chang, Shun-Po Chuang, Hung-Yi Lee</h5>
<h5 align="center">Ching-Ting Chang</h5>
-->
<h4 align="center"><I>Ching-Ting Chang,
Shun-Po Chuang,</I>
<a href='http://speech.ee.ntu.edu.tw/~tlkagk/'><I>Hung-Yi Lee</I></a></h4>
<h4 align="center">Graduate Institute of Communication Engineering, National Taiwan University</h4>
<h5 align="center"><a href='https://arxiv.org/abs/1811.02356'>Article link</a></h5>
<br>
</div>
<div class="container custom-container-width">
<h4 align="center"><B>ABSTRACT</B></h4>
Code-switching is about dealing with alternative languages in speech or text.
It is partially speaker-depend and domain-related, so completely explaining the phenomenon by linguistic rules is challenging.
Compared to monolingual tasks, insufficient data is an issue for code-switching.
To mitigate the issue without expensive human annotation, we proposed an unsupervised method for code-switching data augmentation.
By utilizing a generative adversarial network, we can generate intra-sentential code-switching sentences from monolingual sentences.
We applied proposed method on two corpora, and the result shows that the generated code-switching sentences improve the performance of code-switching language models.
<br>
<br>
<!--
<h4 align="center"><B>INTRODUCTION</B></h4>
Code-switching (CS) is a phenomenon that two or more languages used within a document or a sentence.
It is widely observed in multicultural areas, or countries where
official language is different from native language.
For example, Taiwanese tend to mix English and Taiwanese Hokkien in their text and speech besides their main language, Mandarin.
In this paper, we focus on improving intra-sentential CS with embedded lexeme and phrase.
Specifically, we only deal with words and phrases that are code-switched within a sentence.
<br>
CS is fundamentally challenging due to lack of data.
Applying linguistic knowledge is a solution to this.
Equivalence Constraint and Functional Head Constraint are used to build a better CS language model.
Because of a large amount of monolingual data, monolingual language models for host and guest languages are learned separately, and then combined with a probabilistic model for switching between the two.
<br>
Because CS is mostly used in spoken language, the most practical way of generating data is labeling CS speech.
However, labeling speech requires hundreds of trained human laborer and hours of tedious work.
An alternative way is to generate CS data from existing monolingual text. Unfortunately, there are no clear rules for predicting code-switching points within a sentence, since each person tends to code-switch in a different manner.
These years, people try to synthesize more code-switching text by the models learned from data.
<br>
We propose a novel CS text generation method, by using generative adversarial network (GAN)
, to generate CS data from monolingual sentences automatically. Our method has the following benefits:
<ul>
<li> It doesn't require any CS and non-CS labeled pairs to train the data generator.
<li> It learns CS rules for data generation implicitly with the help of discriminator.
<li> It is able to augment more data of not only specific domain such as LectureSS but general domain such as SEAME.
</ul>
Generative models have been used to generate CS sentences, but previous work uses generative model to generate the sentences from scratch.
Here the generator learns to modify monolingual sentences into CS sentences.
In this way, the generator can leverage the information from monolingual sentences.
<br>
With CS data augmented by our method, it is possible to solve the problem of sparse training data.
We conduct the experiments on two Mandarin-English code-switching corpora, LectureSS and SEAME, which have very different statistics to show that the proposed approach generalizes well in different cases.
The experimental results show that GAN can generate reasonable code-switching sentences, and the generated code-switching sentences can be used to improve language modeling.
<br>
-->
<br>
<br>
<br>
</div>
</body>
</html>