Sign Language Generation System Based on Indian Sign

Language Grammar

SUGANDHI, PARTEEK KUMAR, and SANMEET KAUR, Thapar Institute of Engineering and
Technology

Sign Language (SL), also known as gesture-based language, is used by people with hearing loss to convey
their messages. SL interpreters are required for people who do not have the knowledge of SL, but interpreters
are not readily available. Thus, a machine-based translation system is required to translate the text into
SL. In this article, a system is implemented for translating English text into Indian Sign Language (ISL). It
acts as a tool for human-computer interaction and eliminates the need for an ISL human interpreter for
communicating with people who have hearing loss. The system features a rich corpus of English words and
commonly used sentences. It consists of components such as an ISL parser, the Hamburg Notation System, the
Signing Gesture Mark-up Language, and 3D avatar animation for generating SL according to ISL grammar.
The proposed system has been tested rigorously by SL users. The results proved that the proposed system is
highly efficient and achieves an average score of accuracy (i.e., 4.2 for English words and 3.8 for sentences
on a scale from 1 to 5). The performance of proposed system has also been evaluated using the BiLingual
Evaluation Understudy score, which results in 0.95 accuracy. The proposed system and mobile application
together has the potential to bring individuals with hearing loss and their entourage together.
CCS Concepts: • Human-centered computing → Text input; Graphical user interfaces; • Information
systems → Texting; Multilingual and cross-lingual retrieval; • Computing methodologies → Natural
language generation; • Computer systems organization → Real-time languages; • Software and its
engineering → Parsers;
Additional Key Words and Phrases: Sign Language (SL), Indian Sign Language (ISL), hearing loss people,
human-computer interaction, text to sign, communication, speech to sign
ACM Reference format:
Sugandhi, Parteek Kumar, and Sanmeet Kaur. 2020. Sign Language Generation System Based on Indian Sign
Language Grammar. ACM Trans. Asian Low-Resour. Lang. Inf. Process. 19, 4, Article 54 (April 2020), 26 pages.
[Link]

This work was supported by the Department of Science and Technology (DST), Government of India, under research project
“Automatic generation of SL from Hindi Text for communication and education of hearing impaired people” supported
by the Science for Equity, Empowerment and Development (SEED) division with vide number SEED/TIDE/011/2015 to
implement this research work.
Authors’ address: Sugandhi, P. Kumar, and S. Kaur, Thapar Institute of Engineering and Technology, Bhadson Road, Patiala,
Punjab, India, 147004; emails: {ssugandhi_phd16, [Link], [Link]}@[Link].
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee
provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and
54
the full citation on the first page. Copyrights for components of this work owned by others than the author(s) must be
honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists,
requires prior specific permission and/or a fee. Request permissions from permissions@[Link].
© 2020 Copyright held by the owner/author(s). Publication rights licensed to ACM.
2375-4699/2020/04-ART54 $15.00
[Link]

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
54:2 Sugandhi et al.

1 INTRODUCTION
The advancement in Information and Communication Technology (ICT) has affected human life
in all aspects. It has changed the manner in which humans work, manage business, study, travel,
and communicate. For making its efficient use, there is a need to investigate how ICT can be
used to help the hard of hearing community in defeating their communication issues. There are
7.7 billion people in the world,1 of which approximately 70 million people are deaf according
to the World Federation of the Deaf. People with hearing loss developed Sign Language (SL) to
spread information among the community. SL is a complete language, like other languages, and
is widely used for communicating with people with hearing loss. Approximately 200 SLs are used
by the deaf community all over the world.2 It is a gesture-based language, in which hand gestures,
body gestures, and facial expressions are used to convey information. Anything communicated in
spoken language can be easily communicated via gestures. SL is an independent language and is
used all over the world with native differences, similar to spoken languages. SL varies from region
to region—that is, the sign representation of a word might vary in its native SL.
Today, there is not any international standard for SL, and thus to communicate with people
across the globe, one has to learn the native SL. Every region has its own SL interpretation. How-
ever, some signs are universal. For example, the sign for Sun is the same in American Sign Language
(ASL), British Sign Language (BSL), and Indian Sign Language (ISL). The sign for Sun is shown in
Figure 1.3 In India, an individual with hearing loss uses ISL as a communication medium. There are
1.3 million users of ISL in India.4 ISL has its own syntax and grammar just like any other natural
language [1]. The signs used in ISL are classified in two categories: one-handed and two-handed
signs. Further, these are categorized as static and dynamic signs [2]. A sign involving any kind of
movement is known as a dynamic sign, and a sign without movement is known as a static sign.
Every sign in SL is categorized into manual and non-manual signs. Signs that use facial and body
expression are called non-manual signs, and signs that does not consist of these expressions are
called manual signs. Communication with individuals with hearing loss is not possible without a
translator if the person does not have the knowledge of SL. Moreover, the conversation loses its
privacy due to involvement of the SL interpreter, and there are limited ISL experts in India [1].
Hence, there is a need to develop a system that can generate ISL from English text. This would
aid in communicating with individuals with hearing loss. Therefore, building up a system that
interprets English text into its corresponding SL interpretation and gives output in the form of an
animation will significantly affect the hard of hearing community in India.
To abridge this communication gap, plenty of research work has been carried out, including
INGIT [2], English to ISL [1, 9, 11, 12, 14, 17], Hindi to ISL [16], Malayalam to ISL [13], and Punjabi
to ISL generation [15]. However, each one of them has its own limitations, making them incompat-
ible for use in day-to-day life. The majority of systems that have evolved are not flexible and are
confined to a single language, whereas other systems have a limited corpus. Most of the systems
offer only a word-sign search-based approach. In some systems, pre-stored phrases and videos are
used for SL generation. A few systems have domain and environmental restrictions. Thus, research
so far has been less productive for individuals with hearing loss. Therefore, the requirement for
a solution that could generate SL naturally from speech or text was considered, which could im-
prove the lives of people with hearing loss and would help them gain knowledge. Presently, there

1 [Link]
2 [Link]
3 [Link]
4 [Link]

daughters.

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
Sign Language Generation System Based on Indian Sign Language Grammar 54:3

Fig. 1. Sign for Sun in ASL (a), BSL (b), and ISL (c).3

are more than 566 million mobile Internet users,5 and most of the people use the English language
for browsing the Internet in India. Thus, the proposed system has been designed considering Eng-
lish as the input language. A web interface and a mobile app for generation of SL have also been
developed, which is the first of its kind and portable.
In this article, a system is presented through which natural language (i.e., English text) is trans-
lated into 3D ISL animation. To the best of our knowledge, it is the first fully available system
for public usage. For implementing this system, a Hamburg Notation System (HamNoSys)-based
corpus was developed [3]. Moreover, a rule-based parser was also developed to generate ISL based
on its grammar. This system is efficient for people who do not know SL and feel that it is diffi-
cult to communicate with individuals who have hearing loss. The proposed system can be used
at various places, such as banks, post offices, bus stations, railway stations, and airports, for com-
municating and sharing information. It can be used to provide information and will aid in the
exchange of knowledge. Further, it can also be used in schools and colleges for educational pur-
poses. Therefore, the system will play a vital role in bridging the gap and eliminating all of the
obstacles to communicate with individuals who have hearing loss. Moreover, the proposed system
can be accessed anywhere via the web or mobile interface.
The article is organized as follows. A literature review on gesture-based generation systems is
presented in Section 2. ISL grammar and its various aspects are presented in Section 3. Further,
Section 4 highlights the techniques used by the proposed system, and Section 5 discusses the
features of the proposed system. Section 6 describes the HamNoSys-based ISL corpus, and Section 7
highlights the working of system and rules used for translation of natural language into ISL with
example sentences. Section 8 depicts results on the basis of its quality evaluation of the parser and
sign generation, and Section 9 includes the error analysis of the proposed system. A comparative
study of ISL generation systems is depicted in Section 10, and limitations of the proposed system
are discussed in Section 11. Our concluding remarks about the proposed system and future scope
are presented in Section 12.

2 LITERATURE REVIEW
For the past few decades, researchers have been working on the development process of SL gener-
ation systems. Milestones achieved in the research done in this area are discussed in this section.
A speech or text to SL translation system known as ZARDOZ was developed by Veale
and Conway [4]. The system follows an interlingua approach with a blackboard architecture.
The cross-modal machine translation system converts English into native SL, such as Irish,
British, and American SL, using interlingua representation and the Doll Control Language (DCL)

5 [Link]

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
54:4 Sugandhi et al.

animator [4]. It uses various methods to analyze both the verbal input and its corresponding ges-
tural output generated by the system.
Zhao et al. [5] developed Translation from English to ASL by Machine (TEAM). TEAM was the
first translation system for English to 3D animated ASL that not only considered linguistic but also
visual and spatial information associated with ASL. The translation system uses gloss notation and
Synchronous Tree-Adjoining Grammar (STAG) [6] for intermediate representation of English text.
Further, a sign synthesizer is employed for Parallel Transition Networks (PaT-Nets) [7] to achieve
the smooth transitions between signs.
Kar et al. [2] developed the INGIT system for translation of Hindi text to ISL. The system uses
Fluid Construction Grammar (FCG) [2], ellipsis resolution, and the HamNoSys converter for ISL
generation from text. It uses a graphical simulator without facial expressions for playing signs of
ISL. Moreover, use of the proposed system is restricted to the railway domain, where it is used
for making reservations, and hence it has a limited vocabulary of 90 words [2]. Dasgupta et al. [1]
introduced an English text to ISL generation system. The system uses Lexical Functional Grammar
(LFG) [1] and transfers grammar rules for representation of a sentence in ISL syntax. It has a corpus
of 208 sentences and generates signs with an accuracy of 90%, as reported in the literature [1].
Halawani and Zaitun [8] developed a system for translating Arabic speech to Arabic Sign Lan-
guage (ArSL) using the Microsoft speech recognition engine. This avatar-based translation system
uses ArSL grammar. Faraz et al. [9] proposed a translation model for English text to ISL that uses
direct mapping between English words and corpus entries. The system handles inquiries related to
railway reservations. Moreover, the proposed model’s implementation does not follow ISL gram-
mar rules.
Porta et al. [10] developed a translation system for Spanish text to Spanish SL glosses. The sys-
tem uses a rule-based approach for translation. The transfer-based architecture was implemented
from syntactic functions of dependency analyses. The performance of the system was evaluated
with an open domain test bed, and the test results reported a 42% Translation Error Rate (TER) and
a 0.30 BiLingual Evaluation Understudy (BLEU) score [10]. Mishra et al. [11] introduced a transfer-
based system that translates English text to ISL specifically for endocrinologists, in which a doctor
types text that is then parsed and forms a parse tree. This tree is then converted into an ISL parse
tree according to Subject-Object-Verb (SOV) order. Finally, it generates an ISL that is displayed to
patients with hearing loss on a screen [11]. Kaur and Kumar [12] developed a graphical user inter-
face to generate HamNoSys by selecting various symbols. The tool is used for avatar animation of
200 ISL signs. Joy and Balakrishnan [13] developed a prototype for Malayalam to ISL, which takes
Malayalam text as input. It translates the input using POS tagger that performs morphological
analysis and optimization on sentences.
Kaur and Singh [14] developed the HamNoSys of words based on ISL signs. It uses the JA Signing
Gesture Mark-up Language (SiGML) Player app for avatar animation of ISL signs and has a corpus
size of 100 words. Verma and Kaur [15] extended this system for the Punjabi language, which gen-
erates HamNoSys for Punjabi words. This system has been tested on a corpus of 235 words using
the JA SiGML Player app. Vij and Bhatia [16] developed a system that pre-processes Hindi text
and translates it into ISL grammar using a dependency parser and WordNet.6 It uses HamNoSys
and SiGML for 3D animation. Goyal and Goyal [17] implemented an ISL video dictionary using
synthetic animation, which converts written or spoken English sentences into ISL animation. The
system has a dictionary of 1,818 commonly used words. It uses HamNoSys and SiGML for avatar
animation.

6 [Link]

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
Sign Language Generation System Based on Indian Sign Language Grammar 54:5

After performing the literature review, it was observed that very few systems have features such
as 3D animation, multilingual SL generation, and speech to sign. Most of these systems are limited
to a particular domain. However, no real-time system is available online that can be used by the
masses over the Internet. Moreover, no mobile app is available to translate the English language
into ISL, which provides us with the motivation to develop a system for generation of SL on the
basis of ISL grammar. The ISL grammar and its major features are discussed in the next section.

3 ISL GRAMMAR
ISL has its own grammar that is similar to other spoken languages. The whole grammaticalization
process used by ISL uses a layered approach. ISL uses different word orders depending on its struc-
ture. Generally, the basic structure of ISL is Time-SOV. Zeshan [18] stated that ISL follows its own
ISL grammar rules. Later, Sinha [19] highlighted various structures and use of ISL grammar. The
linguistic study of ISL indicates that ISL grammar follows the same rules all over India [1].
Each sign in ISL is either coinage or finger spelled. ISL neither uses verbs (which show action
or a state of being e.g., is, am, are) nor uses articles (e.g., a, an, the). The finger spelling is used to
introduce acronyms, abbreviations, clippings, and words of a spoken language for which there is
no sign. ISL does not use any inflections (gerunds, suffixes, or other forms); it uses only root words.
Additionally, no conjunctions, articles, or helping verbs are used in ISL. It has some specific spatial
indices of time to represent the tense. Past and future tense has an orientation of the back toward
the signer and the front toward the line of bilateral symmetry. Most of the ISL lexicon is made up
of compounds [1].
The various characteristics of ISL grammar are explained as follows:

(i) Representation of person: The first person is always addressed as a speaker and the second
person as the addressee. ISL handles the orientation of the signer corresponding to time
and place. The pronoun (whose index is R-expression) and anaphora do not exist in ISL.
Signs for personal names are based either on a spoken language word or an individual’s
physical appearance [19].
(ii) Representation of number: The numbers are categorized as singular or plural. In ISL, sin-
gular is unmarked morphologically and plural is carried out through different strategies
in ISL. Reduplication is the common strategy employed for pluralization. In the case of
an indefinite plural of a noun, a noun sign is reduplicated three times with change of
location [1].
(iii) Representation of gender: Gender in ISL is represented by a sign pointing on the side of
the face [19]. To express the sign for male, it is articulated by a sign marked above the
lip level. Similarly, a fingertip on the nose represents female.
(iv) Representation of tenses: In ISL, all three tenses can be visualized using signs. A spatial
time line is used to represent the past, present, and future tense [18]. Figure 2 depicts the
time line used in ISL. The present tense in ISL is unmarked, whereas for representing
the future or past tense of sentences, a spatial time line representation is used, as shown
in Figure 3.
(v) Representation of aspects: ISL also accommodates different aspects of a sentence through
signs such as formational parameters of the verb, namely movement in terms of tense,
dynamics, shape, and size. The continuative aspect usually represents an event continu-
ity without any interruption over a period of time. In the context of ISL, it is expressed
by lengthening the verb movement either by triplication or by a verb with no other
formational changes [19].

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
54:6 Sugandhi et al.

Fig. 2. Time line used in ISL.

Fig. 3. ISL sign for past (a) and future (b).

For representing the perfect aspect of a sentence, it uses the grammatical form of
finish, which can be attached after projecting the sign for verbs such as drink and eat.
[19]. The perfect aspect is used to show that an event was completed recently, and this
could be done using either a single or double hand.
To represent the recurring (repetitive) nature of any verb or an action in a sentence,
there is usually an assignment of aspect to it, which could be a frequentative or iterative
aspect. A frequentative aspect is one in which the verb occurs for a prolonged period
of time, whereas an iterative aspect is one in which the verb recurs, although for some
fixed and definite time period [19]. For articulating the frequentative aspect, the verb
is slowly showcased with a lengthening triplicate movement followed by a rounding
thrust, whereas the iterative aspect in ISL is represented by the sign for again.
The inchoative aspect indicates the state where someone is about to do something.
To represent the inchoative aspect, complete articulation of the verb is reduced to an
abrupt form [1]. Figure 4(a) depicts that the action of break is initiated and is about to
complete, whereas Figure 4(b) shows the complete closure action for break.
(vi) Representation of relations: To represent relationships such as bhai (brother), beta (son),
patni (wife), and other distinct relations in ISL, a combination of two or more words
are used [1]. For example, bhai is expressed as man followed by the sign for sibling
(i.e., brother). Similarly, beta is expressed as man followed by child, whereas patni is
expressed as woman and married.
(vii) Representation of questions: To represent questions, facial expressions and body gestures
are also used with hand gestures. To express questions sentences, eyebrows are raised

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
Sign Language Generation System Based on Indian Sign Language Grammar 54:7

Fig. 4. ISL sign for break: inchoative aspect (a) and non-inchoative sign (b).

Fig. 5. Types of hand shapes.

[17]. Interrogative terms such as when, what, and how are always placed at the end of a
sentence in ISL [20].
(viii) Representation of negation: Non-manual signs like facial expressions and head move-
ments are used to denote negation in negative sentences (e.g., shaking of the head).
Negation in ISL is represented with the sign for not, which is always placed at the end
of sentence [17].

These aspects of ISL grammar have been taken into consideration by the proposed system for
generation of ISL sentences. The proposed system uses techniques such as HamNoSys, SiGML,
and avatar animation for generation of SL. A description of these techniques is described in the
next section.

4 TECHNIQUES USED FOR GENERATION OF SL

In this proposed system, HamNoSys is used for sign representation, SiGML for its conversion to
an XML file, and an avatar for generation of signs. A description of these components is provided
next.

4.1 HamNoSys
HamNoSys notation was developed by Siegmund Prillwitz in 1984 [3]. It is a standard transcription
that has approximately 200 characters and could be used to represent SL in any region. It includes
manual and non-manual components. HamNoSys is a phonetic notation that consists of different
parameters such as the configuration of hand, movement, location, head, and body gestures. These
parameters are named as hand shape, hand orientation, location, and movement, among others.
For representing a sign, there are various hand shape symbols available in HamNoSys [21]. Figure 5
shows some of the various hand shapes and finger positions used in HamNoSys.

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
54:8 Sugandhi et al.

Fig. 6. ISL sign for alone (a) and its HamNoSys (b).

The hand orientation includes the direction of the fingers and palm. It can be left, right, upward,
inward, downward, or outward depending on the position of the signer/body. The location of a
sign can be defined using location symbols such as above the head, shoulders, forehead, head,
nose, mouth, lips, cheek, ears, neck, chest, stomach, or below the stomach. It also defines the
movement of a hand from the initial location to the destination. HamNoSys includes symbols
used for representing two-handed signs. Figure 6 depicts the ISL sign for the word alone and its
HamNoSys.

4.2 SiGML
SiGML [22] was developed at the University of East Anglia for specifying signing sequences. It is
an XML form of various tags for representing each symbol of HamNoSys in the particular sign
[21]. It starts from the <sigml> tag and ends with a closing tag (i.e., </sigml>). Thereafter, a gloss
attribute is defined that usually refers to the word name for which its SiGML is generated. The
non-manual features are defined with their corresponding tags between <hamnosys_nonmanual>
and </hamnosys_nonmanual> tags. For manual features of a sign, the tags are defined within
<hamnosys_manual> and </hamnosys_manual> tag. Figure 7 shows SiGML for the word alone.

4.3 Avatar
The avatar, also known as a virtual human, was developed by Elliott et al. [22] at the University
of East Anglia. It takes SiGML as input that is generated from HamNoSys notation and produces
the animation frames. When played, these frames generate 3D avatar animation. Figure 8 shows
avatar animation for the word alone.7

5 FEATURES OF PROPOSED SYSTEM

The proposed system has various features that make it more compatible, flexible, and easily acces-
sible to users, such as the following:
• The proposed system has both a web interface and a mobile app. Figure 9 depicts the web
interface of [Link] through which a user can give input in English language
and the ISL output is generated using 3D avatar animation.
• The mobile application Sanket is available in Google Play Store at [Link]
store/apps/details?id=[Link]&hl=en_IN.

7 [Link]

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
Sign Language Generation System Based on Indian Sign Language Grammar 54:9

Fig. 7. SiGML for alone.

Fig. 8. Avatar animation of the word alone using Humanoid (a) and Cartoon character (b).7

• The mobile application Sanket has the feature of speech to text conversion. The user can
also control the avatar animation speed (i.e., he can increase/decrease the speed). The same
is depicted in Figure 10.
• It uses real-time HamNoSys for representing gestures instead of pre-recorded videos, which
provides flexibility in terms of generation of signs.
• It has a special keyboard for creating signs and a module for automatic conversion of Ham-
NoSys to SiGML.
• It has the corpus of 2,950 words and 1,000 sentences. There are many filters provided by the
system that the administrator can easily use to search a specific word in the corpus.
• The system uses WebGL for fast 3D avatar animation and makes it compatible with all
browsers.
• The Avatar Application Programming Interface (API) is also developed, which provides var-
ious functionalities of avatar animation such as playing animation of a word or sentence,
speeding up/down animation, fetching current animation speed, replaying last animation,
releasing lock of avatar, and checking for error for its integration with other applications.

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
54:10 Sugandhi et al.

Fig. 9. Web interface of [Link].

• There can be two types of users for the proposed system: admin or ISL user.
• The system has a admin panel for handling the database through which the administrator
can add, delete, search, and modify the words using various filters. Using this interface, ISL
signs can be easily created, deleted, or updated.
• By using the proposed system, an English textbook from the Punjab School Education Board
(PSEB) has been reproduced in ISL. This would be very effective in the education of students
with hearing loss.

6 HAMNOSYS-BASED ISL CORPUS

The major task for the development of the proposed system was the creation of the HamNoSys-
based ISL dictionary. To fulfill this, an online ISL corpus was developed using HamNoSys notation.
This corpus consists of 2,950 various commonly used English words. For each word, HamNoSys
experts watch ISL videos developed by the Indian Sign Language Research and Training Centre
(ISLRTC), and based on the visual representation of a symbol in the video, its corresponding Ham-
NoSys representation is created, which is stored in the database. Further, SiGML files are created
for each of these HamNoSys representations of English words as the system supports automatic
HamNoSys to SiGML conversion [21]. Hence, the online database stores English words with their
SiGML representations. Moreover, the corpus can also be handled through the admin panel for
addition, deletion, and modification of a word and its corresponding HamNoSys for ISL signs.
Details of the corpus on the basis of each category, such as alphabets, numbers, weekdays, ani-
mals, countries, jobs, technical, academic, medical, and legal terms, among others, are depicted in
Table 1.
In Table 2, information about the ISL corpus on the basis of morphological detail or chunks is
shown. The analysis of each word category on the basis of its appearance in 1,000 sentences is
ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
Sign Language Generation System Based on Indian Sign Language Grammar 54:11

Fig. 10. Mobile interface of the Sanket app.

depicted in Figure 11. It consists of various type of sentences, including declarative, interrogative,
imperative, exclamatory, and negative sentences. The average length of these 1000 sentences is
between 9 to 11 words.

7 SYSTEM ARCHITECTURE
The proposed system generates ISL sentences from English text using ISL grammar as depicted in
Figure 12. It takes English words or sentences as an input, which may even consist of numbers.
The input is then parsed using the parser, which performs a morphological and syntactical anal-
ysis. This results in the formation of root words with their respective morphological information.
Through this information, ISL rules are implemented for each of the input sentences. Using these
rules of ISL grammar, an output is generated and the word order is fetched. For each fetched root
words, the corresponding HamNoSys is obtained from the database. Further, the implicit tool for

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
54:12 Sugandhi et al.

Table 1. Categories of ISL Corpus

S. No. Category HamNoSys-Based Corpus

1 Alphabets 26
2 Numbers 46
3 Animals, Birds, and Insects 56
4 Basic phrases 9
5 Body parts 17
6 Colors 15
7 Countries, States, and Cities 84
8 Family and Relationship terms 41
9 Festivals and Holidays 26
10 Fruits 18
11 Jobs, Professions, and Trades 91
12 Months 12
13 Places, Buildings, and Institutions 24
14 Vegetables 18
15 Weekdays 14
16 Deaf studies 23
17 Maths 45
18 Science 60
19 Technical terms 330
20 Legal terms 21
21 Academic terms 188
22 Medical terms 150
23 Others (from books and other sources) 1,636
Total 2,950

Table 2. Word Frequency in Corpus

S. No Category No. of Words

1 Noun (NN) 2,270
2 Adjective (JJ) 216
3 Noun plural (NNS) 107
4 Adverb (RB) 91
5 Verb (VB) 82
6 Verb gerund/Present participle (VBG) 61
7 Personal pronoun (PRP) 32
8 Coordinating conjunction (CC) 29
9 Preposition (IN) 15
10 Wh-adverb (WRB) 23
11 Others 17
Total 2,950

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
Sign Language Generation System Based on Indian Sign Language Grammar 54:13

Fig. 11. Average occurrence of different words in sentences.

Fig. 12. System architecture of ISL generation from text.

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
54:14 Sugandhi et al.

Fig. 13. Morphological information (a) and phrase tree structure (b) of the example sentence (1).

converting HamNoSys code to a SiGML file is used to obtain the file. This file is stored with the
SiGML file of other words. It is then fed to the animation server whenever some word is queried.
Depending on the tags used in the SiGML file, animation frames are generated, which are played
together to animate the 3D avatar. Figure 12 also depicts the web interface for input of English
sentences and shows the generation of ISL animation by the avatar.
Details of each component for the proposed system are described next.

7.1 Implementation of the ISL Parser

The user inputs the English word or sentence through a web interface or mobile app. This input is
parsed using the parser. In this system, the Stanford parser8 is used to process the input sentences.
The parser analyzes the linguistic structure of that sentence and rearranges words together as
phrases on the basis of information such as, subject, object, and verb. It defines the dependency
level of words. The parser uses a Part-of-Speech (POS) tag to represent its morphological informa-
tion. Based on this information, punctuation is removed through the filter and chunks are created
for further processing. The words are stemmed and lemmatized using the Skyeng Php-lemmatizer
to extract the list of root words.
To understand the workings of the proposed system, let us consider the example sentence given
in (1):
She went to the market. (1)
The Stanford parser used in the system divides the sentence into the chunks with their mor-
phological information as shown in Figure 13(a), and its phrase structure tree is also depicted in

8 [Link]

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
Sign Language Generation System Based on Indian Sign Language Grammar 54:15

Fig. 14. Dependency-level information of the example sentence (1).

Figure 13(b). Moreover, the dependency information generated by the parser is shown in Figure 14.
It also includes lemmatized words of the input sentence given in (1).
After fetching the list of root words, ISL grammar rules are applied for reordering the words in
SOV order as English follows Subject-Verb-Object (SVO) order. As discussed in Section 3, ISL has
its own grammar and syntax. For the development of the ISL parser, a rule-based system has been
framed to handle all aspects of ISL grammar, including person, numbers, gender, questions, and
tense, among others. These rules are described with the help of a flow chart, as shown in Figure 15.
As shown in the flow chart, first the user inputs the English word or sentence. The system then
parses the input and applies filters over it. As a result, the list of root words is obtained using
this information. Thereafter, the system compares the morphological information of obtained root
words with existing ISL grammar rules that are pre-stored in the database. When a named entity
appears or the word is not present in the dictionary, its finger-spell is played. To represent present
participle, its sign is repeated three times. Similarly, possessive pronouns are converted into per-
sonal pronouns, and plural is represented by repeating the sign twice or more. Tense in ISL is
represented using a spatial time line, and gender is denoted with he or she. Thereafter, these root

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
54:16 Sugandhi et al.

Fig. 15. Flow chart of the system.

words are reordered in SOV order. In case the rule for a particular input is not found, its root words
are extracted and its SiGML files are fetched, and it generates ISL signs using avatar animation.
Let us consider the example sentence given in (1); after parsing the input sentence, its morpho-
logical information is extracted and punctuation marks are removed. The input is then divided
into chunks such as She, went, to, the, market. Thereafter, its root words (She, go, to, the, market)
are fetched, and on the basis of its grammatical aspect, the corresponding rule is mapped. After
applying ISL grammar rules, the ISL sentence for the input sentence (1) is generated by removing
the preposition (to) and article (the), followed by reordering of words (i.e., She market go).
The ISL rules are framed for each category of English sentences. It includes various aspects of
ISL grammar that are depicted with a set of example sentences given in Table 3.

7.2 Generation of Sign Animation

For each root word in the ISL sentence, the system extracts its HamNoSys from the database as
shown in Figure 15. Thereafter, the proposed system automatically generates SiGML files from its
HamNoSys. Each SiGML file represents the sign for a word. It consists of different tags used for
sign animation. The SiGML file generated from HamNoSys is taken as input for further processing.
It is accessed by the animation server for generating animation frames. Further, these frames are
used to define the pose of an avatar. Finally, the ISL sign is played through these frames in the
form of 3D avatar animation that is displayed to the user as an output of the proposed system.

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
Sign Language Generation System Based on Indian Sign Language Grammar 54:17

Table 3. Translation from English to an ISL Sentence Based on ISL Rules

For the example sentence given in (1), HamNoSys of the ISL sentence is fetched from the
database—that is, HamNoSys of the root words she, market, and go is extracted as depicted in Fig-
ure 16. Thereafter, SiGML is generated using the HamNoSys to SiGML converter, and the avatar
animation server processes it for ISL generation. The avatar animation of the example sentence
given in (1) is depicted in Figure 16.
A detailed algorithm for rule-based ISL generation that explains the workings of proposed sys-
tem is given in Algorithm. The notations used for ISL generation in Algorithm 1 are depicted in
Table 4.

8 RESULTS
The proposed system has been tested on a corpus of 1,000 commonly used sentences, for which SL
experts manually generate its ISL output based on ISL grammar. To assess the proposed system, the
BLEU score is used, which compares the output generated by SL experts with the output generated

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
54:18 Sugandhi et al.

Fig. 16. ISL generation of a sentence (1).

by the proposed system. Based on the resemblance of system output with the manual output of SL
experts, the BLEU score is measured. The BLEU score always lies between 0 and 1 [10]. It depicts
the similarity between translated and expected output, where a value near 1 means that both
sentences are similar, and a value near 0 means that both sentences are not similar. Henceforth,
the BLEU score as shown in (2) is calculated for each of the translated sentences, by comparing
each of them with reference sentences (i.e., generated by experts). Thereafter, an average of these
scores is calculated to estimate the overall translation quality of the system.
N
BLEU = BP ∗ exp  w n ∗ loдpn  , (2)
n=1 

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
Sign Language Generation System Based on Indian Sign Language Grammar 54:19

Table 4. Notations Used for the ISL Generation Algorithm

Notations Description
S Input sentence list
ISL Final ISL sentence
s Input sentence in list S
w List of words in the ISL sentence
W List of root words fetched from the parser for sentence s
M List of morphological information for sentence s
B List of stop words
MS String containing morphological information of root words
MI Rules stored in the rule database
N Stores the indexing of word order corresponding to rule M I
RW List of root words after reordering
HAM HamNoSys of a word
H List of HamNoSys for root words
l Literals of the word
HAM_LIT HamNoSys of a literal
hamnosys HamNoSys stored in the H list
SIGML SiGML file of HamNoSys
SIG List of SiGML files
siдml SiGML stored in the S I G list
RDB Rule database for ISL generation
HDB HamNoSys database
GRW(sentence) Get root words from the parser of the sentence
GET_MI(sentence) Get morphological information of from the parser for
sentence s
COMPARE(morphological_info,rule) Match the morphological information of the sentence with
rules stored in the RDB database
FILTER(sentence,rootWords, Remove punctuation and apply filters
morphologicalInfo,stopWords)
CONCAT(morphologicalInfoList, Concat morphological information in a string
morphologicalString)
INDEX_ORDER(rule) Extract the indexing of word order corresponding to rule M I
WORD_REORDER(indexOrder, Reorder the root words according to the index order
rootWordsList)
APPEND(rootWords,outputSentence) Append the root words in the output sentence
IS_HAM(word,HamNoSysDatabase) Check if HamNoSys of the word is present in the HDB
FETCH_HAM(word,HamNoSysDatabase) Fetch HamNoSys of the word from the HDB
APPENDHS(string,list) Append string in the list
HAMSIGML_CONVERTER(hamnosys) Convert HamNoSys notations into SiGML
PLAY_ANIMATION(sigml) Play 3D avatar animation of ISL signs from the SiGML file

where BP is the penalty for sentence length, pn is the modified N-gram precision, and wn = 1/N.
 
1, for c > r
BP = 1-r/c where c or r is length of translated and reference sentence
e , for c ≤ r
During processing of the BLEU score, the n-gram of both sentences is compared and it counts
the number of position-independent matches. Higher matches in n-gram ensure a better quality

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
54:20 Sugandhi et al.

ALGORITHM 1: ISL generation from English text

Input: S
Output: ISL
1: Procedure ISL_Generation(S )
2: Set (ISL, NU LL)  Set output sentence ISL to NU LL
3: repeat
for each: s ∈ S
4: W ← GRW (s)  Get the root word from the parser for sentence s
5: M ← GET _MI (s)  Get the morphological information of words from the parser
6: F ILT ER(s,W , M, B)  Remove punctuation and apply filters
7: MS ← CONCAT (M, MS )  Concat root words morphological information into string MS
8: if COMPARE(MS, MI ) = T RU E then  Compare morphological information with the
rule stored in database
9: N = I N DEX _ORDER(MI )  Get the indexing of word order corresponding to rule MI
10: RW = W ORD_REORDER(N ,W )  Reorder word order
11: else
12: RW = W
13: end if
14: ISL ← APPEN D(RW , ISL)  Append final root words list RW into an ISL sentence
15: until s! = NU LL (Sentence s is not empty)
16: repeat
for each: w ∈ ISL
17: if IS_HAM (w, H DB) = T RU E then  HamNoSys of a word found in the HamNoSys
database
18: HAM ← F ETCH _HAM (w, H DB)
19: H ← APPEN D(HAM, H )  Append Hamnosys in a HamNoSys list
20: else
21: repeat
for each: l ∈   Break word into list of literals l
22: HAM_LIT ← F ETCH _HAM (l, H DB)  Fetch HamNoSys of each literal
23: H ← APPEN DHS (HAM_LIT , H )
24: until l! = NU LL (List of literal is not empty)
25: end if
26: until w! = NU LL (w is not Null)
27: repeat
for each: hamnosys ∈ H
28: SIGML ← HAMSIGML_CONV ERT ER(hamnosys)  Get SiGML file from HamNoSys
29: SIG ← APPEN DHS (SIGML, SIG)  Append SiGML file in list SIG
30: until hamnosys! = NU LL (HamNoSys list is not empty)
31: repeat
for each: siдml ∈ SIG
32: PLAY _AN I MAT ION (siдml )  Play avatar animation of ISL sign using SiGML
33: until siдml! = NU LL (SIGML list is not empty)
34: End Procedure

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
Sign Language Generation System Based on Indian Sign Language Grammar 54:21

Fig. 17. Score of words.

of translated sentences. The proposed system has achieved a BLEU score of 0.95 for a corpus of
1,000 sentences.
To evaluate the quality of signs, a manual evaluation is also performed in which 2,950 words
had been tested by six SL experts. During manual testing, an avatar animation is played for each
of the words from this corpus, and based on their accuracy or resemblance with the real sign, ISL
experts assigned the score to them [23]. This evaluation is performed on a scale from 1 to 5, where
1 denotes very poor performance, 2 refers to poor, 3 is average, 4 denotes good, and 5 is the best
performance. As depicted in Figure 17, the maximum number of words (i.e., 1,594) was awarded
the highest score, and only 200 words were assigned with a low score of 1 or 2.
Apart from words, the proposed system has also been evaluated for testing the accuracy of sen-
tences. About 1,000 sentences were manually tested based on their output generated by the system.
As shown in Figure 18, there are 694 sentences that can be considered as accurately generated, as
their score lies between 4 and 5 (i.e., good and excellent performance). The weighted mean as given
in (3) is also measured for the scores of words and sentences. From the result as depicted in Fig-
ure 19, the weighted mean for the score of words is 4.2, whereas the weighted mean for sentences
is 3.8.
n
wi ∗ xi
x= (where n = 5), (3)
i=1
wi

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
54:22 Sugandhi et al.

Fig. 18. Scores of sentences.

where x is the weighted mean of the score, x i refers to the score, and w i refers to the number of
words or sentences lying under x i .
The results of the proposed system are very encouraging, and the proposed system has potential
for use in areas such as education, railway reservations, banking systems, and airports.

9 ERROR ANALYSIS
It has been observed that some signs in ISL are very complex, and to depict these sign via the pro-
posed system, a lot of animation is required. In addition, HamNoSys consists of limited notations.
Thus, the proposed system does not showcase a swift animation in the case of complex words and
is also devoid of facial expressions that are usually portrayed by human sign interpreters. Another
point worth mentioning is the parser dependency of the proposed system. Sometimes the proposed
system’s parser outputs the wrong morphological information of a sentence, which further leads
to wrong mapping with ISL rules, and ultimately the final output is a wrong sentence that does
not affirm ISL rules. Presently, the system is not capable of handling complex sentences.

10 COMPARATIVE STUDY OF SL GENERATION SYSTEMS

The comparative analysis on the basis of different parameters has been done between the existing
SL generation system and the proposed system as depicted in Table 5.

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
Sign Language Generation System Based on Indian Sign Language Grammar 54:23

Fig. 19. Average score of words and sentences.

From the analysis, it has been concluded that there is no system available for desktop and mobile
devices for real-time communication with people with hearing loss. The existing systems do not
have the functionality of the speech to ISL generation system. Moreover, the existing systems have
a limited corpus and are bounded to a particular domain only.
To the best of our knowledge, this system is the first of its kind available for ISL generation. The
main features of proposed system that make this system unique are a rich vocabulary database,
an admin panel to handle the database, a HamNoSys to SiGML conversion module, 3D avatar
animation, and speech to ISL. The proposed system has a scalable architecture, and a corresponding
API has been developed to extend it to other languages.

11 LIMITATIONS OF THE STUDY

The proposed system has coverage of almost 3,000 words, but some signs are not present in the
system. Thus, vocabulary size is one limitation of this system. Therefore, to extend its coverage,
continuously building our own HamNoSys database will help. The system supports ISL rules for
simple sentences and sometimes for complex sentences, it failed to generate accurate ISL sen-
tences. The proposed system also lacks in representing complex animation and facial expressions
of ISL signs, as there are limited facial/body expressions in HamNoSys notation for representing
non-manual signs. Moreover, in ISL, there are different criteria for handling directional signs
based on its context. Hence, depending on the direction, there can be a change in the position of
a sign from beginning to end than its usual sign. The proposed system does not handle indexing
of signs based on the context. Currently, the proposed system has the limitation of word-sense
disambiguation, which may lead to incorrect word sign animation. There is no standard bench-

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
 54:24

Table 5. Comparative Analysis of Existing and Proposed ISL System

Undertaking
S. No. Name Year Corpus Size Technique Domain Speech to Text Access Type URL Mobile App Accuracy
1 ZARDOZ: English to ASL 1994 Not available Blackboard architecture General No None Not available No Not mentioned
2 TEAM: English to ASL 2000 Not available Uses ASL grammar rules General No None Not available No Not mentioned
3 INGIT: Hindi to ISL 2007 90 words and Hybrid-Formulaic Railway No None Not available No 80%
230 sentences grammar approach and reservations
Pre-recorded videos
4 English to ISL 2008 208 sentences Transfer Grammar Rule, General No None Not available No 90%
Lexical Functional
Grammar
5 Arabic to ArSL 2012 29 Arabic Rule-based approach General Yes None Not available No Not mentioned
alphabets and
words
6 English to ISL 2013 Not available Corpus-based MT Railway No None Not available No Not mentioned
reservations
7 Spanish to LSE 2014 Not available Rule-based approach General No None Not available No 42% TER and 0.30
BLEU
8 Malayalam to ISL 2014 Not available Morphological Analyzer General No None Not available No Not mentioned
and Optimizer
9 English to ISL 2014 Not available Transfer-based system No None Not available No Not mentioned
Endocrinologists
10 Punjabi to ISL 2015 235 words JA SiGML Player General No None Not available No Not mentioned
11 English to ISL 2016 1,818 words Probabilistic Context General No Private Not available No Not mentioned
Free Grammar (PCFG),
3D avatar animation
12 Proposed System 2018 2,950 words and Probabilistic Context General Yes Public [Link] Yes 0.95 BLEU
1,000 sentences Free Grammar (PCFG) [Link]
and Dependency parser,
Rule-based approach, 3D
Avatar animation

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
Sugandhi et al.
Sign Language Generation System Based on Indian Sign Language Grammar 54:25

mark available for SL generation system evaluation. Thus, evaluation of the proposed system has
been based on the manual review performed by ISL experts and users of the system.

12 CONCLUSION AND FUTURE SCOPE

This article proposes a system that translates English text into its corresponding ISL animation.
The system consists of an important component (i.e., an ISL parser) that is being used to parse
sentences based on ISL grammar rules. With a corpus of 2,950 English words and 1,000 sentences,
the proposed system could be used efficiently for communicating with individuals with hearing
loss. The system outputs the ISL representation of a sentence or word using 3D avatar animation
in real time rather than using pre-stored pictures or humanoid videos. The system highlighted in
the study can be experienced by visiting [Link] or on the Android app Sanket,
which is free to download from the Google Play Store, available at [Link]
apps/details?id=[Link]&hl=en_IN.
In the future, coverage of the HamNoSys-based dictionary will be extended. To make the system
more robust, some additional ISL rules will be added to handle complex sentences. Further, word-
sense disambiguation, animation for complex signs, and sign indexing will be included to enrich
the systems’ functionalities. However, the results and feedback received from ISL users are very
positive and encouraging, as the proposed system will help people with hearing loss get in touch
with their entourage.

ACKNOWLEDGMENTS
We thank Mr. Aditya Ghodgaonkar and Mr. Divanshu Singh for their support in carrying out this
project.

REFERENCES
[1] Tirthankar Dasgupta, Sandipan Dandpat, and Anupam Basu. 2008. Prototype machine translation system from text-
to-Indian sign language. In Proceedings of the IJCNLP-08 Workshop on NLP for Less Privileged Languages.
[2] Purushottam Kar, Madhusudan Reddy, A. Mukherjee, and Achla M. Raina. 2007. INGIT: Limited domain formulaic
translation from Hindi strings to Indian sign language. ICON 52 (2007), 53–54.
[3] Thomas Hanke. 2004. HamNoSys—Representing sign language data in language resources and language processing
contexts. In Proceedings of the 4th International Conference on Language Resources and Evaluation (LREC’04), Vol. 4.
1–6.
[4] Tony Veale and Alan Conway. 1994. Cross modal comprehension in ZARDOZ an English to sign-language translation
system. In Proceedings of the 7th International Workshop on Natural Language Generation. 249–252.
[5] Liwei Zhao, Karin Kipper, William Schuler, Christian Vogler, Norman Badler, and Martha Palmer. 2000. A machine
translation system from English to American Sign Language. In Proceedings of the Conference of the Association for
Machine Translation in the Americas. 54–67.
[6] Stuart M. Shieber. 1994. Restricting the weak-generative capacity of synchronous tree-adjoining grammars. Compu-
tational Intelligence 10, 4 (1994), 371–385. DOI:[Link]
[7] Norman I. Badler, Martha S. Palmer, and Rama Bindiganavale. 1999. Animation control for real-time virtual humans.
Communications of the ACM 42, 8 (1999), 64–73.
[8] Sami M. Halawani and A. B. Zaitun. 2012. An avatar based translation system from Arabic speech to Arabic sign
language for deaf people. International Journal of Information Science and Education 2 (2012), 13–20.
[9] Syed Faraz Ali, Gouri Sankar Mishra, and Ashok Kumar Sahoo. 2013. Domain bounded English to Indian sign language
translation model. International Journal of Computer Science and Informatics 3, 1 (2013), 41–45.
[10] Jordi Porta, Fernando Lopez-Colino, Javier Tejedor, and Jose Colas. 2014. A rule-based translation from written
Spanish to Spanish sign language glosses. Computer Speech and Language 28, 3 (May 2014), 788–811. DOI:https://
[Link]/10.1016/[Link].2013.10.003
[11] Jagriti Mishra, Gouri Sankar Mishra, Kiran Ravulakollu, Ravi Rastogi, and K. M. Rafi. 2014. Machine translation of
Indian signs for endocrinologist. International Journal of Emerging Technology and Advanced Engineering 4, 4 (2014),
112–116.

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.
54:26 Sugandhi et al.

[12] Rupinder Kaur and Parteek Kumar. 2014. HamNoSys generation system for sign language. In Proceedings of the In-
ternational Conference on Advances in Computing, Communications, and Informatics (ICACCI’14). IEEE, Los Alamitos,
CA, 2727–2734.
[13] Jestin Joy and Kannan Balakrishnan. 2014. A prototype Malayalam to sign language automatic translator.
arXiv:1412.7415.
[14] Sandeep Kaur and Maninder Singh. 2015. Indian sign language animation generation system. In Proceedings of the 1st
International Conference on Next Generation Computing Technologies (NGCT’15). IEEE, Los Alamitos, CA, 909–914.
[15] Amit Verma and Sandeep Kaur. 2015. Indian sign language animation generation system for Gurumukhi script. In-
ternational Journal of Computer Science and Technology 6, 3 (July 2015), 117–121.
[16] Paras Vij and Parteek Bhatia. 2016. Translator of Hindi Text to ISL and Extension of ISL Dictionary with WordNet.
Master’s Thesis. Thapar University.
[17] Lalit Goyal and Vishal Goyal. 2016. Development of Indian sign language dictionary using synthetic animations.
Indian Journal of Science and Technology 9, 32 (Aug. 2016), 1–5. DOI:[Link]
[18] Ulrike Zeshan. 2000. Sign Language in Indo-Pakistan: A Description of a Signed Language. John Benjamins Publishing
Company. DOI:[Link]
[19] Samar Sinha. 2017. Indian Sign Language: A Linguistic Analysis of Its Grammar. Gallaudet University Press.
[Link]
[20] Enoch O. Aboh, Roland Pfau, and Ulrike Zeshan. 2005. When a Wh-word is not a Wh-word: The case of Indian sign
language. In The Yearbook of South Asian Languages and Linguistics, R. Singh (Ed.). Amsterdam Center for Language
and Communication, Amsterdam, the Netherlands, 11–43. DOI:[Link]
[21] Sugandhi, Parteek Kumar, and Sanmeet Kaur. 2018. Online multilingual dictionary using Hamburg notation for
avatar-based Indian sign language generation system. International Journal of Cognitive and Language Sciences 12, 8
(2018), 120–127.
[22] Ralph Elliott, Javier Bueno, Richard Kennaway, and John Glauert. 2010. Towards the integration of synthetic SL
animation with avatars into corpus annotation tools. In Proceedings of the 4th Workshop on the Representation and
Processing of Sign Languages: Corpora and Sign Language Technologies. 84–87.
[23] Daniel Stein, Jan Bungeroth, and Hermann Ney. 2006. Morpho-syntax based statistical methods for sign language
translation. In Proceedings of the 11th Annual Conference of the European Association for Machine Translation. 169–
177.

Received May 2019; revised December 2019; accepted February 2020

ACM Trans. Asian Low-Resour. Lang. Inf. Process., Vol. 19, No. 4, Article 54. Publication date: April 2020.