Building a vietnamese dialog mechanism for v dlg~tabl system

International Journal on Natural Language Computing (IJNLC) Vol. 3, No.1, February 2014
DOI : 10.5121/ijnlc.2014.3104 31
BUILDING A VIETNAMESE DIALOG MECHANISM
FOR V-DLG~TABL SYSTEM
An Hoai Vo and Dang Tuan Nguyen
Faculty of Computer Science, University of Information Technology, Vietnam National
University – Ho Chi Minh City
ABSTRACT
This paper introduces a Vietnamese automatic dialog mechanism which allows the V-DLG~TABL system to
automatically communicate with clients. This dialog mechanism is based on the Question – Answering
engine of V-DLG~TABL system, and composes the following supplementary mechanisms: 1) the mechanism
of choosing the suggested question; 2) the mechanism of managing the conversations and suggesting
information; 3) the mechanism of resolving the questions having anaphora; 4) the scenarios of dialog (in
this stage, there is only one simple scenario defined for the system).
KEYWORDS
Semantics, Automatic Dialog Mechanism, Question – Answering Engine, Vietnamese Language
Processing.
1. INTRODUCTION
In this paper, we focus on introducing the Vietnamese automatic dialog mechanism of our V-
DLG~TABL system. This automatic dialog mechanism allows the system to communicate
effectively with clients, helps them know the needed information about the tablets which they
expect to find.
The V-DLG~TABL system’s automatic dialog architecture composes the following
supplementary mechanisms: 1) the mechanism of choosing the suggested question; 2) the
mechanism of managing the conversations and suggesting information; 3) the mechanism for
resolving the questions having anaphora; 4) the scenarios of dialog (in this first time, we only
developed one simple scenario for the system). This automatic dialog mechanism is used to build
the component “V-DLG~TABL Dialog” of the system.
In addition, the automatic dialog mechanism of our system bases on an important engine of
answering Vietnamese questions about tablets. This Vietnamese language based Question –
Answering engine is built by applying the theoretic methods and logic programming techniques
of computational semantics proposed by Patrick Blackburn and Johan Bos [1], and by reusing and
referring to some basic implementations (including technical solutions and program source codes)
of Nguyễn Thành and Phạm Minh Tiến [2].
2. USING SIMPLE LAMBDA EXPRESSION FOR REPRESENTING THE
SEMANTICS OF VIETNAMESE QUESTIONS
To represent and analyze the semantics of Vietnamese questions, we apply the approaches,
methods and implementation techniques which are proposed in [1], [2]. According to [1], the

32
semantics of sentences, including questions, can be represented by lambda expressions (cf. [1],
[2]):
 Following [1], each unidentified part of a lambda expression is symbolized by a variable
going after an operator λ, and the operator @ is used to assign an argument to a variable
in a lambda expression (cf. [1], [2]).
 Following [1], the β-transformation uses operator @ to combine the lambda expressions
of syntactic elements with each others. We use the algorithm proposed in [1], which is
based on β transformation, to combine the semantic components of the Vietnamese
question (cf. [1], [2]).
Based on [1], we propose a semantic analysis algorithm for Vietnamese questions in V-
DLG~TABL system as follows:
ALGORITHM:
1) Step 1: Remove the supplementary syntactic elements in the syntactic tree of the
question. The principal syntactic elements are retained: n_tablet, pn_tablet_name,
n_component_<name>, n_property_<name>, and “literal”.
2) Step 2: Identify a lambda expression corresponding to each syntactic element which is
retained after removing the supplementary syntactic elements.
Based on the proposed methods in [1] and their applications in [2], the lambda
expressions corresponding to the retained syntactic elements are defined as follows (cf.
[1], [2]):
Syntactic element Lambda expression
pn_tablet_name λx.x@<tablet_name>
n_component_<name> λx.component_name(x)
n_property_<name> λx.λy.property_name(x, y)
literal λx.x@<literal>
3) Step 3: Apply β-transformation [1] to combine the lambda expressions of syntactic
elements with each other (cf. [1], [2]). The complete FOL (First-Order Logic)
expression can represent the semantics of Vietnamese question.
Example 1: “Máy tính bảng Nexus 7 có kích thước màn hình là bao nhiêu?”
(English: “How much Nexus 7 tablet screen size is”)
The system resolves this question in the following steps:
 Step 1: The syntactic tree of the question in example 1 is presented in Figure 1.

33
Figure 1: The syntactic tree of the Vietnamese question in example 1
 Step 2 (a): Remove the supplementary syntactic elements in the syntactic tree. The
syntactic tree becomes as in Figure 2.
Figure 2: The syntactic tree of the Vietnamese question in example 1 (after removing the
supplementary syntactic elements)
 Step 2 (b): Determine the lambda expressions corresponding to each retained syntactic
element:
+ The lambda expression for the syntactic element n_property_size:
λx.λy.kích_thước(x, y).
+ The lambda expression for the syntactic element n_component_screen:
λx.màn_hình(x).
+ The lambda expression for the syntactic element pn_tablet: λx.x@Nexus_7.

34
Figure 3: The lambda expressions corresponding to retained syntactic elements of the Vietnamese
question in example 1
 Step 3: The lambda expressions representing retained syntactic elements will be
combined by using β transformation, in the following steps:
- Combine ingredients lambda expressions to create a new lambda expression:
[λt.t@Nexus_7]@[λx.λy.kích_thước(x, y)]@[λz.màn_hình(z)]
- Apply β transformation for ingredients lambda expressions to create a completed lambda
expression:
[λt.t@Nexus_7]@([λx.λy.kích_thước(x, y)]@[λz.màn_hình(z)])
= [λx.x@Nexus_7]@[λy.kích_thước(λz.màn_hình(z), y)]
= [λx.x@Nexus_7]@[λz.λy.kích_thước(màn_hình(z) , y)]
Figure 4: Each step for applying β transformation to create a completed lambda expression of the
Vietnamese question in example 1
- Continue to apply β transformation for the other side of the lambda expression:

35
= [λz.λy.kích_thước(màn_hình(z), y)]@Nexus_7
= [λy.kích_thước(màn_hình(Nexus_7), y)]
= λy.kích_thước(màn_hình(Nexus_7), y)
Figure 5: The completed lambda expression of the Vietnamese question in example 1 (after
applying β transformation)
- Finally, the completed lambda expression of the question in example 1 is:
λy.kích_thước(màn_hình(Nexus_7), y). In this lambda expression, y is the unidentified
variable, and represents the value of the screen size asked in the question.
3. ANSWERING VIETNAMESE QUESTIONS ABOUT TABLETS
To answer Vietnamese questions in V-DLG~TABL system, we need to define the data to
describe the information about the tablets. Based on [1], we apply the methods and techniques of
Nguyễn Thành and Phạm Minh Tiến [2] to define descriptive facts and organize the modules in
Prolog.
When a Vietnamese input question is entered into the system, the component “V-DLG~TABL
Syntactic Parsing and Processing” will analyze it to return its syntactic tree. The retained
syntactic elements of the syntactic tree, which are determined by our algorithm mentioned above,
will correspond to the components of the “information structure model” proposed in [2]. These
retained syntactic elements are then processed by the component “V-DLG~TABL Semantic
Analyzing” to determine the lambda expression representing the semantics of the Vietnamese
question. To create a lambda expression for each Vietnamese question, we use the methods and
implementation techniques proposed in [1] (cf. [2]).
For example, the lambda expression of the question in example 1 is λy.kích_thước
(màn_hình(nexus_7), y). In this lambda expression, variable y indicates the unidentified
element in the question.
Basing on the methods and techniques of implementation proposed in [1], and referring to their
applications presented in [2], the system uses the β-transformation to reduce the lambda
expression of the Vietnamese question, and acquires the complete FOL expression; then, this
FOL expression is used to query the database of facts in Prolog and returns the found result.
Following the techniques proposed in [1] (these techniques have been applied in [2], [3], [4], [5]),

36
the querying mechanism is implemented by using the unification between the FOL expression of
the question and the facts of the system (cf. [1], [2]).
Example 2: Based on [1], [2] the lambda expression λy.kích_thước(màn_hình(nexus_7),
y) of the Vietnamese question in example 1 will be unified with the fact
kich_thuoc(man_hinh(nexus_7), ‘7 inch’)) in the database of facts. The unification result
of y is “7 inch”. The value of y is the screen size of the tablet nexus_7.
To create Vietnamese answer in V-DLG~TABL system, we apply the method which is developed
in [2] (the original idea of this approach was proposed in [7], [9]): based on the unidentified node
in the syntactic tree of the question, the system queries the database of facts to find the data, and
creates the answer.
Example 3: The system performs the following steps to resolve Vietnamese question in
example 1:
 Step 1: The syntactic structure of the question is presented in Figure 6.
Figure 6: The syntactic tree of the Vietnamese question in example 1
 Step 2: Query and take the result: “7 inch”.
 Step 3: Find the value which answers the interrogative node in the syntactic tree: see
Figure 7.

37
Figure 7: The interrogative information is answered in the syntactic tree of the Vietnamese
question in example 1
 Step 4: Create the answer based on combining the information components in the
syntactic tree, from left to right: “Máy tính bảng Nexus 7 có kích thước màn hình 7 inch”.
4. BUILDING DIALOG MECHANISM OF SYSTEM
The dialog mechanism of V-DLG~TABL system is built basing on the following mechanisms:
- The mechanism for choosing the suggested question.
- The mechanism for managing the dialog and suggesting the information.
- The mechanism for resolving the questions having anaphora.
- The sample scenarios.
4.1. The mechanism for choosing the suggested question
The mechanism of choosing the suggested question is based on some predefined samples of
suggested questions. Depending on the communication scenario between client and the system,
the system will create the appropriate suggested questions.
- Type 1: These are the suggested questions about other information of the tablet which the
client has just been interested in.
For example, as soon as the client asks the information about screen, size, and
weight of a tablet, the system suggests a question about its color.
- Type 2: These are the suggested questions about the tablets which similar to the one
which client has just been interested in.
For example, after the client asks about tablet Nexus 7 having screen size ‘7
inch’, the system can suggests some more different tablets having the same
screen size.
- Type 3: These are the suggested question about other information when the above cases
do not occur. These questions can be: “Bạn có muốn biết thêm thông tin nào không?”
(English: “Would you like to know any information more?”), or “Bạn có muốn biết thêm

38
thông tin về máy tính bảng nào khác không?” (English: “Would you like to know any
information more about another tablet?”), ...
Table 1 presents the samples of suggested questions which are defined in V-DLG~TABL system.
Table 1: The samples of suggested questions
Suggested question types Elementary questions
Type 1 Question 1.1: <component>
Question 1.2: <component> <tablet>
Question 1.3: <property>, <component>, <tablet>
Question 1.4: <component>, <tablet >
Type 2 Question 2.1: <tablet>
Question 2.2: <same_tablet>
Type 3 Question 3.1: <what_tablet>
Question 3.2: <what_component>
Question 3.3: <what_info>
Table 2 presents some examples for the suggested question samples.
Table 2: Some examples for the suggested question samples
Suggested question types Elementary questions
Type 1 Question 1.1: Bạn có muốn biết thông tin <thành phần> không?
Question 1.2: Bạn có muốn biết thành phần <thành phần> của
máy tính bảng trên không?
Question 1.3: Bạn cần biết thông tin <thuôc tính> của <thành
phần> máy tỉnh bảng <X> không?
Question 1.4: Bạn có biết thông tin về <thành phần> máy tỉnh
bảng <X> không?
Type 2 Question 2.1: Bạn có muốn biết thêm thông tin máy tính bảng
<X> hay không?
Question 2.2: Bạn có muốn biết thêm thông tin các máy tính bảng
cùng loại không?
Type 3 Question 3.1: Bạn có muốn biết thêm thông tin máy tính bảng
nào không?
Question 3.2: Bạn có muốn biết thêm thông tin các thành phần
nào khác không?
Question 3.3: Bạn có muốn biết thêm thông tin nào không?
4.2. Managing the conversations and suggested information
The mechanism of managing the conversations and suggested information is integrated into the
Question – Answering engine, and based on the following tasks:
- Management of time: operate based on managing downtime between each question and
answer, and between user and system.
- Auto generation of suggested question: after identifying the suggested question type, the
suggested question is created based on the suggested question samples.

39
- The mechanism for managing the dialog: This mechanism is integrated into the Question
– Answering engine, implemented some basis tasks:
+ Saving to communication information between the client and the system.
+ Managing time with the component “Manage time” and giving the suggestion
with the component “Auto-generate the suggested question”.
4.3. Revolving the questions having anaphora
In a series of Vietnamese questions and answers following in a dialog, usually there are the
existences of anaphora. According to [8] and [9], the anaphora can be omitted, or can refer to a
queried object in previous questions.
Based on [8] and [9], we propose a method for treating the anaphora problems existing in
Vietnamese questions about tablets.
4.3.1. Anaphora in questions about tablets/components
Example 4: Consider following dialog:
User: Nexus 7 sử dụng loại màn hình gì?
(English: What type of screen does Nexus 7 use?)
System: Nexus 7 sử dụng loại màn hình WXGA.
(English: Nexus 7 uses WXGA screen)
User: Máy tính bảng đó có kích thước màn hình bao nhiêu?
(English: What screen size of the tablet is?)
In the above example, “máy tính bảng đó” at the second question of the client refers to the same
object with the “Nexus 7” at the previous question.
4.3.2. Anaphora referring to queried tablets/components in previous questions
User: Máy tính bảng nào có chức năng đàm thoại?
(English: Which tablets have conversation function?)
System: Nexus 7 có chức năng đàm thoại.
(English: Nexus 7 has conversation function)
User: Máy tính bảng đó có kích thước màn hình bao nhiêu?
(English: What screen size of the tablet is?)
In the above example, the information about the tablet at the second question of the client can be
identified via the queried information in the first question: this anaphora can only be identified
after answering the previous question.
4.3.3. Elliptical anaphora in questions about tablets/components

40
User: Có kích thước màn hình bao nhiêu?
(English: What screen size is?)
In the above example, the second question of the client does not contain any information about
the tablet although this is mandatory information of the question. Therefore, this information is
omitted. The system can identify it basing on the previous question.
4.3.4. Resolution of anaphora in V-DLG~TABL system
The anaphora appearing in syntactic tree of question is processed by an algorithm which we
propose based on the information in Table 3.
Table 3: Anaphoric nodes in V-DLG~TABL system
Anaphora types Anaphoric nodes Descriptive words
Type 1 this_tablet, that_tablet “máy tính bảng/tablet” + “này / đó /
trên”Type 2
Type 3 na_tablet Representing the omitted information
in question
We propose an algorithm for resolving the anaphora existing in Vietnamese questions about
tablets:
Algorithm:
Input: The syntactic structure of Vietnamese question.
Step 1: Searching anaphoric nodes based on the listed cases in Table 3.
Step 2: For each anaphoric node, searching a node corresponds to the anaphoric node in the
syntactic structure of the question.
- Type 1: The corresponding node of an anaphoric node in the current question is the
node of noun phrase describing tablet or components in the previous question.
- Type 2: The corresponding node of an anaphoric node in the current question is a
queried node needing to be answered in the previous question.
- Type 3: The corresponding node of an omitted anaphoric node in the current question is
node of noun phrase describing tablet or components in the previous question.
Step 3: Replacing an anaphoric node with corresponding node in syntactic structure of the
question.

41
4.4. Building the scenario
To set up and build the dialog system, we build a scenario based on mechanism of choosing the
suggested question, mechanism of managing the dialog, and mechanisms of resolving anaphora.
Example 7: A dialog follows the scenario:
User: Có kích thước màn hình bao nhiêu?
(English: What screen size is?)
System: Có kích thước màn hình 7 inch
(English: the screen size is 7 inches)
…20s
System: Bạn có muốn biết thông tin bộ nhớ không?
(English: Would you like to know the information of its memory )
User: Có
(English: Yes)
System: Bộ nhớ trong là 115 MB.
(English: Memory is 115 MB)
…20s
System: Bạn có muốn biết thêm thông tin nào không?
(English: Would you like to know any other information more?)
User: Không
(English: No)
5. CONCLUSIONS
This paper presents the model of Vietnamese automatic dialog mechanism that we propose for V-
DLG~TABL system. This dialog mechanism bases on supplemental mechanisms: choosing the
suggested questions, managing the conversations and suggesting information, revolving the
anaphora in conversion, and Question – Answering engine. Among these supplemental
mechanisms, the engine of answering Vietnamese questions, which is built based on [1] and [2],
is an important element of our dialog mechanism.
We have implemented this dialog mechanism to build V-DLG~TABL system. We have tested the
system according to the defined scenario. Although at the current time, the system only operates
according to one simple dialog scenario, but this scenario is quite general to cover some simple
communication of the user with the system. However, we see that need to developer the other
scenarios for the system.
ACKNOWLEDGEMENTS
This research is funded by University of Information Technology, Vietnam National University –
Ho Chi Minh City (VNU-HCM) under grant number C2011CTTT-06.

42
REFERENCES
[1] Patrick Blackburn, Johan Bos, Representation and Inference for Natural Language: A First
Course in Computational Semantics, September 3, 1999.
[2] Nguyễn Thành, Phạm Minh Tiến, "Xây dựng cơ chế hỏi đáp tiếng Việt cho hệ thống tìm kiếm sản
phẩm máy tính bảng", B.Sc. Thesis in Computer Science, University of Information Technology,
Vietnam National University – Ho Chi Minh City, 2012.
[3] Vương Đức Hiền, “Xây dựng công cụ truy vấn tiếng Việt về các phần mềm máy tính”, B.Sc.
Thesis in Computer Science, University of Information Technology, Vietnam National University
– Ho Chi Minh City, 2013.
[4] Phạm Thế Sơn, Hồ Quốc Thịnh, "Mô hình ngữ nghĩa cho câu trần thuật và câu hỏi tiếng Việt trong
hệ thống vấn đáp kiến thức lịch sử Việt Nam", B.Sc. Thesis in Computer Science, University of
Information Technology, Vietnam National University – Ho Chi Minh City, 2012.
[5] Lâm Thanh Cường, Huỳnh Ngọc Khuê, "Biểu diễn và xử lý ngữ nghĩa dựa trên FOL (First Order
Logic) cho các dạng câu đơn tiếng Việt trong hệ thống hỏi đáp kiến thức xã hội", B.Sc. Thesis in
Computer Science, University of Information Technology, Vietnam National University – Ho Chi
Minh City, 2012.
[6] Son The Pham and Dang Tuan Nguyen, "Processing Vietnamese News Titles to Answer Relative
Questions in VNEWSQA/ICT System", International Journal on Natural Language Computing
(IJNLC), Vol. 2, No. 6, December 2013, pp. 39-51. ISSN: 2278 - 1307 [Online]; 2319 - 4111
[Print].
[7] Dang Tuan Nguyen, An Hoai Vo, Phuc Tri Nguyen, “A Semantic Approach to Answer
Vietnamese Questions in OpenCourseWare Retrieval System”, Proceedings of the 2011
International Conference on Software Technology and Engineering (ICSTE 2011), August 12-13,
2011, Kuala Lumpur, Malaysia, pp. 314-320.
[8] Dang Tuan Nguyen, An Hoai Vo, Phuc Tri Nguyen, "Answering a Series of Vietnamese Questions
in Library Retrieval System", Proceedings of the 2011 2nd International Conference on Future
Information Technology (ICFIT 2011), September 16-18, 2011, Singapore, pp. 445-449. ISBN:
978-981-08-9916-5. IPCSIT, Vol. 13, 2011.
[9] Võ Hoài An, Nguyễn Trí Phúc, “Mô hình xử lý cú pháp và ngữ nghĩa cho các chuỗi câu hỏi tiếng
Việt trong hệ thống tìm kiếm thư viện”, B.Sc. Thesis in Computer Science, University of
Information Technology, Vietnam National University – Ho Chi Minh City, 2011.

Building a vietnamese dialog mechanism for v dlg~tabl system

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Building a vietnamese dialog mechanism for v dlg~tabl system (9)

Recently uploaded (20)

Building a vietnamese dialog mechanism for v dlg~tabl system