Classification of serialverb constructions

International Journal of Artificial Intelligence & Applications (IJAIA) Vol. 6, No. 4, July 2015
DOI : 10.5121/ijaia.2015.6408 79
CLASSIFICATION OF SERIALVERB CONSTRUCTIONS
IN THAI
Surasa Sookgont1
, Thepchai Supnithi2
and Rachada Kongkachandra3
1,3
Department of Computer Science, Faculty of Science and Technology, Thammasat
University, Klongluang, Pathumthani, 12120, THAILAND.
2
Language and Semantic Technology Laboratory, National Electronics and Computer
Technology Center, Klongluang, Patumthani, 12120, THAILAND.
ABSTRACT
The objective of the research is to classify the serial-verb constructions in Thai automatically by using the
word classes from Thai WordNet to classify verbs in the sentence. Due to the Thai language has the extend-
to-the-right structure and put the adjective after the noun. Its overall grammar characteristic is the
"Subject-Verb-Object" or SVO type. And Thai language can be communicated using one verb after another
within the same sentence, that we called "Serial Verb". Today we already have many researches about this
serial-verb constructions, but no research is about its automatic classification.
KEYWORDS
Thai WordNet, Serial Verb Contruction, Automatic Classification
1. INTRODUCTION
Thai language is an isolating language. A Thai sentence is constructed by putting words in
sequence from left to right. The form of each word is not changed according to tense, case, mood,
or voice. Also, Thai words have neither gender nor number. Many words in Thai cannot identify
their static meanings, but have to see their context before interpret. Different from English
language, all adjective used for clarifying noun are placed after noun instead. The syntactical
characteristic of Thai language is the "Subject-Verb-Object". In English language, a sentence
should contain only one main verb, but in Thai many verbs in one sentence could be happened. It
is called the "Serial Verb" such as "ฉันนั่งอานหนังสือ" (I sit reading a book.), "เขาเดินคุยโทรศัพท" (He
walks talking on the phone.), "เขาเด็ดดอกไมดม" (He picks a flower to sniff.), and so on.
In Thai language, a sentence is divided into 2 parts, i.e. subject and predicate. Subject Part is an
important part used to tell the readers/listeners who the subject of the sentence is. Most of the
subject words are the noun and pronoun. Predicate part describes the state or behaviour of its
subject, to tell the readers/listeners which state or behaviour that the subject expresses. Most of
the predicate part is the verb of the sentence that can be either the transitive verb or intransitive
verb.
"Serial verb constructions has been widely interested in Linguistic researches both Thai and other
languages such as Chinese and some languages in East Asia."[1] "The sentence structure has 2
serial verbs or more within the same sentence. These verbs use the same subject, and are put
altogether without any conjunction between verbs. But the sentence may have or not have the

80
noun as the object, such as “เขาตะโกนตอบ” (He shouts answering.), “เขายืนรองเพลง” (He stands
singing.), “เขาถูกรถชน” (He gets hit by the car.), etc.
The examples of researches that study serial verb constructions in Thai are [2], [3], [4], and [5].
Factors relating to properties of serial verb constructions absolutely affect translating structure
which mean translating a language to another couldn’t be perfectly translated in every single
word. However, to translate a language to another even though the language that will be
translated doesn’t have serial verb constructions, after translated better has. Most of the
researches tend to focus on the understanding of the constructions of the sentences, consequences,
and meaning which semantics will be a tool that play an important role to help emphasize the
meaning of the serial verb constructions to be clearer. In general, the conceptual framework to
analyse properties of serial verb constructions in Thai language, most of the researches will study
the unit of serial verb constructions based on syntax and semantics. Most of the researchers have
mutually agreed on these following properties of serial verb constructions [2] (1) No conjunctions
between serial verb constructions (2) Every verb in serial verb constructions is mutual in term of
aspects, timing, rejection. (3) Verbs in serial verb constructions that are mutual will contain at
least one argument (4) Serial Verb Constructions will indicate only a circumstance at a time, but
complicated.
With the fact that serial verb constructions have many aspects in term of views according to the
study of each researcher. The researches which the author have found, most of them hard to apply
linguistically analysis to classify serial verb constructions by programming, except the study
conducted by [5] which clearly classified each verb class which can be applied to develop the
program and the researcher not found any research that automatically classify the serial verb
constructions. Therefore, the author deems the study of Cholthicha [5] can be the beginning of
development useful and help in machine translation between Thai and other languages better.
2. RELATED WORKS
The study related research is intended to gathering ideas and theories about SVCs to guide the
application of this research follows:
2.1. Serial Verb Constructions in Thai
In the study above describes the grammar and meaning of serial-verb constructions that can be
classified to 8 types as below:
2.1.1. Motion Serial Verb Constructions
Consist of the first verb from Motion Verbs such as เดิน (walk), วิ่ง (run), ขับรถ (drive), followed by
Deictic Verbs (ไป (go), มา (come)) or Directional Verbs (such as ไป (go), มา (come), ขึ้น (go up), ลง
(go down), เขา (go in), ออก (go out))
1. กานดา เดิน ออก ไป รองเพลง (Kanda walk exit go sing)
2. การดา เดิน เขา โรงเรียน ไป (Kanda walk enter school go)
2.1.2. Posture Serial Verb Constructions
Consist of the first verb from Postural Verbs such as ยืน (stand), นั่ง (sit), นอน (sleep), followed by
any verb that we call “Open Class Verb”.
3. กานดา ยืน เคาะ ประตู (Kanda stand knock door)
4. กานดา นั่ง รองเพลง (Kanda sit sing)

81
2.2.3. Take-Serial Verb Constructions
Consist of the first verb which must be, “เอา” or take, followed by any verb (Open Class Verb)
which usually the verbs are activity verbs.
5. กานดา เอา ผา ใส ตะกรา (Kanda take cloth put basket)
2.2.4. Use-Serial Verb Constructions
Consist of the first verb which must be, “ใช” or use, followed by any verb (Open Class Verb)
which usually the verbs are activity verbs
6. กานดา ใช มีด หั่น ไก (Kanda use knife cut chicken)
2.2.5. Open Class Serial Verb Constructions
Consist of any verb (Open Class Verb) and verbs in the sentence must not be all the Stative Verb.
The example verbs from Stative Verb are “เชื่อ” or believe, “คิด” or think, “ดีใจ” or glad, etc.
7. กานดา หุง ขาว กิน (Kanda cook rice eat)
2.2.6. Give-Serial Verb Constructions
Consist of the first verb which must be, “ให” or give (or let), followed by any verb (Open Class
Verb.) But this “ให”, or give, cannot be used with the subject or object that is not the living things
because it would be ungrammatical sentence as (9)
8. กานดา ให จุม อาน หนังสือ (Kanda give Jum read book)
9. นวล ทาสี ให บาน (Nuan paint give house)
2.2.7. Causative Serial Verb Constructions
Consist of the first verb which must be, “ทํา” or make, followed by any verb (Open Class Verb)
that is the Intransitive Verb because if the followed verb is not intransitive, it would be
grammatically wrong as (11).
10. กานดา ทํา เด็ก รองไห (Kanda make child cry)
11. กานดา ทํา เด็ก อาน หนังสือ (Kanda make child read book)
2.2.8. Resultative Serial Verb Constructions
Consist of any verb (Open Class Verb) that the second verb in the sentence is the result from the
first verb.
12. กานดา ผลัก วีระ ลม (Kanda push Weera fall)
2.2. Thai WordNet
The researcher use the Thai WordNet (Thai WordNet is a large lexical database of Thai) to assist
in classifying words and group the words altogether. Thai WordNet was constructed in 2007
based on the Princeton’s word network version 3.0 and the two-language dictionary within the
WordNet database. It classifies Thai words into 4 groups: Noun, Verb, Adjective, and Adverb.
Each group is classified into separated subgroups as below:

82
Table 1. Noun class in WordNet
Noun class Meaning
noun.act action activity
noun.animal Animal
noun.artifact things that human created
noun.attribute properties
noun.body body
noun.cognition knowledge and understanding
noun.communication communication
noun.event event or incident
noun.feeling feeling
noun.food food
noun.group group
noun.location location or place
noun.motive motive, objective, reason
noun.object natural phenomenon
noun.person human person
noun.plant plant
noun.possession possession, asset
noun.process process
noun.quantity quantity
noun.relation relationship, cousin
noun.shape shape, form
noun.substance substance, matter
noun.time time
Table 2. Verb class in WordNet
Verb class Meaning
verb.body Verb about maintaining and functioning in each part of the
body
verb.change Verb about changing
verb.cognition Verb about thought
verb.communication Verb about communication
verb.competition Verb about competition
verb.consumption Verb about consumption
verb.contact Verb about contacting
verb.creation Verb about creating or making
verb.emotion Verb about feeling and emotion
verb.motion Verb about motion
verb.perception Verb about perception, classified to 5 senses of perception.
verb.possession Verb about possession
verb.stative Verb that tells the state/condition.
verb.social Verb about behavior in society
verb.weather Verb about weather and climate

83
2.3. Compound Word in Thai
This work studied about the word construction, type of compound word that came from the verb
as the first word to combine.
Compound word is the word that came from 2 words combined together. It may create a new
meaning, or the same meaning, or more concise meaning (Compound word = First word +
Additional word.) It may use the first word as the verb, and additional word as noun, verb, or
adjective. [6][7]
Table 3. Example of compound word with the verb as the first word
Compound word First word Additional word
กระจายเสียง (Broadcast) กระจาย (Distribute) เสียง (Sound)
อุ้มท้อง (Pregnant) อุ้ม (Hold) ท้อง (Pregnant)
3. METHODOLOGY
The researcher has analyzed the word groups compare with the word class in WordNet as below:
3.1. Motion Verb can be compared with verb.motion because the meaning from the WordNet is
the verb about motion.
3.2. Postural verb can be compared with verb.body (verb about function of each body part) such
as นอน (sleep), แตงตัว (dress, put on) and verb.contact (verb about contacting) such as ยืน (stand)
3.3. Stative verb can be compared with verb.stative (verb that tells the state/condition),
verb,emotion (verb that tells the feeling or emotion), verb.possession (verb that tells the
possession), verb.cognition (verb about thought) and verb.perception (verb about 5 senses of
perception) because the Stative verb means the verb that describe the state of something that
cannot visibly express, just to tell or explain the state or condition that happened.
3.4. Activity Verb. The researcher conclude that the word groups that are not in Stative verb can
be included in this group, that are verb.body (verb about maintaining and functioning of each
body part), verb.change (verb about changing), verb.communication (verb about communication),
verb.competition (verb about competition), verb.consumption (verb about consumption),
verb.creation (verb about creating and making), verb.motion (verb about motion), verb.social
(verb about behavior in society) and verb.weather (verb about weather and climate)
Table 4. Show comparable between groups of words with WordNet class.
Word group WordNet Class
Motion verb verb.motion
Postural verb verb.body
verb.contact
Stative verb verb.stative
verb.emotion
verb.possession
verb.cognition
verb.perception

84
Activity verb verb.body
verb.change
verb.communication
verb.competition
verb.consumption
verb.contact
verb.creation
verb.motion
verb.social
verb.weather
3.5. Mapping the [PPRS] from Orchid to noun.person from WordNet because [PPRS] is personal
pronoun
3.6 In case of Give-Serial Verb Constructions that has to use noun as living things (animate
subject), the researcher would use the noun class in WordNet as below:
Table 5. Show comparable between an animate with WordNet class
Noun WordNet Noun Class
Noun in the form of living things noun.animal
noun.person
3.7 Determined that Directional Verb (DV) [1] consists of verbs as: ไป (go), มา (come), ขึ้น (up),
ลง (down), เขา (enter), ออก (exit)
Summary of 8 types of serial verb constructions as below:
Table 6. Eight types of serial verb constructions in Thai
Types of serial verb constructions Verb group
Motion SVCs Motion verb + directional verb
Posture SVCs Postural verb + open class verb
Take-SVCs “เอา” (take) + open class verb (activity verb)
Use-SVCs “ใช” (use) + open class verb (activity verb)
Open class SVCs Open class verb + open class verb
*both verbs must not be the stative verb.
Give-SVCs “ให” (give) + open class verb (not stative verb)
*Required animate subject.
Causative SVCs “ทํา” (make) + open class verb
Resultative SVCs Open class verb + open class verb (result)
3.8 Experiment procedure
Gather the sentences from data collecting and save as the .txt-extension file. Then bring this file
to segment the sentences in words and identify the word function of Part of Speech (POS) with
Orchid. Filter the result to identify the noun, verb, and/or the words that Orchid cannot identify
their POS. Do the additional search from Thai WordNet to identify the word class. After finished
the data preparation, compare the verb group in the sentence with the 8 types of serial verb
constructions.

85
Input
Segmentation &
Coarse POS
Find Word Class Rule-based Analysis Output
Figure 1. Overview of the process
4. EXPERIMENTS
This topic described the result from the experiment that use the type of serial-verb constructions
that linguists classified into 8 types in Table.6 to apply to be the automatic sentence-classifying
program by using the data from Thai WordNet for verb class identification. This experiment use
the sentences from the reference research.
13. เขา@NPRP เดิน@VACT ไป@XVAE รองเพลง@VACT
เขา(noun.person) เดิน(verb.motion) ไป(DV) รองเพลง(verb.communication)
Expect output = Motion SVCs
Actual Output = Motion SVCs
14. เขา@NPRP นอน@VACT อาน@VACT หนังสือ@NCMN
เขา(noun.person) นอน(verb.body) อาน(verb.cognition) หนังสือ(noun.communication)
Expect output = POSTURE SVCs
Actual Output = POSTURE SVCs
15. เขา@NPRP เอา@VACT ผา@NCMN ใส@VACT ตะกรา@CMTR
เขา(noun.person) เอา(take) ผา(noun.artifact) ใส(verb.body) ตะกรา(noun.artifact)
Expect output = Take SVCs
Actual Output = Take SVCs
16. เขา@NPRP ใช@VACT มีด@UNK หั่น@VACT ไก
เขา(noun.person) ใช(use) มีด(noun.artifact) หั่น(verb.contact) ไก(noun.animal)
Expect output = Use SVCs
Actual Output = Use SVCs
17. เขา@NPRP หุง@VACT ขาว@NCMN กิน@VACT
เขา(noun.person) หุง(verb.creation) ขาว(noun.food) กิน(verb.change)
Expect output = Open class SVCs
Actual Output = Open class SVCs
18. เขา@NPRP ให@JSBP ฉัน@NPRP อาน@VACT หนังสือ@NCMN
เขา(noun.person) ให(give) ฉัน(noun.person) อาน(verb.cognition)
หนังสือ(noun.communication)
Expect output = Give SVCs
Actual Output = Give SVCs
19. เขา@NPRP ทา@VACT เด็ก@NCMN รองไห@VACT
เขา(noun.person) ทา(make) เด็ก(noun.person) รองไห(verb.body)
Expect output = Make SVCs
Actual Output = Make SVCs
20. เขา@NPRP ผลัก@VACT ฉัน@NPRP ลม@VSTA
เขา(noun.person) ผลัก(verb.contact) ฉัน(noun.person) ลม(verb.social)
Expect output = Resultative SVCs
Actual Output = Open Class SVCs
5. CONCLUSIONS
During the development of the program for automatically classifying the serial-verb constructions
by using the word functions or POS, and word classes in Thai WordNet to assist, we found that
verb class form between the Open Class SVCs and Resultative SVCs is similar, so the researcher

86
cannot classify. And for the Causative SVCs, the researcher think that we should do additional
analysis about open class verb because of the issues, such as in the form of “ทํา + open class verb”
if the system process the sentence such as “เขาทําเด็กอานหนังสือ”, the result would be Causative
SVCs that is incorrect because the open class verb in Causative SVCs must be intransitive. So the
additional analysis of this issue is recommended.
The author believes that Open Class SVCs, Resultative SVCs, and Causative SVCs are verbs that
currently classified by using open class verb which is any verb. These verbs should be classified
to one of the other verb classes in Thai WordNet in order to replacement of the original pattern,
however, sufficient data of serial verb constructions sentences is essential to classify.
REFERENCES
[1] Wanlee Sutthichatchawanwong, (2006) “A Study of the Translation of Thai Serial Verb Constructions
with Directional Verbs and Their Semantic and Syntactic Equivalence in English” Retrieved June 25,
2012, from http://www.paaljapan.org/resources/proceedings/PAAL11/ pdfs/19.pdf
[2] Thepkanjana Kingkarn, (1986) “Serial Verb Construction in Thai”, Ph.D. Dissertation, the University
of Michigan.
[3] Wilawan Supriya, (1993) “A Reanalysis of so-called serial verb constructions in Thai, Khmer,
Mandarin, and Yoruba”, Ph.D. Dissertation, the University of Hawaii at Manoa.
[4] Nuttanart Muansuwan, (2002) “Direction Serial Verb Constructions in Thai”. Retrieved June 22,
2012, from http://www.stanford.edu/group/cslipublications/cslipublications/HPSG/1/hpsg00
muansuwan.pdf
[5] Cholthicha Sudmuk, (2006) “The Syntax and Semantics of Serial Verb Constructions in Thai”, Ph.D.
Dissertation, the University of Texas at Austin.
[6] http://ict.siit.tu.ac.th/kindml/thainest/index.php?option=com_content&view=article&id=7&Item
id=10. Retrieved September 14, 2012
[7] http://ir.swu.ac.th/xmlui/handle/123456789/956?show=full. Retrieved September 14, 2012
[8] http://wordnet.princeton.edu/man/lexnames.5WN.html#sect4. Retrieved September 14, 2012
[9] Ni Luh Windiari, (2012) “Indonesia Serial Verbs with Directive Verb “PERGI” and their translation
into English”. Retrieved September 14, 2012, from http://www.academia.edu/
1467923/INDONESIAN_SERIAL_VERBS_WITH_DIRECTIVE_VERB_PERGI_AND_THEIR_T
RANSLATION_INTO_ENGLISH
Authors
Surasa Sookgont received a BBA degree from Rajamangala University of Technology Phra
Nakhon North Bangkok Campus, Thailand. She is Master student in Computer Science in
Thammasat University and work as system analyst in Enterprise Business Operation,
ProsperSOF Consulting.
Thepchai Supnithi received the Ph.D. degree in Electrical and Computer Engineering from
Osaka University, Osaka Japan. Currently, he works as supervisor of Language and Semantic
Technology Laboratory (LST), National Electronics and Computer Technology Center
(NECTEC), Thailand. His research interests include Education System, Knowledge
Engineering, Natural Language Processing, and Machine Learning.
Rachada Kongakchandra received the Ph.D. degree in Electrical and Computer Engineering
from King Mongkut’s University of Technology Thonburi, Thailand. Currently, she works
as Assistant Professor at the Computer Science department, Faculty of Science and
Technology, Thammasat University. Her research interests include Artificial Intelligent,
Natural Language Processing, Semantic Processing, and Machine Learning.

Classification of serialverb constructions

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Classification of serialverb constructions (20)

Recently uploaded (20)

Classification of serialverb constructions