Machine Translation Approach to Translate Amharic Text to

Ethiopian Sign Language

Minilik Tesfaye
Lecturer, Faculty of Informatics, St. Mary’s University College, P.O.Box
18490, Addis Ababa, Ethiopia

Abstract
Sign language is a communication system using gestures that are interpreted
visually. Many people in deaf communities around the world use sign languages as
their primary means of communication creating a strong sense of social and cultural
identity. About 0.4% of the whole population of Ethiopia is deaf and out of this,
more than half is illiterate because of the fact that there are very limited studies
conducted on Ethiopian Sign Language. This research document presents a Machine
Translation system to translate Amharic text to an equivalent Ethiopian Sign
Language (ESL) by finger spelling representation with the help of a 2-dimensional
animating avatar rendering the equivalent ESL finger translation. Animating an
avatar for rendering Ethiopian finger spelling requires a dynamic and periodic
switching of hand shapes and hand movements. Macromedia Flash 8.0 and
ActionScript 2.0 are used as a tool to model and design the avatar, the hand shapes
and the hand movement definitions. The translation system is tested by 10 deaf
people and 66.6% of them responded that the overall performance is good and
supportive.

Keywords: Ethiopian Sign Language, Ethiopian Finger Spelling, avatar based

translation
-----------------------------------------------------------------------------------------------------

Introduction
According to Ethiopian National Association for Deaf (ENAD)
referring Ethiopian national statistical agency's study, about 1.5 million deaf
people lived in Ethiopia in 1994 E.C. and latter in 2000 E. C., the deaf

252
population raised to approximately 2.5 million. About 8% of the whole
population of Ethiopia is disabled people. Out of these, 2% counts for deaf
and hard of hearing people.
Deafness, characterized by visual communication, creates a strong
sense of identity among the society of any country. Visual communication
like sign language is the primary means of communication among the deaf.
Except for parents of deaf people, teachers of deaf students and people
interested on sign language; most of the hearing society does not know sign
language. This created a huge communication gap between hearing people
and the deaf. The researcher believes that a research attempt on translating
either from the auditory language (Amharic) to that of sign language (ESL)
or the vice versa eases the communication gap. There is either two-way
translation (Bilingual translation) or one-way translation among natural
language translation systems (Emiko, n.d.). Bilingual translations are
commonly observed for languages of same nature (spoken/written
languages). A translation made between sign languages and spoken/written
languages like Amharic is more manageable if accompanied in one-way
translation approach because of the nature of the two languages. In one-way
translation, the translation direction is best identified based on previous
research attempts, available resources and the worth of beneficiaries. In the
case of Amharic and ESL, there are two directions of translation: from
Amharic to ESL or from ESL to Amharic. Such translations are conducted
considering that ESL is of the deaf side and Amharic is of the hearing
society. Translating Amharic to ESL benefits the deaf because it helps the
deaf understand what is stated in Amharic. It also helps the Amharic
speakers express what they want to say to the deaf. Translating ESL to
Amharic on the other hand, benefits the hearing people because it helps the
hearing people understand what is said in ESL.

Scope and limitation

253
 Sign languages all over the world including Ethiopian Sign Language
make use of finger spelling and signing. Signing by itself include sign
language, sign-spoken language, signed-spoken language. The intended
system is supposed to take a text input, undergoes through a machine
translation path and finally, delivers an animation output. On the input part,
this study is specific to a Unicode Amharic text typed with the phonetic
mode of Geez alphabet. Associated with the processes, an adoption of
machine translation system adopts a multi-path machine translation approach
takes part to analyze the internal process of the system. On the output part,
this research is limited to only the application of Ethiopian finger spelling to
convey meaning by ESL. The reason to select only the finger spelling
ignoring the conceptual signs is that finger spelling is quite manageable to be
presented by animation with some machine translation rule. Conceptual ESL
words can of course be animated also; but no refined rule can be defined to
generate animations for even a very specific corpus of words in a particular
domain. The application of set of animations corresponding to some corpus
of words in a specific domain would be no better than the correspondence of
pre-recorded video sets corresponding to domain specific corpus of words,
which is already studied on the context of hypermedia tool development for
ESL (Endale, 2005).

Sign Structure in Sign Language

Sign Languages involve simultaneous manual and non-manual
components for conveying meaning. Non-manual features are comprised of
the posture of the upper torso, the orientation of the head and facial
expressions. Manual features have been often been decomposed as hand-
shape, hand orientation, hand position and motion (Stokoe, 1978). The
Hamburg Notation System (HamNoSys) is an established phonetic
transcription system for sign languages comprising more than 200 iconically

254
motivated symbols to describe these manual and non-manual features of
signs (Prillwitz, Leven, Zienert, Hanke, Henning, et al, 1989).
Visual languages systems especially like sign language are expressed by
combining different components of the sign. Signs have the following
components (Daniel, 2006) (Kadous, 1995)
I. Hand shape
Hand shapes are usually used to represent finger alphabets of sign languages.
Hand shapes are created by bending and stretching the finger and the palm.
There are signs created using only one hand like Ethiopia, America and two
hands like British.
II. Location
Location refers to the place where the signing is performed. The signing can
be performed either by hand, eye or other body. There are signs that are the
same with other components of sign and only defers in their location. For
example, signing the word "mother" and the word "father" in ESL have the
same hand shape except for the sign for "mother" is signed locating hand at
the fore-head and that of "father" at the chin.
III. Movement
There are a wide variety of movements possible with sign language
including arcs straight lines and wavy patterns. Hand movement is
dominantly used to make the major component of ESL. Head movement, eye
movement and the torso movement may also be applied in ESL.
IV. Orientation
There are three possible orientations of the hand: roll, pitch and yaw.

Finger Spelling
Finger spelling is a manual representation of written language, and it
is used as a substitute for speech as a live, or face-to-face, medium of
communication. It is a means of communication rather than spelling of any

255
spoken language words. It is used in combination with sign language for
proper nouns, names and addresses and for words that have no specific sign
(Riekehof, 1987). The term finger spelling stands for both sign language
production and perception (Evans, 1982).
When finger spelling, people use their dominant hand to make hand shapes,
one corresponding to each letter of the word. Fluency in sign language
includes the ability to produce and recognize finger spelling at a rate of four
characters per second (Wolfe, 2004).

Ethiopian Sign Language

Ethiopian Sign Language is used as the primary communication
language among the Ethiopian deaf community. It is derived from American
Sign Language since the first school for deaf people was established by
American missionaries and it served as an instruction tool. The major
differences that exist between American and Ethiopian sign languages are
(Legesse, 2008):
• Ethiopian Sign Language has borrowed too some signs from Nordic
countries Sign Languages, especially from the Finnish one.
• There are local signs created and used in some specific deaf schools
in the country, later included into Ethiopian Sign Language.
E.g. E  
EA
 
• The finger alphabet used in Ethiopian Sign Language is constructed
based on the shapes and orientations of Amharic alphabet as shown in Figure
1.
Very limited attempt has been made to find out the grammatical structure
and rule of Ethiopian Sign Language, due to non-existence of local experts in
the field of sign languages. However, there are fruitful attempts in the
development of modern sign language dictionaries in Ethiopian local spoken
languages (BERTAT, 1997)

256
Ethiopian Finger Spelling
ENAD (Ethiopian National Association for Deaf), has developed the
Ethiopian Finger Spelling in 1971(Legesse, 2008). Latter, Ethiopian Finger
Spelling was accepted and recognized by the Ministry of Education.
Currently, it is used in all deaf schools along with the American finger
spelling. The American Finger Spelling is used when we want to express
words or concepts in English and local languages based on Latin alphabet,
like Oromiffa, while Ethiopian finger spelling is used when we need to
express words or concepts in language
anguages that use Gee’z alphabet, like
Amharic and Tigrigna (Legesse, 2008). Ethiopian Finger Spelling has 34
unique hand shapes to represent the first order Amharic alphabet
alphabet, where 33
y’) have been used until 2009 (Legesse,
of them (except the Amharic letter ‘y
2008). The hand shape for letter ‘y’’ (shown in Figure 2) is lately introduced
as the 34th finger spelling of ESL by ENAD. Figure 1 shows the 33 Ethiopian
finger spelling with the corresponding Gee’z alphabet except letter ‘y’ as
found from ENAD.

Figure 1: Ethiopian Finger Spelling

257
Figure 2: the lately introduced letter ‘y’ (v)

ESL Linguistic Issues

In ESL, several parts of the body convey meaning in parallel: hands
(location, orientation, and shape), eye gaze, mouth shape, facial expression,
head-tilt, and shoulder-tilt. Signers may also interleave lexical signing (LS)
with classifier predicates (CP) during a performance (Legesse, 2008). During
LS, a signer builds ESL sentences by syntactically combining ESL lexical
items (arranging individual signs into sentences). The signer may also
associate entities under discussion with locations in space around their body;
these locations are used in pronominal reference (pointing to a location) or
verb agreement (aiming the motion path of a verb sign to/from a location).
It is also necessary to remember that ESL is signed dominantly by the
conceptual and conventional gestures, and body signs and ESL finger
spelling assist in conveying nouns and pronouns. Commonly, these finger
alphabets are applied together with ESL conceptual signing to convey some
doubtful words by initialization (Legesse, 2008).

Gee’z alphabet on Power Ge’ez 2005

Amharic language use Gee’z alphabet as its scripting latter.
According to Power Ge’ez 2005, there are a total of 44 base Gee’z letters in
Amharic texts. Each of these Gee’z base latter is multiplied by 7. On the base
letters, there is no vowel sound identified by the phonetic mode of power
Gee’z. Therefore the 7 is to refer for 1 base letter and 6 different Amharic
vowel sounds. Out of these letters, the Ethiopian Sign Language finger
spelling uses only 34 base letters of Gee’z alphabet. The developer of power
Gee’z software, Concepts Data Systems P.L.C., has provided different
modes like phonetic (P), phonetic Unicode (PU), typewriter (T), typewriter
Unicode (TU) and English (E) modes of using the power Gee’z. In this

258
research, the phonetic mode (P) is selected to take Amharic input from
keyboard because the phonetic mode is the simplest and common for people
who do not type Amharic frequently. The researcher believes that the
simplicity may help both the deaf and hearing people in easing the
application of this new study results.
Concepts Data Systems P.L.C presented the phonetic mode of Gee’z
alphabet in such a way that it corresponds with the English sound of that
letter by English alphabet, as presented in Table 1.

Grouping Gee’z alphabet

It appears manageable if the seven columns of Gee’z alphabet are
grouped from one to seven on the basis of the 6 vowel group and the first
consonant (base) group. In this study, the seven groups of Gee’z letters are
used as indicated in table 1.

Table 1: Amharic vowels and their Groups

Consonant u I A y e o
Ó°´ Ÿw° XMe ^w Ue dÉe dw°
Group 1 Group 2 Group 3 Group 4 Group 5 Group 6 Group 7

As mentioned before, the ESL finger spelling uses only 34 consonant

letters with 6 Amharic vowels for each. According to the above table, the 34
unique letters are under group 1 and the Amharic vowels generate the
remaining groups (group 2 through group 7). ESL Finger spellings of group
1 letters (Ó°´) are signed without hand movements; whereas the remaining
groups are signed by hand movements as shown by the arrows in figure 3.

259
 No
movement
Ó°´ Ÿw° XMe ^w Ue dÉe dw°
Consonant u i a y e o
Figure 3: Amharic vowels associated with ESL finger spelling hand
movement.

In all of the 33 selected consonants with their vowels, the phonetic

mode of power Gee’z uses either one or two English letters. While typing a
consonant Gee’z letter, the phonetic mode uses only one English letter
(mostly that corresponds with the sound of the Gee’z letter). But while
typing the combination of consonants and vowels, the phonetic mode of
power Gee’z uses two letters of English alphabet. Should be mentioned that
there are cases in ESL finger spelling associated with Gee’z letters that need
the Caps Lock – On (%,¤,g, ç, and ì).

Translation architecture
This research investigates the conversion of Amharic text into the
equivalent ESL finger spelling sign. ESL finger spelling translation needs an
“heuristic” approach that can be mapped in the direct and transfer paths of
Machine Translation architecture. These are defined in the experimentation
part of this study as methods of hand-shape modeling and hand movement
definitions. The pyramidal model of machine translation architecture is
customized in this research as in Figure 4.

260
 Interlingua

Analysis Unicode Generation

transfer

Direct translation

Amharic Target ESL

text finger
spelling
Figure 4: Machine translation architecture

The direct translation involves a simple mapping between the corpus

of Amharic text and ESL finger spelling words. On the transfer level of the
model, the mapping derives from the Unicode equivalent of the input
Amharic text and the equivalent ESL finger spelling word. The Interlingua
system incorporates the different intermediate rules of ESL finger spelling as
a correspondence to the Unicode mapped word of Amharic text. The rules of
ESL finger spelling considered in this level (Interlingua) are hand shapes and
hand movements.

Modeling the system

The translation system has some basic components that are integrated
to one other to show the right ESL finger spelling equivalent of an input
Amharic text, Hand shape definitions library, Hand movement definitions
library, Integrator and the avatar are the basic component of the translation
system model. The following figure 5 shows these components with their
interaction and the process flow of each component:

261
 Amharic text input

Break into
consonant and vowel
Unicode
reference

Take one letter

has
vowel?
No Consonant

Vowel sounds Yes Consonant

Hand
movement Hand shape
definitions library for
group1 letters

Integrate
Hand movement shape and
definition movement
Hand shape
definition

Output

Animating An other Yes

avatar letter left?

No

Show word by END

Finger-text

262
 When an Amharic text is typed from the keyboard, the Unicode
equivalent of each letter is collected from a module. Note that the Unicode
equivalent of each base letter of Gee’z alphabet is defined in a module for
the purpose of displaying the Amharic font at the input box. Each letter of
the input text is then broken into a consonant (group1) and vowel-containing
(Non-group1) category and stored in an array. Here, if a letter is categorized
as vowel-containing, it is identified by the base letter and the following
vowel. For example, in the word “S`"„”, the phonetic mode key sequence
becomes “m, re, ka, to”. Only the first letter “S” is characterized by a
consonant category (group1). But the remaining letters are all characterized
by a vowel-containing category (Non-group1). For instance, “`” is
phonetically typed as “re” where “r” represents the base letter “[” from
group1 and the vowel “e” is appended to it. The following table illustrates
the consonant / vowel category on the above example.

Table 2: Categorizing Individual Letters of an Input Text

Alphabet of the typed word Keyboard Equivalent Category
Consonant Vowel

S m (S) Group1

` r ([) E Non-group1

" k (Ÿ) A Non-group1

„ t (}) O Non-group1

The system checks the category of each individual letter as consonant

or vowel-containing. This is carried out just by deciding whether a letter has
a vowel (Non-group1) or not. If a letter does not have a vowel, then it is a
consonant which is directed to hand shape library to refer for the right hand
shape from group1 set. But if the letter contains a vowel (Non-group1), the
263
letter is supposed to be directed to both hand movement definition set and
hand shape library, so as to prepare the right input for the avatar. Hand
movement definition is a set of defined rule that is directly associated with
the 6 hand movements corresponding with the 6 vowels. These movement
definitions are applied on the shoulder-joint and the elbow joint of the
avatar’s hand (see figure 6). The hand shape definition library is the set of
pre-defined hand shape movie clip for each of the group1 letters of Geez
alphabet as shown in Figure 1. These two components, the hand movement
definition set and the hand shape library integrate their information and
provide the right information to the avatar. This process is repeatedly iterated
until there is no other letter is left in the queue to generate an ESL finger
spelling word.

Translation algorithm
The translation algorithm shows a step-by-step transition of the
translation process passing through each component.
Step 1: Enter an Amharic word by using the phonetic mode of power Gee'z.
Step 2: If unsigned character is available:
pick one character and identify the character's behavior based on the two
categories, Group 1 letter or none group 1 letter.
Sub procedure:
If the input letter contains either of the 6 vowel sounds (a, u, i, a, y and o),
then it is identified as vowel containing (none-group 1).
Otherwise, it is a group 1 letter.
Step 3: If the letter is group 1, then it picks the correct hand shape for the
hand shape library. Otherwise, pick the correct hand shape for the hand
shape library and refer the right type of hand movement corresponding with
the attached vowel.
Step 4: Integrate the hand shape information and the hand movement
definition and provide- input it to the avatar.
264
Step 5: Go to step 1.

Modeling the avatar

An avatar in this research is an animating object that resembles and
acts as a human ESL finger spelling translator. The avatar is constructed with
only a torso, and the upper torso limbs: two hands, and a head. Only the right
hand of the avatar is dynamically associated with the different translation
operations, and is by considering the avatar as a right-handed translator. The
basic joints of the hand to determine hand motion are taken as the shoulder-
joint and the elbow. Shoulder-joint connects the avatars shoulder with its
upper-arm and an elbow connects the fore-arm with the upper arm.

The white point

at the avatar’s
right shoulder is
the shoulder joint
and the dotted
selection shows
the rotational
movement of its
right hand from
the shoulder.

Figure 6: Shoulder joint hand movement of the avatar

As shown in the figure 6, Rotational operation is applied on the

shoulder-joint and the elbow to generate the required hand movement on
ESL finger spelling. The shoulder-joint rotation and the elbow joint rotation
integrate to create a virtual realism to the 6 hand movements associated with
the 6 vowels of the Geez alphabet. The hand shapes of the avatars right hand
are carefully modeled having similar shape with the actual hand shape and
stored in the library window of Macromedia Flash. Each of the 34 base
letters are designed and stored in the library as a movie clip (one of

265
Macromedia Flash’s symbol type that can be manipulated and called by an
action script language). The resemblance sample between the avatar’s hand
at design mode and the real picture of the letter “u” is shown in figure 7.

Figure 7: Design of avatar’s hand shape as compared to real hand shape

picture

Hand shape modeling needs to consider each of the five fingers

independently in five different layers. And each finger consists three sections
(knuckles). Here also, rotational operation is applied on each fingers and
finger knuckles. Note that each finger is separately treated as a movie clip
symbol and even the three knuckles of each finger are treated as movie clips.
These components are converted to movie clips because movie clips are
highly interactive with action script of Macromedia Flash.
Each of these components of the avatar are named and layered in
Macromedia Flash 8. The following figure shows the snapshot of the avatar
in Macromedia Flash 8.

Text matching analysis

Matching the Amharic text input with the hand shape library and
associated with hand movement definitions of the system follows the
procedure demonstrated in the machine translation model. But the internal
text matching analysis passes through some technical steps. As mentioned in
the model, texts are first broken into their individual constituents as
consonant and vowel-containing. There is a method that identifies the
category of each letter of the input text as consonant or vowel-containing by
266
returning Boolean information as True for Consonant (group1) input and
false otherwise. This method identifies the category of each letter simply by
referring the phonetic mode key-board input for each letter. If a letter is
phonetically typed being accompanied by any of the 6 vowels (a, u, i, y, e
and o), then this indicates that it contains a vowel. For instance, this module
returns True if “u” is typed and False if “u?” is typed. Note that we can use
this module only by associating True with group1 alphabets and False with
the remaining none-group1 alphabets. It is only the None-group1 elements
(Vowels) that need hand movement definition. Any letter, group1 or none-
group1 element, have its own hand shape definition. Hence, hand shape
definition does not depend on this module but hand movements are referred
only if this module returns False.
As mentioned earlier, hand shape definitions are designed by
modeling the avatar’s right hand as shown in figure 7 for the 34 base letters
of ESL finger alphabet. Each of these hand shapes is stored in a library of
Macromedia Flash 8.0 being compatible with action script operation. The
action script code is responsible to switch between the different hand-shapes
needed for each alphabet of the typed input text; 34 hand shapes are designed
and stored in the macromedia flash library as movie clip. Switching between
these hand shapes is done by taking different layering display levels that
hides the image at the lower display level compared with the current display
level. Note that hand shapes are changed only for different base letters
(group1) or in other words, for different consonants. For example there is no
hand shape change between “S” and “T@” because they have common base
letter “S”. Therefore the two letters have common hand shape. There is a
module that can switch hand shapes by matching the Unicode of the input
letter with that of the stored hand shape movie clip in the library associated
with the input Unicode. This module returns a link to the right movie clip for
a given Unicode input.

267
 Hand movement definitions are defined based on the avatars two
joints: the elbow (Hinge joint between the forearm and upper arm) and the
shoulder joint (joint between upper arm and shoulder). Flexible rotational
operations on these two joints suffice to show the 6 different hand
movements for group2 though group7. First, a default rotational angle is set
to 00 with a reference of the vertical avatars standing direction for both joints.
Vertical and horizontal referencing is frequently applied to the rotational
operation to result on the right hand movement. Additional operations like
scaling and translation (movement) are slightly applied to create virtual
realism.

Rotation with vertical

reference Shoulder joint

Elbow
Rotation with
horizontal reference

Figure 8: Rotational Movements

Stick picture shows the rotational operations applied on the elbow of the
avatar’s hand with respect to the vertical and the horizontal references.

Testing the avatar

Ethiopian Signs Language uses two basic approaches to
communicate: conceptual signing and finger alphabet, but there is even a
hybrid approach called ‘initialization’ that uses both finger spelling and
conceptual signing (Legesse, 2008). As stated on the scope of this document,
268
this study is limited to ESL finger spelling only. Even if finger spelling is
very crucial for the fluency of ESL and can convey meaning by itself, it is
least frequently used in ESL. ESL most commonly uses the conceptual
signing, so identification of the target group of this system is the first step of
testing. Hence both deaf people and hearing people are identified. Testing
this system obviously needs the involvement of deaf people who
communicate by ESL and who can read and write Amharic text, so I tried to
get as many of such people as possible. The avatar translation system is
provided for 10 different criteria fitting people, with a questionnaire to
evaluate the avatar where some of them have also repeatedly tested it (two of
them responded twice for two different tests).

The testing parameters are identified as:

1. quality of rendering ESL finger alphabets by avatars hand shapes;
2. hand movement performance;
3. quality of animation transition from one ESL finger alphabet to the
other;
4. general amount of Amharic sample texts correctly translated;
5. quality of rendering an accurate ESL finger alphabet word;
Based on these parameters, 12 different responses have been collected from
10 deaf people, where two of them responded twice.

269
 Table 4: Statistical Illustration of the Respondent on the first 4 Objective
Parameters
Parameter Options Respondents
Frequency Percentage

Quality of hand shape for ESL finger Was correct 6 50

alphabet Medium 4 33.3
Was not correct 2 16.67
Hand movement quality Was correct 4 33.3
Medium 4 33.3
Was not correct 4 33.3
Was there Animation transition quality No problem 4 33.3
problem? Little problem 5 41.67
Serious 3 25
problem
Amount of Amharic sample texts correctly More than half 7 58.33
translated Half - 0
Less than half 4 33.33

Even if almost all respondents have appreciated the study mentioning

that such studies are quite uncommon, they also pointed out some serious
problems. These problems are observed on the hand movement quality and
animation transition between two ESL finger spellings as shown in the above
table. Such problems appeared because the avatar translates in a 2-
dimensional plane. In fact 3-dimensional spaces are very crucial for
rendering any sign language. Most of the respondents from the user
satisfaction analysis, recommended a 3D animation technique to cover the
failures of this system.

Conclusions
ESL lacks a standard writing system and besides very limited studies
has been conducted on ESL finger spelling, despite the fact that it is quite

270
necessary to help the existing communication gap between hearing people
and the deaf. Animating any character is a challenging task, especially when
the animation is humanoid and intended to act as a human. Avatar
construction is quite different from other animations because, avatars need to
have some more heuristics of human operations.
In this research, an attempt is made to adopt a machine translation
approach to translate Amharic text into ESL finger alphabet word. The
machine translation architecture is adopted from Interlingua system of the
three-level Machine Translation Architecture. The adopted architecture and
the translation model are designed in such a way that it can be a good input
for any researcher who want to improve this system or even to construct ones
own avatar on other platforms or other media.

Recommendations
A number of shortcomings can be observed from this research that
may initialize further similar studies. The following issues need to be
considered to complement the study area.
• It is quite convincing to use avatar based machine translation
especially for Sign language related researches. From the literatures
reviewed in this research and performance analysis, it has been observed that
using a 3D avatar can definitely better convey sign languages than 2D
avatars.
• ESL is complete in both signing and finger spelling. Signing includes
the conceptual sign expression which is dominantly applied to convey
meaning in ESL. Hence, there is a need to study a Machine Translation of
conceptual ESL expressions with a high level heuristic-based animation
technique. It is even better if both, the finger spelling and the conceptual
expression, could be incorporated together in a single application.

271
• This research reads only a single word to translate in to ESL finger
alphabet expression which is not applicable for a phrase, sentence, or a
document translation. Therefore it is necessary to extend the system so that it
handles sentences or even a document.
• Identification and classification of nouns and action words (verbs) of
ESL is worth studied issue because ESL or any sign language use finger
spelling fully or partially for nouns and proper nouns. On the other side, sign
languages use of conceptual expression for action words or verbs. This
would ease the translation of Amharic text to ESL or other sign language.

Reference
Augustine Su Project. (n.d.). Accessed October 1, 2002, from
http://www.csdl.tamu.edu/su/asl
Augustine Su Project. (n.d.). Retrieved October 1, 2002, from
http://www.csdl.tamu.edu/su/asl
Becky Sue Parton, Sign Language Recognition and Translation: A
Multidisciplined Approach from the Field of Artificial Intelligence,
2004
Daniel Assefa: Amharic Speech Training for the Deaf, Masters Thesis, Addis
Ababa University, Computer Science Department, 2006
Dorr, P. Jordan, and J. Benoit. 1998. “A Survey of Current Paradigms in
Machine Translation.” Technical Report LAMP-TR-027, Language
and Media Processing Lab, University of Maryland
Emiko SUZUKI, Mariko H., Koyoko K. Bilingual Sign Language (n.d.):
Dictionary for Learning a Second Sign Language with out Learning
the Target Spoken Language
Endale Asefa, Design and Development of Hypermedia Learning Tool for
Ethiopian Sign Language,2005
Ethiopian National Association for Deaf: BERTAT (Yearly Magazine).
Addis Ababa: Ethiopia, 1997
Evans, L.: Total Communication: Structure and Strategy. Florida: Gallaudent
College press, 1982
Francik, J., & Fabian, P. (2002). Animating sign language in the real time.
Paper resented at the 20th IASTED Conference on Applied
Informatics, Innsbruck, Austria.
H. Kamp and U. Reyle. 1993. From Discourse to Logic Introduction to
Model theoretic Semantics of Natural language, Formal Logic and

272
 Discourse Representation Theory. Kluwer Academic Publishers,
Dordrecht.
Ian Marshall, E´ va Sa´fa´r, A Prototype Text to British Sign Language
(BSL) Translation System
Kadous, M. W.:Recognition of Australian Sign Language using
Instrumented Gloves, Honours Thesis, School of Computer Science
and Engineering, University of New South Wales, 1995
Kyle, J : Sign and School: Using sign in Deaf child’s development. 1987
Liddell. 2003. Grammar, Gesture, and Meaning in American Sign Language.
UK: Cambridge U. Press
Matt Huenerfauth, English Text into American Sign Language Animation,
Computer and Information Science University of Pennsylvania
Philadelphia, PA 19104
Morford and J. MacFarlane. “Frequency Characteristics of ASL.” Sign
Language Studies, 3:2.
Sáfár and I. Marshall. 2001. “The architecture of an English-text-to-Sign-
Languages Translation System.” In G. Angelova, ed., Recent
Advances in Natural Language Processing. Tzigov Chark, Bulgaria.
Sandler, W. : An Overview of Sign Language Linguistics. In Keith Brown,
(ed.), Encyclopedia of Language and Linguistics. 2003.

273