INSTITUTE OF TECHNOLOGY

DEPARTMENT OF COMPU TER SCIENCE

SCHOOL OF GRADUATE STUDIES

MORPHOLOGY BASED SPELLING CHECKER FOR GEEZ LANGUAGE

MASTERS OF SCIENCE THESIS

FIKADE CHANE FELEKE

HAWASSA UNIVERSITY, HAWASSA, ETHIOPIA

NOVEMBER 2023
 MORPHOLGY BASED SPELLING CHECKER FOR GEEZ LANGUAGE

MASTERS OF SCIENCE THESIS

FIKADE CHENE FELEKE

A THESIS SUBMITTED TO HAWASSA UNIVERSITY

INSTITUTE OF TECHNOLOGY,
FACULTY OF INFORMATICS,
DEPARTMENT OF COMPUTER SCIENCE,
SCHOOL OF GRADUATE STUDIES
HAWASSA, ETHIOPIA

IN PARTIAL FULFILLMENT OF THE

REQUIREMENTS FOR THE
DEGREE OF

MASTER OF SCIENCE IN COMPUTER SCIENCE

NOVEMBER 2023

i
 Declaration

This thesis is my original work carried out by me and has never submitted to this or any other
institution or university to get any other degree or certificates.

Name of the student Fikade Chane Signature _____________

Date of submission: __________________

Place: Hawassa

ii
 ADVISORS’ APPROVAL SHEET

SCHOOL OF GRADUATE STUDIES

HAWASSA UNIVERSITY ADVISORS’ APPROVAL SHEET

This is to certify that the thesis entitled “Morphology based Spelling checker for Geez

language” submitted in partial fulfillment of the requirements for the degree of Master's with

specialization in Computer Science the Graduate Program of the Department of Computer

Science, and has been carried out by Fikade Chane , under our supervision. Therefore, we

recommend that the student has fulfilled the requirements and hence hereby can submit the

thesis to the department.

Name of major advisor Signature Date

Dr.Tesfaye Bayu Bati (PhD) __________ _____

Name of co-advisor Signature Date

Teshager Kassa (MSc) __________ _____

iii
 Acknowledgment

Before all, I would like to thank Almighty God, his mother holly Virgin Mary, and all the
saints for leading and following me in each steps, for giving me the strength when I get weak
and for showing me the way when I get lost.

Next, I would like to acknowledge my advisor Tesfaye Bayu Bati (PhD) for supporting and
reading my document line by line and giving me comments. And, I would like to
acknowledge my co-adviso Teshager kassa (MSc) for supporting from improving my title to
reading my document.

I wish to express my deepest gratitude to my wife Beletie Asmare for supporting and
encouraging me in my journey. And also, I would like to thank my father and mother and
other family members who encouraged me to go on.

iv
 Abstract

Geez is one of the ancient languages. It belongs to Semitic language family. Many ancient
literature and books have written in Geez. Currently, Geez course are offered in various
colleges, universities and in some primary schools. However, still developed NLP applications
are insufficient for this language. In order to write error free Geez text in less time, spelling
checker application is a critical NLP application. Spelling checker is a tool used to detect
spelling error in a block of text and gives closer suggestions to the error words. A previous
attempt has made to develop a spelling checker for Geez language. This attempt was focus
only homophone alphabet interchangeably error.

In this study, we proposed morphology based (dictionary lookup and morphological analyzer)
approach to Geez language spelling checker. The system have three main compenents.These
are text preprocessing, error detection, and error correction. To achieve the objective of this
study the researcher builds one main dictionaries that contains Geez language lexicon and
morphological feature. The researcher built 6115 unique Geez lexicon and 955 rules had
defined. We adopt the Hunspell dictionary and affix file format to design a lexicon (i.e. the
knowledge base component) and hashing algorithm for searching. Hunspell is an open source
spelling checker tool. It has designed especially for languages that have complex morphology.

Finally, the researcher has developed a prototype of a system to test the functionality and
performance of the Geez language spelling checker. The accuracy of error detection expressed
in terms of precision and recall. In addition, the accuracy of suggestion expressed in terms of
suggestion adequacy. Therefore, we got the result of lexical recall 91.9%, error recall 83.7%,
lexical precision 97.2%, error precision 62.2% and correct suggestions provided by GLSC
87.5%. The overall performance of the system is 90.05%. We conclude that increase the size
of the dictionary and develop well organized rule will increase the overall performance of the
Geez language spelling checker.

Key words: Error Detection, Error Correction, Spell Checker, Morphology, Non Word Error,
Real Word Error, Geez language, dictionary lookup

v
Contents
Acknowledgment ..................................................................................................................................... iv
Abstract ..................................................................................................................................................... v
CHAPTER ONE .......................................................................................................................................1
1. Introduction .......................................................................................................................................1
1.1. Background ...................................................................................................................................1
1.2. Motivation .....................................................................................................................................3
1.3. Statement of Problem ....................................................................................................................4
1.4. Objective .......................................................................................................................................8
1.4.1. General Objective..................................................................................................................8
1.4.2. Specific Objective .................................................................................................................8
1.5. Scope and Limitation of the Study ................................................................................................9
1.6. Significant of the Study.................................................................................................................9
1.7. Methodology .............................................................................................................................. 10
1.7.1. Literature Review ............................................................................................................... 10
1.7.2. Data Collection................................................................................................................... 11
1.7.3. Implementation Tool .......................................................................................................... 11
1.7.4. Evaluation Metrics ............................................................................................................. 11
1.8. Organization of the Thesis ......................................................................................................... 12
CHAPTER TWO ................................................................................................................................... 13
2. Literature Review ........................................................................................................................... 13
2.1. Introduction ................................................................................................................................ 13
2.2. Overview of Spelling Checker ................................................................................................... 13
2.3. Technique of Spelling Checker .................................................................................................. 14
2.3.1. Error Detection ................................................................................................................... 15
2.3.2. Error Correction ................................................................................................................. 16
2.4. Performance Evaluation Technique for Spelling Checker ......................................................... 19
2.5. Related Works ............................................................................................................................ 21
2.5.1. Spell checker for foreign language..................................................................................... 21
2.5.1.1. Arabic Spell Checker ................................................................................................. 21
2.5.1.2. Spelling Checker for Marathi ..................................................................................... 22

vi
 2.5.2. Spell checker for Ethiopian language................................................................................. 22
2.5.2.1. Afaan Oromo Spelling Shecker ................................................................................. 22
2.5.2.2. Amharic Spelling Checker ......................................................................................... 23
2.5.2.3. Tigrigna Spelling Checker ......................................................................................... 27
2.5.2.4. Geez Spelling Checker................................................................................................. 28
2.5.2.5. Morphological Analyzer for Geez language .............................................................. 29
2.5.3. Summary of Related Works ................................................................................................... 31
CHAPTER THREE................................................................................................................................ 36
3. Overview of Geez Language .......................................................................................................... 36
3.1.1. History of the Geez language ............................................................................................. 36
3.1.2. Writing System of Geez language ...................................................................................... 36
3.2. Geez word class ......................................................................................................................... 38
3.2.1. Affixation in Geez word..................................................................................................... 43
3.2.1.1. Prefixes....................................................................................................................... 44
3.2.1.2. Suffixes ...................................................................................................................... 46
CHAPTER FOUR .................................................................................................................................. 48
Design of Geez Spelling Checker .......................................................................................................... 48
4. Overview ........................................................................................................................................ 48
4.1. Morphological analyzer for Geez Language .............................................................................. 48
4.2. System Architecture for Geez Spelling Checker........................................................................ 51
4.2.1. Preprocessing Component .................................................................................................. 53
4.2.2. Error Detection Component ............................................................................................... 54
4.2.3. Error Correction Component.............................................................................................. 56
CHAPTER FIVE.................................................................................................................................... 59
5. Experiment ..................................................................................................................................... 59
5.1. Overview .................................................................................................................................... 59
5.2. Testing Data Preparation ............................................................................................................ 59
5.3. Prototype of the System ............................................................................................................. 60
5.4. Performance Evaluation ............................................................................................................. 63
5.4.1. Performance of Spelling Error Detection ........................................................................... 63
5.4.2. Performance of Spelling Error Correction ......................................................................... 64

vii
5.5. Result and Discussion ................................................................................................................ 65
5.6. Challenge and Limitation ........................................................................................................... 66
CHAPTER SIX ...................................................................................................................................... 68
6. Conclusion and Recommendation...................................................................................................... 68
6.1 Conclusion ....................................................................................................................................... 68
6.2. Recommendation ............................................................................................................................ 69
Reference ............................................................................................................................................... 71
Appendix A: Geez Alphabet and Number .............................................................................................. 75
Appendix B: Sample Affix file Rule in Hun spell ................................................................................. 76
Appendix C: Sample Corpus .................................................................................................................. 80
Appendix D: Sample Suggestion ........................................................................................................... 83

List of Abbreviations

NLP: Natural language processing

EOTC: Ethiopian orthodox Tewahido church

GLSC: Geez language spelling checker

ASR: Automatic speech Recognition

MT: Machine Translations

OCR: Optical Character Recognition POS:

POS: Part of Speech

TTS: Text To Speech

NER: Named Entity Recognition

IR: Information Retrieval

LED: Levenshtein Edit Distance

viii
List of table

Table1. 1 Geez homophone alphabet ........................................................................................................3

Table1. 2 comparing Geez from Amharic and English ............................................................................6
Table1. 3 Geez inflected in 10 pronouns .................................................................................................7

Table2. 1 Summary of literature related works ...................................................................................... 35

Tatble3. 1 sample example to show modern representation of Geez alphabet ...................................... 37

Tatble3. 2 Example of singular to plural noun ....................................................................................... 38
Tatble3. 3 Example of Geez Amharic adjective suffix and prefix to change in to plural form ........... 39
Tatble3. 4 Tense-mood Identified by Ethiopian Scholars [38] ............................................................. 40
Tatble3. 5 Geez root verb ...................................................................................................................... 41
Tatble3. 6 five stem verb with tense mode [46] ..................................................................................... 44
Tatble3. 7 Basic out breeding phones .................................................................................................... 45
Tatble3. 8 Geez subjective suffix adapted from..................................................................................... 46
Tatble3. 9 Geez objective suffix ............................................................................................................ 47

Table5. 1 Result of cause of pattern error .............................................................................................. 60

Table5. 2 Error detection result .............................................................................................................. 64
Table5. 3 sample example to calculate SA ............................................................................................ 65

List of figure

Figure1. 1 sample screen shot from android .............................................................................................5

Figure4. 1 (a and b) Morphological feature (affix and rule) and Dictionary ....................................... 49
Figure4. 2 Geez circumfix...................................................................................................................... 50
Figure4. 3 General system architecture ................................................................................................. 52

Figure5. 1 sample example of error detection ........................................................................................ 61

Figure5. 2 sample prototype to rank suggestion .................................................................................... 62

ix
List Algorithm

Algorithm4. 1 for Tokenization process ................................................................................................ 53

Algorithm4. 2 Dictionary lookup .......................................................................................................... 54
Algorithm4. 3 Algorithm for morphological analyzer [3] ..................................................................... 55
Algorithm4. 4 for morphological generator ........................................................................................... 57

x
 CHAPTER ONE

1. Introduction

1.1. Background

Language is one of the fundamental aspects of human behavior. It uses either in written form
or spoken form. In written form, it serves as a means of recording information and knowledge
for long time and transfer to the next generation. In spoken form, it serves as the way to
support corporate human activities [1]. Language can classified in to two major parts. These
are natural language and artificial language [2]. A language that, we learn from environment
and used as a means of communication by human known as natural language like Amharic,
English, Geez e.t.c. whereas artificial language is a language developed based on a set of rule
to perform specific activity. For instance, Java, C++ e.t. c.

From natural language, Geez is one of the ancient languages. Geez language is the classical
language of Ethiopia and belongs to the Semitic language family [3]. This language was a
governmental language especially at the time of Axumite civilization for the kingdom and
imperial court [4] .At that time Geez language becomes the only national language in Ethiopia
and got an acceptance among people [5]. Due to this, different types of religious and secular
books and literature have written in this language. From secular books, ancient literature,
ancient philosophies, tradition, histories, romances, legal, mathematical, and medical texts are
some of the examples [6]. From religious books, Bible, Apocrypha, liturgical literature,
homiletic, theological, and magical texts, stories of martyrs and saints, religious poetry,
hymns in honor of Christ, the Virgin, the martyrs, the saints, and angels are some of
the examples.

After the end of Axum Empire Geez language have dominated by other Semitic language [4]
and Geez language also becomes as liturgy language in Ethiopian orthodox Tewahido church
(EOTC), Eritrean Orthodox Tewahido Church, the Ethiopian Catholic Church, the Eritrean
Catholic Church [5] [7].

1
Although Geez language used as liturgy language after the end of Axum Empire, currently,
Geez course are offered in various colleges, religious centers and higher institute and different
Geez books and other reference materials are publishing [8]. For example, Ethiopian Orthodox
Church Theology Colleges are giving Geez course and publishing various books. In addition
to this, various higher institutes teach Geez language as a course and organize in its own
department. In foreign country, like European and United States of America Universities [7]
and In Ethiopia like Wollo University, Bahir Dar University starts to teach Geez language. In
order to minimize errors and save time when we write Geez text, some Natural language
processing (NLP) applications must be developed.

NLP is a branch of artificial intelligence that helps computer to understand, interpret and
manipulate human language and how to program computer to process and analyze large
language of natural language data [4]. NLP focuses on designing and implementing of
tools, techniques, frameworks to enable computers communicate effectively with humans
[3]. It is the means for accomplishing different types of applications. Such as grammar
checking, spelling checker, Named Entity Recognition (NER), Information Retrieval (IR),
Speech Recognition, Machine Translation, Question Answering etc. From these NLP
applications, the proposed research focuses on developing the interactive spelling checker mod
for Geez language.

In computing, spelling checker is a tool that detects misspelled word(s) in a paragraph or

sentence and provides the alternatives to correct them [9]. The idea for spelling checker
development is not new. Still many researchers conduct a research on spelling checker in
foreign language as well as local language using different method [10]. For instance, for the
foreign language like Kanada [11],Punjabi [12],Hindi [13].The same research had also
been conducted on Ethiopian language like, Afan Oromo [9],Amharic [14] [10],Tigrigna
[15],Hadiyyisa [16] [17] ,Kambatissa [17] ,Awngi [16] [17]. However, most of African
languages have no complete spelling checker tool to integrate word processor [10].

2
Spelling checker application has two main phases [18]. The first phase is detect or identify the
misspell word. The second phase is to suggest the alternative word form in order to replace the
misspelled word.

1.2. Motivation

Now a day, some of foreign and Ethiopian universities start to teach Geez language and some
organization and individual are developing android application that contains ancient books and
bibles. Additionally, Geez course are offered in some primary schools. Therefore, the
development of spelling checker application for this language is required for easy preparation
of documents, notice and lecture note. The absence of a spelling checker tool for Geez
language has made document preparation activities difficult, needs excessive effort to check
and correct the misspelled from the document.
Previous attempt has made to develop a Geez language spelling checker. This attempt was
made using N-gram technique to solve homophone alphabets interchangeably error for Geez
language. In Geez some alphabets (ፊዯልች/fīdelochi) have the same sound but have
different shape and meaning like 3Ha‟s (ሀ/hā፣ ሐ/ ā፣ ኀ/ ā) se‟s (ሰ/se፣ ሠ/še) tse‟s
(ጸ/ts‟e፣ፀ/t s‟e) and A‟s (አ/ā ዏ/„ā) are homophone [5] [4].

Look at the following table that shows homophone and known name.

Letter Known name Reason

ሀ ሃላታው “ሀ” Since it is the beginning of the Geez word ሃላ
ሐ ሐመሩ “ሐ” Since it is the beginning of the Geez word ሐመር
ኀ ብዘኃኑ “ኀ” When the word ብዘኃን written it is the one used
ሰ እሳቱ “ሰ” When the Geez word እሳት written it is the one used.
ሠ ንጉሡ “ሠ” When the Ge‟ez word ንጉሡ written it is the one used
አ አሌፋው “አ” The word አሌፋ is always written using it
ዏ ዏይኑ “ዏ” The shape is like Eye and the Geez word ዏይን is written
ጸ ጸልቱ “ጸ” it word ጸልት written it is the one used.
The Geez
using
ፀ ፀሐዩ”ፀ” The shape is like sun and used to write the Geez word ፀሐይ
Table1. 1 Geez homophone alphabet

3
In Geez, these alphabets have their own meaning when we write a word [5] [20]. If the writers
uses this alphabet interchangeable, it consider as invalid word (ፀያፍ) [20].

Example

Instead of ንጉሥ/king/ the writer writes ንጉስ/king/

Instead of ሀገር/country/ the writer writes ሐገር or ኀገር are invalid (ፀያፍ) words.

Instead of መጽአ/he came/ the writer writes መፅዏ/metsia/

Therefore, the previous study worked by Aleka T [4] focus on to fix Geez homophone
alphabet interchangeably errors (ፀያፍ). However, interchangeably alphabets are not the only
spelling errors in the Geez language. The work proposed by Aleka T [4] did not cover Geez
abbreviations, Geez Non-word errors and other real word errors. There for performing Non-
word error detection and correction can improve the performance of the Geez language
spelling checker.

1.3. Statement of Problem

Currently, the world is digital world and people interact to digital computer and handheld
device. In this digital world, the data must organize in digital way. Therefor inputting data is
the critical activity for organize data in a digital way. Additionally, different organizations and
individuals attempt to develop different mobile application like machine translation. In this
case, a spelling checker is required to minimize error, save time and to motivate for all Geez
writers and readers can easily communicate without misinterpretation caused by spelling
errors. Look at the following figure that takes screen shot form mobile application .

4
 Figure1. 1 sample screen shot from android
As shown in the figure above the underline word is Non-word error. This type errors and
other type of homophone interchangeably errors makes the reader doubt to the developed
application. We proposed a system to solve this problem.
In Geez, there is no dictionary that covers all possible inflection, derivation and compounds
words. Because of it generates many words from a single verb. In practically, it is impossible
to store all words in dictionary. There for dictionary based approach is not suitable for
morphologically rich language. Error free readymade Geez corpora in electronic media also
not available so far, so corpora-based spell-checker is not suitable. Although some Geez text
(like bible) available in online, it contains a lot of Non-word and homophone error [4].The
rich morphological nature of the language makes a morphology-based approach more suitable
[21].
An attempt have already made on Semitic language using morphological information. like
[14] Merhawit S was done Amharic language spelling checker based on morphology,
Berhihun H. was done Tigrigna spelling checker using rule based and morphological analyzer
[15]. However, for Amharic spelling checker and Tigrigna spelling checker cannot be
applicable for Geez language. Since these languages did not incorporate some grammatical

5
rules of Geez and alphabet variation [5]. The main reason that makes it unique from the
previous Semitic language spelling checker is:-
1. The spelling checker application is language dependent:-The spelling checker
application for one language is does not work directly for another language.
2. The nature of the language: - Each verb in Geez can inflect in Gender (Masculine or
Feminine), Number (Singular or Plural), nearness or farness [22]. All of this represent
in different unique words. For example in English the word “you” represents may be
plural or singular, may be represent female or male. However, in Geez male and
female, plural and singular are representing in different unique word. Show the
following table.

Subject Plural and Singular Male and Female For Third person
English You You(may be plural or you(may be male or She(for female)and
singular) female) He (for male)
They (may be male or
female)
አማርኛ አንተ(you) አንች/አንተ(for አንች(for female)አንተ(for እሱ(for male) and
አንች(you) singular) male) እሷ(for female)
እናንተ(you) እናንተ(plural) እናንተ(for male and እነሱ(for male or
female) female)
ግዕዜ አንተ(you) አንተ(for singular አንተ(for male) ውእቱ(for singular
አንቲ(you) male) አንቲ(for female) male)
አንትሙ(you) አንቲ(for singular አንትሙ(for male) ይእቲ(for singular
አንትን(you) female) አንትን(for female) female)
አንትሙ(for plural ውእቶሙ(for plural
male) male)
አንትን(for plural ውእቶን(for plural
female) female)

Table1. 2 comparing Geez from Amharic and English

6
The above table shows the word “you” in English may be plural or singular, female or male
and the word “እናንተ” column 3 and row 3 in Amharic may be female or male. However, in
Geez plural and singular, female and male are representing in different word. Look column 3
and row 4. This indicates Geez language have complex morphology from English as well as
from sematic language, Amharic. Look at the following table the word “ ነበረ”/nebere.=ተቀመጠ
In Geez Subject In English In Amharic

ነበርኩ Neberku I I sat ተቀመጥኩ

ነበርነ Neberne We We sat ተቀመጥን

ነበርከ Neberke You(singular male) You sat ተቀመጥክ

ነበርኪ Neberki You(singular female You sat ተቀመጥሽ

ነበርክሙ Neberkimu You(plural male) You sat ተቀመጣችሁ

ነበርክን Neberkin You(plural female) You sat ተቀመጣችሁ

ነበረ Nebere He He sat ተቀመጠ

ነበረት Neberet She She sat ተቀመጠች

ነበሩ Neberu They They sat ተቀመጡ

ነበራ Nebera They They sat ተቀመጡ

Table1. 3 Geez inflected in 10 pronouns

This table shows in English 6 unique subject (I, we, you, he, she, and they) represent 10
unique Geez words (ነበርኩ፤ነበርነ፤ነበርከ፤ነበርኪ፤ነበርከሙ፤ነበርክን፤ነበርነ፤ነበርት፤ነበሩ፤ነበራ) and 7 unique
Amharic words. This indicates Geez language have high inflection.
3. The existence of phonetic redundancies could create spelling errors;- The third
reason that makes it unique from the previous Semitic language spelling checker is, as
mention above in Geez some alphabet (letter) have pronounced the same but different
shape and meaning. This letter also uses to write Amharic text. Even though there is a
convention to write Amharic text [14] . Currently, most of the people use this alphabet
7
 interchangeably to write Amharic text. For instance, the word tsahay/ፀሐይ can be typed
like ጸሐይ, ፀሐይ, ጸሃይ or ፀኻይ. However, according to Merhawit [14] there is a
convention to write the word tsahay start by ፀ. For Geez use these letters as
interchangeably considers as invalid (ፀያፍ) [20] [4] or change the meaning of the word.
For instance, the word ሰረቀ and ሠረቀ are the same pronouncing but different in meaning
and shape. ሰረቀ/sereqe means he/she stolen and ሠረቀ/sereqe means something raises.
4. Some Geez word written in combination of alphabet and number: In Geez, some
words the collection of Geez number and alphabet, to represent number and
abbreviation. For example ፲ወ፪ for 12 and ፳ኤሌ for እስራኤሌ, ፪ኤ for ክሌኤ. This is also
another challenge to develop spell checker application for Geez language. The research
conducted on Semitic language previously not fill the gap, that mentioned in the above
four challenge .There for the proposed morphology-based Geez spell checker solve
the challenge that mention in the above,

At the end of conducting this research, the following basic research questions will have
answered

 To what extent spelling errors in Geez can be detected using dictionary lookup and
morphological analyzer?
 To what extent spelling error in Geez can be corrected using replacement rule and
Levenshtein edit distance (LED) algorithm?

1.4. Objective

1.4.1. General Objective

The main objective of this study is to design and implement interactive spelling checker for
Geez language using morphology-based approach.

1.4.2. Specific Objective

1. To explore Geez word formation and review related documents for Geez language
spelling checker.

8
 2. To build the lexicon and affix file of Geez language
3. To design the General architecture of the proposed system
4. To develop prototype of the proposed system
5. To conduct an experiment(on both error detection and error correction components)
6. To evaluate the performance of the system

1.5. Scope and Limitation of the Study

It is important to identifying the Scope or boundary of the study. In this study, the researcher
attempt to collect the data and analyze different related documents in order to design and
implement the Geez language spelling checker by considering internal Inflection of and
compound words. The proposed study mainly focuses on Non-word error detection and error
correction of Geez language. Non-word error is one class of error. It may be multiple/ single
insertions, deletions, substitutions, transpositions in single word. After developing the system,
the performance of the system has been evaluated.
Geez language has some common abbreviations like ፳ኤሌ for እስራኤሌ and ፲ቱ for ዏሥርቱ/10.
There for this study consider for this abbreviations. However, real world error have not
include in this scope. The first reason to exclude real word error is shortage of time to solve
real world error or contextual error. The second limitation is lack of well-organized material in
Geez language.

1.6. Significant of the Study

Spelling checker is a tool that used for everyday activity in today‟s generation. Students,
teachers, organization and anyone who wants to learn and prepare documents use spelling
checker application. In order to save time and reduce error in document preparation spelling
checker is mandatory for every language. The proposed study offers different benefits from
different perspectives. Some of the benefits are listed below.

 To motivate for all Geez writers and readers can easily communicate without
misinterpretation caused by spelling errors and easily understand Geez literature.

9
 Since, there are so many ancient literatures and book written in Geez language. This
literatures and books used as sources of philosophy, creativity, knowledge and
civilization both to Ethiopia and the rest of the worlds [6].
 To organize the ancient literature and books in digital form with in less time and free
error
 For academic exercise in preparation of handout, report, assignment and lecture note.
Since Geez course are offered in various colleges, religious centers and higher
institutes
 It uses an input for Researchers who want to conduct on higher-level application of
NLP for Geez language such as Parsing, grammar checker, speech recognition,
Question Answering, machine translation, text –to speech translation.
 To minimize spelling error and save time during document preparation
 Geez language is widely used in every activity of the Ethiopian orthodox Tewahedo
church (EOTC) in worship of God. Holy books like bible, anaphora/kidassie/, psalm,
miracles, biography of martyrs, wudasie Maryam are reading in Geez language. These
ancient religious books are duplicated to address the newly established churches and
monasteries. Therefore, correcting Geez spelling error is used to preserve the dogmatic
and canonical faith of the religion.

1.7. Methodology

In order to achieve the research objective and provide valid and reliable result identifying the
right methodology is a crucial step [14]. There for to conduct this study the following methods
and techniques we have used.

1.7.1. Literature Review

Literature review has a vital role for finding the gaps, for exploring the comparison of
different approach, detail understanding of the problem, identifying the methodology etc. We
have reviewed global and local relevant literature Such as, book, article, conference paper,
report and relevant resources from internet. In addition to this, the researcher assessed Geez
language components such as phonemes, word classification and word formation.

10
 1.7.2. Data Collection

So far, there is no readymade Geez word dictionary for Geez language spelling checker
(GLSC). Therefor the researcher has constructed two dictionaries. One main dictionary
contains stem word and another dictionary, which contains affix files. For this study the
dictionary built from ሉቀ ሉቃውንት ያሬዴ ሽፈራው መጽሐፍ “መጽሐፈ ግስ ወሰዋሰው መርኆ መጻሕፍት”(like
likawunt Yared Shiferaw “metsihafe gis wesewasewu merho metsahift) [23].The testing data
also directly collected from bible. In addition to this, the researcher communicated with the
linguistics experts to understand the structure of the word, inflection and characteristics of the
language words.

1.7.3. Implementation Tool

For implementing the proposed system, the researcher used Python programming language for
designing prototype of GLSC. The major reasons for choosing a Python programming
language are: the first reason, it is an object oriented programing language which is
suitable and flexible for text processing and NLP applications. Second, the researcher has a
good programming knowledge and skill in the selected programming language. Additionally,
the researcher used another software and tool. Like

Sublime text editor:- used for developing dictionary and affix file.

Microsoft word office 2010:-for document preparation

Hunspell:-a tool used for morphological analyzer. It is a spell checker and morphological
analyzer library and program designed for languages with rich morphology and complex word
compounding or character encoding [24].

1.7.4. Evaluation Metrics

To determine the functionality and performance of the system, testing is a mandatory task. In
order to measure the performance of the system, we have used recall and precision for spelling
error detection and for suggestion, we have used suggestion adequacy (SA).

11
 1.8. Organization of the Thesis

This study organized in six different but interrelated chapters. The first chapter starts by
discussing the background of the research, motivation of the research, statement of the
problem, objective of the study including general and specific, methodology of the study,
scope and limitation of the study and significance of the study are discuss in detail.
Chapter two discusses basic theoretical concepts about the spelling checker, spelling error
types, existing spell checking approaches and spell checking technique for Semitic language
and other languages and some of the previous related works.

Chapter three discusses Geez language in detail. By including, explain Geez writing system,
identify Geez word classification, explore Geez word formation and identify Geez word
inflection. Like internal Inflection of words

Chapter four discusses the general architecture and design of the proposed system, the
designed Geez spelling error detection and correction algorithm and the rule definition
structure designed as an input for the algorithm. Moreover, stored knowledge base
components used by the prototype are discusses detail.

Chapter five discusses the testing data collection and preparation for experiment, conduct
experiment and measure the performance and discus the result. Chapter six discusses the
conclusion of the whole work and future work.

12
 CHAPTER TWO

2. Literature Review

2.1. Introduction

This chapter deals with the basic theoretical concept of type of spelling checker in written text,
review previous work for spelling checker in local and foreign language, understand existing
spelling checker approach.

2.2. Overview of Spelling Checker

Spelling checker is tools that identify misspelled word in a paragraph or in a sentence and
provide an alternative word to correct the misspelled words. The spelling checker application
is a basic requirement for any language to be digitized [25].In addition to this, the spelling
checker application has a great role to develop higher NLP application. Like machine
translation, question answering, Entity named Recognition, Text-to Speech (TTS) system,
Automatic Speech Recognition (ASR) system e.t.c [12].

Type of spelling error

Many researchers try to develop spelling checker application using different method for
different language. The error encounter in text may be either Non-word error or Real word
error based on the meaning that gives [14].

Non-word error occurs when the user writes incorrect or misspelled word [10]. This type of
error does not give any meaning. For example, “we iss coming”. The word “iss” is invalid
word.

Real word (semantic) error occurs when a user writes correct word but not give meaning in
the context. It is difficult to identify the word is wrong or correct without some contextual
information [10]. For example, “we is coming”. The word “is” is correct word in English
language. But not have a meaning in the context of the above sentence. Because of in the

13
above sentence the subject/ we/ indicates plural person. But the verb “is” used for singular
subject.

According to Merhawit sh [14] the spelling error also grouped in to three based on the error
patterns produced .These are topographic error, cognitive error and phonetic error.

I. Topographic error

This type of error occurs the correct spelling of the word known by the writer but the writer
make mistake unknowingly. The cause of this error mostly related to wrongly key press [13].
This type of error may be encounter in single character of a word or in multiple character of a
word [14]. Typographic errors fall into one of the following four categories, such as
substitution error, deletion error, insertion error, and transposition error.

II. Cognitive error

This error occurs when the correct spelling of the word is not known. In cognitive error the
pronunciation of misspelled word is the same or similar to the pronunciation of the correct
word [17].Cognitive error encounter especially in languages having homophone alphabet [4].

III. Phonetic error

Phonetic error one class of cognitive error [26].It occurs when a user substitute a character
with its phonetically equivalent character [14] [4].

2.3. Technique of Spelling Checker

The spelling checker application has two main phases. This are error detection and error
correction phases [18].Error detection concerns to identify wrongly spelled words in the
paragraph or checking either the word is found in the dictionary or not. Whereas error
correction concerns to provide suggestion or an alternative word in order to correct the
misspelled word.

14
 2.3.1. Error Detection

Spelling error detection is process of identifying misspelled words from a sentence or

paragraph by using error detection technique. Dictionary lookup and N-gram analysis are the
most common existing error detection techniques [10] [4].

A. dictionary lookup

Dictionary lookup technique used to check every word of input text present in the dictionary
or not. If the word presence in the dictionary, then that word consider as correct word.
Otherwise, put in to the list of error words.

All words with their inflection and derivation of the word store in the dictionary then, check
each input text from the dictionary takes many times. As result, the dictionary looks up
technique needs large amount of memory size and may take long time to search. Therefore,
the drawback of this technique is needs more space and inefficient search time [15]. Because
of practically, it is impossible to store all word with their inflection and derivation for
morphologically rich language. The most significant searching techniques for dictionary
lookup are hashing, binary search and finite state automata [27] [14].

Hashing involves searching for an input string from a pre-compiled hash table via a key. In the
context of spelling checker, if the word stored at the hash address is the same as the input
word, the input string marks as correct. However, if the input word and the retrieved word are
not the same or the word stored at the hash address is null, the input word marked as a
misspelled. Hashing technique provide fast access [26].

Another predominant method in dictionary lookup technique is binary search technique [28]
[29].This method performs by using median split. Median split tree used to access high-
frequency words faster than to low-frequency words without sacrificing the efficiency of
operations.

Finite State Automata (FSA) [14] method is another dictionary lookup method. This method
used to representing a language, as a set of strings with a sequence of notations of some
symbol or script [30].

15
 B. N-gram technique

N-grams are a method to find misspelled words in the text [31].It is a contiguous sequence of
n items from a given sample text. These items can be phonemes, syllables, letters or words
according to the application. Where n€ {1 3…}.If n = 1 3 they are referred to as
unigrams, bigrams, trigrams respectively.

Instead of comparing each entire word in a text to a dictionary, just n-grams are controlled. N-
gram technique is work by examining each n-gram in an input string and looking it up in a
pre-compiled table of n-gram statistics to ascertain either its existence or its frequency. The
advantage of n-gram algorithm they do not require knowledge of the language. Often it is
called language independent [15].

2.3.2. Error Correction

After error detecting, the next step in spelling checker is error correcting. Spelling error
correction is a process of generating or suggesting closer words to the misspelled word, ranks
the suggested words and replaces the misspelled word with the correct one [14].There are
two type of spelling error correction [26].These are automatic and interactive error
correction. Automatic error correction is the system decides one best correction and
automatically replaces the misspelled word. However, interactive error correction the system
provide one or more suggestions and the user select one of them. Some of the existing error
correction techniques are minimum edit distance, rule-based, similarity key, n-gram-based,
probabilistic, noisy channel and neural networks techniques [14] [26].

A. Minimum Edit Distance Technique

This technique stands on counting the minimum number of editing operations (i.e., insertions,
deletions, and substitutions) required to transform one string into another. As stated by
KAREN KUKICH [27] the first edit distance algorithm was implement by Damerau. This
technique is a simple and the most effective technique to generate the suggestion. Minimum
edit distance technique compute the minimum edit distance between a misspelled strings and
dictionary entry. It have applied to virtually for all spelling correction tasks, including text
editing, command language interfaces, natural language interfaces, etc. However, it is not

16
quite as good for correcting phonetic spelling errors [32].Given two-character strings s1 and
s2, the edit distance between them is the minimum number of edit operations required to
transform s1 to s2.Minimum distance technique perform by using different algorithm
[15].Some of this algorithm are Levenshtein algorithm, Hamming, Longest Common
Subsequence [33].

a. The Levenshtein algorithm: This algorithm is a weighting approach to appoint a

cost of 1 to every edit operations. For instance, the Levenshtein edit distance
between “cat” and “dog” is 3 (substituting “c” by ” d” , “a” by “o”, ” t” by “g”
).
b.The Hamming algorithm: This algorithm used to measure the distance between two
strings of equal length [34]. For instance, the hamming distance between “bad” and
“bag” is 1.Becuase of the corresponding character mismatch (g and d).
c. The Longest Common Subsequence algorithm: The longest common
subsequence algorithm is a popular technique to find out the difference between
two words. The longest common subsequence of two strings is the mutual
subsequence.
B. N-gram Technique

As stated by [35] N-gram analysis can work in two ways, either together with dictionary or
without dictionary. When n-gram implemented without a dictionary, used to find in which
positions the misspelled word occurs. If there is a unique way to change the incorrect word so
that, it contains only valid n-grams, this is taken as the correction. The performance of this
technique is limited. It is simple and does not require any dictionary. Together with a
dictionary, n -grams are used to define the distance between words, but the words are
always checked against the dictionary [35] [26] [31] [25].

C. Rule Based Techniques

Rule based technique works by defining a set of rule for common spelling errors patterns in
the form of rule for transforming misspellings into valid words [14] [27]. This technique needs
language expert to define the rule of the language. If a misspelled word falls in the set of

17
common spelling error patterns, candidate words will generated by applying all applicable
rules that can be applied on the identified error pattern. Some of the drawback of rule-based
technique is, it needs plenty of manual effort and requires complete grammar rules to cover all
types of sentence constructions.

D. Similarity key

The similarity key technique works by mapping every string into a key such that similarly
spelled strings will have identical or similar keys. This technique works based on transforming
words into similarity keys that shows the relations between the characters of the words. Thus,
when a key was compute for a misspelled string it will provide a suggestion to similarly words
in the dictionary words. The importance of this similarity technique is, speed [27]. Because it
is not necessary to directly compare the Non-word to every word in the dictionary. This
similarity can be positional similarity, material similarity, and ordinal similarity [30] [14].

Positional similarity:-This similarity shows the degree of the matching character in two
strings is in the same position.

Material Similarity:-This similarity shows two strings consist of exactly the same characters
but the order is different.

Ordinal similarity: - This similarity shows the matching characters in two strings are in the
same order. Similarity key techniques can found in the Soundex system. Soundex system has
invented to solve the problem of phonetic errors [30].

E. Neural network

Another interesting technique and needs in small dictionary word is neural network [32].This
technique are potential candidates for spelling correction due to their ability to do associative
recall based on incomplete or noisy input. Neural nets have the ability to adapt to the specific
error patterns of a certain user's domain because they can be trained on actual spelling errors.
Hence, the correction accuracy for that domain is potentially improved.
The back propagation algorithm is widely used in order to training the neural network. Back
propagation has 3 layers. These are input layer, hidden layer and output layer. The nodes

18
 within a given layer are not directly connected. However, each node in the input layer is
connected to every node in the hidden layer by a weighted link and each node in the hidden
layer follows the same fashion and is connected to every node in the output layer by a
weighted link. The activities on input and output nodes of the network indicate input and
output information respectively. In this input information is represented by on-off pattern (i.e.
when a node is turned on, a 1 is indicated; when a node is turned off, a 0 is indicated [30]).

In spelling checker, a misspelled represented as binary in n-gram vector may be taken as input
pattern. The output pattern is the activation of some subset of m output nodes, where m is the
number of words in the dictionary [30] [26].Hidden layer connects with the input and output
nodes.

F. Probabilistic Techniques

Probabilistic technique works based on statistical characteristics of the language [32]. There
are two types of probabilities [30] [33]. These are transition and confusion probabilities.

Transition or Markov probabilities:- They represent a certain given letter will be followed
by another given letter. These are language dependent. These probabilities can be estimate by
collecting n-gram frequency statistics from a large corpus.
Confusion or error probabilities:- estimate the probabilities that a certain letter substitutes
another given letter, given that an error has been made. This probability is source dependent.
Because different OCR devices have different techniques and features in order to recognize
Characters. Each device will produce a unique confusion probability distribution [30].
G. Noisy Channel
This model assumes a natural language text passes through noisy communication channel
[14].

2.4. Performance Evaluation Technique for Spelling Checker

Evaluation technique is a technique used to measure the performance of the proposed system
or the model [29]. Different researchers have proposed various criteria for evaluation of a
given spelling checker. The existing common performance evaluation metrics for spelling

19
checker is recall, precision and f- measure [29].Suggestion adequacy (SA) also another
evaluation metrics, used to measure the performance of the spelling error suggestion [36].

Recall: - is a measure of completeness.

Lexical recall (Rc): is defined as the number of valid words in the text that are recognized by
the spelling checker (i.e. true positives), in relation to the total number of valid words in the
text (i.e. the sum of all true positives and false negatives)

Rc= ………………………. …….(3.1)

Error Recall (Ri): is defined as the number of invalid words in the text that are flagged by the
spelling checker (i.e. true negatives), in relation to the total number of incorrect words in the
text (i.e. the sum of all true negatives and false positives)

Ri = ………………………………….(3. )

Precision:- indicates exactness of the spelling checker in catching up and responding to the
erroneous words.

Lexical Precision (Pc) is computed by dividing all correct non-flags (true Negative) by the
total number of non-flags (i.e. true positives plus false positives)

P c= …………………………………..(3.3)

Error Precision (Pi): is defined as the number of correct flags (true negatives) in relation to the
total number of flags assigned by the spelling checker (i.e. true negatives plus false negatives)
gives an indication of the spelling checker.

Pi= …………………………………………(3.4)

Where TP- True Positive, TN- True Negative, FP- False Positive, FN- False Negative(see
chapter five section 5.4.1 for detail)

20
F-measure: this also another performance measure that combines Recall and precision into a
single measure of performance, this is just by taking into account the product of Precision and
Recall divided by their sum

𝑭-𝒎𝒆𝒂𝒔𝒖𝒓𝒆 = …………………………………..(3.5)

2.5. Related Works

This section explains and summarizes the previous works on spelling checker. Many
researchers tried to develop spelling checker application for both global and local language in
order to minimize spelling error problem.

2.5.1. Spell checker for foreign language

2.5.1.1. Arabic Spell Checker

The authors of this study try to develop for an Independent Spell-Checking System from the
Arabic Language Vocabulary. This study introduces morphological analysis concept in the
Levenshtein algorithm using small size dictionary that contains Arabic language stem words
[37].These paper are mainly focus on comparing and evaluating their method with
Levenshtein approach.

In order to compare the new approach with Levenshtein‟s the researchers focus on the
following three indicators. Those are-The correction average time, the rate of rectified words
and the size of each system lexicon. According to his experiment the average correction time
for the proposed approach is faster than the average correction time for Levenshte in distance
is , the average correction rate by the proposed system is 33.7% higher than Levenshtein
distance and the size of the lexicon in proposed system less than that of Levenshtein distance.

The authors concluded that the result of proposed approach highly positive and much better
than that of Levenshtein distance. Due to this fact, the new approach appreciates the validity
of new concept and shows the importance of their new approach.

21
 2.5.1.2. Spelling Checker for Marathi

These authors [21] try to develop morphology-based spelling checker for Marathi language.
For Indian languages such as Marathi and Hindi have complex morphology. For example, for
a single noun in Marathi, over 200 words are formed that are either adjectives or
adverbs. Similarly, a verb may exhibit over 450 forms [21]. Due to this reason, for
morphologically complex language not maintain in the dictionary-based approach. Because it
is impossible to store all root words with their possible inflection word in the dictionary. In
addition to according to the author, there is no readymade corpus for this language. Due to
this reason, corpora-based spelling checker may be achieving poor performance. The author
collects 13,000 root words and classifies in to part of speech, and then apply morphological
rule for all part of speech.

The author explain morphological rule for all class of part of speech such as postposition
morphology, noun morphology, pronoun morphology, verb morphology, adjective
morphology, Adverb morphology, Conjunction morphology and Interjections morphology.
The performance of this work is 99.7%.The author conclude that the reason for traced this
error is missing incomplete vocabularies and exceptional cases.

2.5.2. Spell checker for Ethiopian language

2.5.2.1. Afaan Oromo Spelling Shecker

a. Design And Implement Morphology Based Spell Checker for Afaan Oromo

Gaddisa Olani [9] try to design and implement of morphology based spell checker for afaan
Oromo language.

The author uses morphology-based spell checker that combines dictionary look-up with
morphological rules. The system architecture of this study contains Tokenizer, Error detection,
Error correction, knowledge base, Morphological analyzer, Morphological generator,
Suggestion ranker and Word Assembler components.

22
The tokenize component divide block of text in to individual words, digits and punctuation
marks. The knowledge base component stores lexicons, affixes and rules of the language.
Error detection receives a word from the word tokenize and looks for the word inside the root
word. If the word found in the root word, it marks as correct word and no need to further
process. If the word not found in root word, pass to Morphological analyzer. The
Morphological analyzer components accept list of words from error detection and decompose
in to root word and affixes and then pass to the error detection to check whether the
decompose root word and affix are valid or not.

After affix stripping in morphological analyzer the error correction component, classify either
the error encountered from affix or from root word. Then, make closer correction based on
their class using LED algorithms. Morphological generators rebuild the decompose parts of
words (morpheme). The word assembler to combine correct word with the one flagged as
misspelled word. Finally, the evaluation of the proposed system resulted in 100% error recall,
88.62% lexical recall and 28.62% precision.

2.5.2.2. Amharic Spelling Checker

a. Automatic Amharic spelling error detection and correction using hybrid

approach

Getinet A [18] developed spelling checker for Amharic language. This researcher tries to
identify non-word error in the document and after identifying, correct them. To achieve his
objective the author uses the Metaphone algorithm for error detection and edit distance
algorithm for error correction. The main component of this study is data preprocessing,
spelling error detection, spelling error correction and ranking.

The author use python programming for developing prototype that works single word at a
time. In addition to this, the author uses VB programming language for integrating spelling
checker with word processing. The prototype accepts a candidate word and checks the validity
of the words against the dictionary of words. If it the word not found the candidate word mark
as misspelled and the system will suggest closer words from the dictionary using edit distance
algorithm.

23
The author collects about 125,000 words in order to construct dictionary and 500 words for
testing. From this, 100 of them misspelled words. After prototype development the whole
system tested by using testing data. Based on this, the author gets 98% effectiveness in its
system error detection and correction functionalities.

The author concluded that the proposed system is effective and accurate in its functionality.
However, the processing speed was not considered in this study. It needs improvement for
future research. Then, the author recommends improving the processing time and integrating it
with other Microsoft office systems in future works.

b. Multilingual Spelling Checker for Selected Ethiopian Languages

Wubetu B. [16] try to develop Multilingual Spelling Checker for Selected Ethiopian
Languages such as Amharic, Afan Oromo, Tigrinya, Hadiyyisa and Awngi. This work done
by using dictionary based approach and hashing. The proposed system marked as the
candidate word is correctly spelled word; if the candidate word matches the word at the hash
address. Otherwise, the candidate word is incorrect. The author build corpus-based dictionary
for each language. This author collect 993,072 for Amharic,866,328 for Afan Ormo,966,328
for Tigrigna,987,176 for Hadiyyisa and 676,534 word for Awngi.

The performance of the system measured using evaluation metrics precision, recall and F-
measure. There for the average performance of the system is 81%. The author concluded that
the suggested approach is able to detect diverse classes of spelling errors in the mentioned six
selected languages.

c. Automatic Spelling Checker In Amharic language

Melkamu T [10]try to design and develop an automatic Amharic spell checker integrated with
Open office word processor. In this study the author, mainly focus on non-word spelling error
detection with considering internal inflection of words, repeated words and compound words
of Amharic word. In this paper, the spelling error correction is not included.

The author uses Hunspell tool for morphological analysis. Hunspell is morphological analyzer
library and program designed for languages that have complex morphology. According to this

24
study, the spell checker has five components this are, input component, normalization, error
detection, morphological analyzer, error correction and suggestion components.

The input component focus on tokenize the text file in to word and remove punctuation from
file. Normalization identifies characters with syllographic redundancy or homophone
alphabets that can be used interchangeably. Example ሀገር፣ሐገር፣ኀገር must change in to one
common word.

Error detection this component search the candidate word inside the dictionary. If the word is
found, it consider as valid word. Otherwise, it will be pas to morphological analyzer
components. Morphological analyzer component receives the word that is not found in
dictionary from the previous component, splits it into stem, and affixes. The author
recommended to future research; need to full of error detection and error correction of
Amharic word. However, in this paper, the author not conducts experiment on error correction
component.

In order to evaluate the system, the author conducts five experiments. For the first experiment,
the text had taken directly from Amhara Science Technology and information communication
commission 2009 annual report. From this, source 199 Amharic words out of which 5(2.5%)
words are misspelled. From this experiment, all misspelled words are detected by the system.
and one correct word mark as misspelled.

For the second experiment had taken data directly from Amhara national regional state science
technology and information communication commission annual report and from this source
the author collected 840 Amharic words out of which 16(1.9%) words are misspelled. The
result of this experiment all misspelled words are detected by the system and 20(2.4%) correct
words are marked as misspelled. The third experiment, 181 words taken from afar Region ICT
2009 annual report out of which 7(3.9%) words are detected as misspelled. From this, all
misspelled words are detected but 3 correct words are marked as incorrect.

The fourth experiment, 94 words taken from Harari Region ICT 2009 annual report out of
which 9(9.5%) words are detected as misspelled. The result of this experiment all misspelled
word is detected and 9.5% correct word are marked incorrect. The fifth experiment, 94 word

25
taken directly from experiment 4 evaluated by language expert out of which 7(7.4%)
misspelled. From the last experiment precision and recall 97.75 and 92.55 respectively. The
average performance of the system shows 97.4%

d. Amharic spelling detection and correction system morphology based approach

Merhawit Shi [14] present spelling checker for Amharic language using morphology based
approach. This work is focus on Non-word spelling error detection and correction for Amharic
word. In this work, rules are generated by analyzing the morphological behavior of the
language. To formulating the rule for the morphology of language needs linguistic expert
[15].The author develop morphological rule manually. This rule contains different parts of
speech like nouns, adjectives, verbs, pronouns, and compound words.

The architecture of the system contains three components and three. These are preprocessors,
error detectors, and error correctors. In addition to that, the architecture contains three
knowledge bases. Those are morphological rules, dictionary lookup, and a database holding
Amharic characters. The dictionary is one of the knowledge bases that stores stem words of
Amharic language. The stem word may be noun, verb, adverb or adjective. If searching a word
in the dictionary is empty, then the candidate word split into n-grams (unigram, bigram, and
trigram) and checked in the morphological rules, which are used to derivate and inflect a word
into different forms. All morphological rules and affix files are stored in morphological rules
knowledge base. The author develops 717 morphological rules. The Amharic character
database stores every Amharic character.

Preprocessing is the first phase of the architecture. This phase contains tokenization.
Tokenization works by splitting block of text in tow word, punctuation mark removal except
hyphen used for identifying compound word and replicated word removal. The second
component is error detection component. Under error detection component three main
activities were performed. Those are N gram splitter, stem extractor and rule filter. The error
detection component receive candidate token (word) from preprocessing component and
divide in to N-gram (unigram, bigram, and trigram) and comparing them with the defined affix
rules. After that select the rule and extract the probable stem word. If the derivation of the root

26
word matches with the rule, then the word is flagged as valid otherwise it will be an invalid
word.

The last component is the error correction component. This component have responsible to
provide suggestions for the misspelled word by taking rule weight, location of the error, and
the invalid word itself from the stem extraction and rule filter.

The challenge of this approach is time consuming to test every possible scenario in the
language. Moreover, processing on a single word passes a long path starting from searching
from dictionary, if not found split it into n-grams and comparing each n-gram with the
derivation rules containing suffixes, prefixes, conjunctions and prepositions.

Finally, the author measures the performance of the spell checker prototype by using
precision, recall, and accuracy. In order to test and evaluate prototype the author prepare two
test cases. The first test case contains1551 valid words and 173 invalid words. The second test
case contains 88 unique words Amharic bible. The result obtained from the evaluation is 96%
precision, 99% recall, and 95% predictive accuracy.

2.5.2.3. Tigrigna Spelling Checker

a. Spell checker for Tigrigna language using rule based morphological analyzer and
unsupervised approach

Berihun Hadi [3] try to develop spelling checker for Tigrigna language using rule based and
morphological analyzer and unsupervised approach. The author performs two experiments.
The first experiment conducted on rule-based dictionary look up and morphological analysis.
Even though the language and size of data is difference, the dictionary and affix file
preparation of this experiment is similar Gaddisa Olani as mention above [9] . Knowledge
Base builds by adopt Hunspell dictionary and affix file format. It contains all information
(morphological rule) and dictionary file.

The second experiment also conducted on unsupervised Morphological based approach using
Morfessor tool. Unsupervised approach learning is a type of machine learning algorithm used
to draw inferences from datasets consisting of input data without labeled responses. In order to

27
conduct the second experiment the architecture containing, input component, morfessor tool,
morfessor model, Viterbi algorithms, morfessor segmentation model, Knowledge base.

The input component is the contains Tigrigna text. morfessor tool: is a tool that used for
morphological segmentation of a word in unsupervised machine learning. morfessor mod,
Viterbi algorithms, morfessor segmentation model, Knowledge base in this experiment
knowledge build from morfessor segmentation mode.

The author uses ½ million words collected from different source and extracted using Text
STAT software. From this word use 787 words for testing the system and conduct two
experiments as mention above. Based on the result of the experiment rule-based approach out
performs well than the unsupervised approach. The author conclude that increasing the
dictionary and morphological rule to the knowledge base and increasing corpus size to the
unsupervised approach will increase the performance of the proposed methods.

2.5.2.4. Geez Spelling Checker

a. Context-Dependent Spelling Error Detection and Correction For Geez Language

Aleka T. [4] Try to developed Context-Dependent Spelling Error Detection and Correction for
Geez language. In my knowledge, this work is the first work for Geez language spell checker.
This study designed by using design science research methodology. The study focuses the
problems of using homophone Geez alphabets incorrectly, that they bearing completely
different meanings in a given sentence. In Geez, some alphabets have the same sound different
shape and meaning called homophone alphabet [7] [4].In order to meet the objective of this
paper, the author uses N-gram techniques. The proposed context-dependent spell checker
architecture contains three main components: text preprocessing, error detection, and error
correction.
Text processing component responsible to remove special character, punctuation marks, geez
numbers and Arabic number. In addition to split in to sentence level using full stop (።) and
question mark (፧ ?). For each sentence, a statement beginning (<s>) and ending with (</s>) are
added to differentiate a sentence from another. In addition to that, the segment sentence split
into different tokens to create unigram, bigram, and trigram using white space as a separation
mark.
28
In the error detection component the author proposed and implemented an n-gram (bigram and
trigram) language model to detect and correct context dependent spelling errors that occurred
in the Geez language. To detect the error, the error detection algorithm checks the presence of
every trigram [𝑤i-1, 𝑤i, 𝑤i+1] and bigrams [𝑤i-1, 𝑤i] by looking up into a precompiled
trigram and bigram language model. Back off smoothing technique is use to eliminate the zero
probabilities of unseen n-gram and estimates the probabilities higher-order model by also
making using of information from the lower-order (n-1 gram) model.
In error correction component provide suggestion by ranking them based on their probability
values. The most probable word pops at the top. Firs algorithm checks whether the word
contains homophone Geez alphabets or not and the word is correctly spell or not based on
bigram and trigram language model. Then, the algorithm generates all possible word
combinations (bigrams and trigrams) and computes the most probable correction.
The performance of the system has measured by precision and recall. From 16918 sentences
for training 650 sentences randomly selected for testing data. The author conduct bigram and
trigram experiment. Based on his experiment the author concludes that the trigram-based
context-dependent spell checker scored better performance than the bigram-based experiment.

2.5.2.5. Morphological Analyzer for Geez language

a. Morphological analysis for Geez language using memory based learning

Ytayal Abate [38] try to develop a morphological analyzer for Geez language using
one of the supervised machine learning approaches known as memory-based learning. The
proposed system is focus on only Geez verbs and trained with manually annotated sample
verbs.

The proposed system architecture has two phases. This are training phase and morphological
analysis phase. In the first phase of the architecture involves morpheme annotations, to detect
the morpheme from a set of words. Inflected words are segments by prefix, stem and suffix.
Each segment marked by their representative notations. In Geez, prefixes indicate many
features [7]. Such as negation marker, preposition marker, conjunction marker, Causative
marker, causative-reciprocal marker, Reflexive marker and reciprocal.

29
The representative letters represent those features. The annotated words are stored in a
morphological database. Then, the instances of the annotated word extracted from the
morphological database using windowing method. Then, a memory-based learning model is
developed using a learning tool TiMBL. Classifier algorithms called TRIBL2 and IB1 are used
to construct databases in memory. The suffix also indicates for two features in geez language.
These feature either subject marker or object marker.

Once the learning model is developed, the second phase (morphological analysis) continues.
In this phase, a new word is analyzed based on previous training data. The word is segmented
and represented as instances and they are compared with the training set. The word will be
classified as the closest instance. If the inflected word is unknown, its morpheme is identified
and the class that shares the most common features is inferred and predicted as the class of the
new instance.

The next step is reconstructing the given word into meaningful units. The system searches for
similar stem patterns from the stored training set. If it is not found, distance to the most nearest
neighbor is calculated. The last step is extracting the root from the stem. This involves
removing the vowel in stems for words with more than three characters.

The author annotated manually 1105 verbs to be suitable to TiMBL algorithms. From these
annotated verbs, the author extracted 12135 instances automatically. This data set was divided
into training and testing data from which 90% for training and 10% for testing.

Generally, the author concludes that IB2 is good at memory usage on both default and
optimized settings (with 91.72% and 93.24% accuracy). However, it has low processing
speed, which in turn takes more time. On the other hand, TRIBL2 algorithm achieves a little
bit different from IB2. It performs 91.19% and 92.31% with default and optimized parameter
settings respectively.

30
 2.5.3. Summary of Related Works

Title author, Approach/ Strength Limitation

Year methodology
Automatic Melka -Dictionary looks up -The study considers the -Real word error
spell checker mu T. technique. internal inflections of words, detection and
for Amharic ,2020 -The Hunspell tool is compound words, and correction are not
language [10] an open-source spell repeated words. considered.
checker with a -Even though
-Developing affix (prefix,
morphological developing an
infix, and suffix) rules:
analyzer library. automatic spell
inflected words are generated
checker is the
by applying affix rules on the
researcher‟s aim it
root words. This minimizes
is not implemented
the problems of requiring
spelling correction.
extremely large spaces,
searching time, and
impracticability of listing all
correct words to develop a
dictionary.

-Integrating the spell checker

into an open office word
processor

Amharic Merha -Morphology-based Define morphological rules -Detecting and

Spelling Error wit S. , approach: for to derive and inflect Amharic correcting real word
Detection and 2010 detecting misspelled stem words into different errors are not
Correction words and generating forms. included.
System [14] suggestions. .
-Distance calculator -morphological rules

31
 -N-gram to split words algorithm used to find closer are prepared only for
unigrams, bigrams, stem words verbs. However,
and trigrams other parts of speech
like nouns,
adjectives, pronouns,
and compound
words are not
considered.

-Defining a
complete set of rules
for morphologically
complex language is
an inefficient task.

Multilingual Wubetu -Dictionary-based -The system is able to -The system not well
Spelling B., approach with hashing identify diverse classes of performed for the
Checker for 2020 spelling errors. languages that have
selected -Preparation of corpus from complex
Ethiopian different source is good morphology.
Languages thing. -Not include real
[17] -The speed of the system is word error detection
well b/c Hashing search and correction.
algorithm provide fast
access.

Automatic Getnet -Metaphone algorithm -The proposed system is -Metaphone

Amharic A., for detection and edit effective in its error detection algorithim mostly
Spelling Error 2018 distance for correction and correction functionality. effective in less
Detection and of misspelled word morphological

32
Correction complexity language
using Hybrid -The proposed
Approach system inefficiency
[18] in speed
-The study not
considers real word
error.

Design and Fikru -Dictionary look up Developing affix (prefix, -Real word error
Implementatio T.,2018 and morphological infix, and suffix) rules: detection and
n of analyzer for error inflected words are generated correction are not
Morphology- detection by applying affix rules on the considered.
based Spell -Levenshtein edit root words. This minimizes
Checker [26] distance has used for the problems of requiring
correcting errors. extremely large spaces,
-Hunspell tool for searching time, and
morphological impracticability of listing all
analyzer correct words to develop a
dictionary.

Integrating the spell checker

into an open office word
processor

Designing Abdulje -Dictionary lookup -Using the hash function to -Dictionary based
Dictionary bar with hash function and minimize the response time method requires
Based K.,2019 n-gram approach to of the dictionary lookup more space for
Spelling detect the error and languge, that have
Checker for correct -The corpus collected from complex
Afaan difrent sources. morphology.
Oromoo [28] -Real word spelling

33
 error detetion and
correction not
include
Spell checker Berhun -Unsupervised -Developing affix (prefix and -Real word error
for Tigrigna H.,2020 machine learning suffix) rules: inflected words detection and
language approach are generated by applying correction are not
using rule -Hunspell for affix rules on the root words. considered.
based morphological This minimizes the problems -developed rule not
morphological analysis of requiring extremely large concider infix and
analyzer and -Edit distance for error spaces, searching time, and circumfix affixes
unsupervised correction impracticability of listing all
approach. [3] . correct words to develop a
dictionary.

-Integrating the spell checker

into an open office word
processor

Context- Aleka N-gram model -Corpus preparation from -The proposed

Dependent T.,2022 different source systems only focus
Spelling Error -This work is the first on homophone error.
Detection and attempt for Geez languge. -This work not
Correction attempts to solve
For Geez non-word error and
Language [4] real word error.
-Addtinally, this
work did not cover
Geez
abbreviations.Becua
se of Geez language
have many

34
 acceptable
abbreviations.

Table2. 1 Summary of literature related works

35
 CHAPTER THREE

3. Overview of Geez Language

This chapter explains the theoretical concept about Geez language. Especially, elaborates the
historical background of Geez language, its scripts, writing system as well as its linguistics
properties such as syllable structures, phonetics and morphology.

3.1.1. History of the Geez language

Ethiopian scholars [20] the word Geez has two meaning due to the variation of the letter. The
first meaning of the word Geez is the first and foremost language when we write with as
“ግዕዜ/giezi”. For instance Arabic language belong to the people who live in Arab, French
language belong to the people who live in France. However, Geez does not expect this
practice [39], the first and foremost language. Geez does not describe any tribe (ጎሳ), the
language of all [39].The second meaning is the language spoken by free people when we write
as “ግዕዜ/giezi”.

Around 3600BC, Semitic peoples who live in the South Asia migrate to South Arab and
Yemen then they migrate to Ethiopia, construct village at Aksum, develop, and spread their
language around the country. At that time the most dominant and well-known languages are
Saba (ሳባ) and Geez (ግዕዜ/gizi) [40]. Especially until Zagwe dynasty (13th centuries), Geez
becomes the only national language in Ethiopia, and used as governmental language. For this
evidence, many stone inscription and monument inscription have found in Aksum. After the
13th century, as the remains of Latin were making the metamorphoses into the romance
languages, spoken Geez also split into many closely related tongues, mainly Tigrigna in the
north and Amharic in the south [5].

3.1.2. Writing System of Geez language

Writing systems shows how to represent sets of symbols to represent the sounds of
speech and may also have symbols for such things as punctuation and numerals. In
Ethiopia, the only language that has its own alphabet (ፊዯሌ) is Geez. Other Semitics languages
like Amharic and Tigrigna adopted these alphabets from Geez language [38]. Before 5 century

36
BC the ancient Geez writing system from right to left up to Abune Selama like Arabic [41].
Currently, the Geez language writing system changes (i.e., left to right).

In Geez, 26 base alphabet/ፊዯሌ vertically listed with seven columns. Each of the columns are
labeled as ግዕዜ /geez/ (first order), ካዕብ /kaeb/ (second-order), ሣሌስ /salɨs/ (third-order), ራብዕ
/rabɨɁ/ (fourth-order), ኃምስ /hamɨs/ (fifth-order), ሳዴስ /sadɨs/ (sixth-order), and ሳብዕ /sabɨe/
(seventh-order) of alphabets [39].Look at the following example how to represent current
Geez alphabet.

ግዕዜ ካዕብ ሣሌስ ራብዕ ኃምስ ሳዴስ ሳብዕ

ሀ ሁ ሂ ሃ ሄ ህ ሆ
ሇ ለ ሉ ሊ ላ ሌ ል
Tatble3. 1 sample example to show modern representation of Geez alphabet
In addition to 26 base alphabet, there are four alphabets which have not the second and the
third order. So 26 *7+5*4=202. Totally 202 alphabets are used to type Geez text [38].This
alphabet is known as ፊዯሌ/fidel.

Geez language also has its own numerals for representing number.
፩/ አሐደ/1 ፰/ስምንቱ/8 ፷/ስዴሳ/ስሳ/60

፪/ክሌዔቱ/2 ፱/ተስዏቱ/9 ፸/ሰብዏ/70

፫/ሠሇስቱ /3 ፲/ዏሠርቱ/10 ፹/ሰማንያ/80

፬/አርባዕቱ /4 ፳/እሥራ/20 ፺/ተስዏ/90

፭/ኃምስቱ/5 ፴/ሠሊሳ/30 ፻/ምዕት/100

፮/ስዴስቱ /6 ፵/አርብዓ/40 ፼/እሌፍ/10000

፯/ሰብዏቱ/7 ፶/ሃምሳ/50 ፻፼/አእሊፋት/1000000

In Geez, has no symbol to represent zero [22]. Sometime Geez number combined with
alphabet. For example, ፲ወ፪ => 12 ፣ ፫ተ => 3.

37
 3.2. Geez word class

Geez word can categorize into eight: Noun, Adjective, Pronoun, Verb, Adverb, Preposition,
Conjunctions and Interjection [42].
Noun

Noun is a name that represents a person, animal places, feeling, things and idea. According to
[42] there are different types of nouns. Such as common noun, proper noun, collective noun,
abstract noun, pronoun, independent pronoun and suffix pronoun.

The objective of this study is not to explain types of nouns, rather how to derivate and inflect
noun and identify the plural marker of noun in Geez language. In Amharic most of the noun
have suffix -ኦች/ዎች(Plural marker) [10].But in Geez language there is no simple method in
order to change singular noun to plural noun [7]. In addition to adding suffix at the end and
prefix at the beginning, there is also internal inflection to change singular noun to plural noun.

Example ዴንግሌ to ዯናግሌ. Generally, there are two way of forming plural forms of noun.
These are - 1. Pattern replacement 2. Addition of an ending

Pattern replacement: for example ዯብር/ dabr --------አዴባር /adbar/ ፤ ሀገር/heger to አህጉር/ahigur ፤
ቤት/bet to አብያት/abyat.

Addition of an ending: for example አመት/amet ---------- አመታት/ametat. In Geez we use the
alphabet አ ፤አ……ት ፣ ው ፣ት ፣ሌ፤ሙ፤ይ to change a singular noun to Plural noun [23] [7] show
the following table.

Alphabet(letter) Singular noun Plural noun

ሌ ኪሩብ ኪሩቤሌ
ሙ አንተ አንትሙ
ሱራፊ ሱራፌሌ
ት ካህን ካህናት
አ ዯብር አዴባር
ው እዴ እዯው
ሕዜብ አሕዚብ
አ…ት ገብር አግብርት
አብ አበው
Tatble3. 2 Example of singular to plural noun

38
In addition to this plural formation of nouns can occur by changing the Fidel to ራብዕ ፣ ሳዴስ.
For example አንቀጽ to አናቅጽ ፣ ዯብተራ to ዯባትር ፣ መክሉት to መካሌይ ፣ መዴልት to መዲሌው. Some
nouns end with Fidel ሣሌስ ፣ ኃምስ ፣ ሳብዕ add ያት or ያን. For example ዯዌ to ዯዌያት ፣ አረጋዊ to
አረጋዊያን ፣ ቅዲሴ to ቅዲሴያት.

Adjectives

An Adjective is a word that describes nouns or pronoun. It used to identify the behavior, color,
shape character and situation of noun‟s or pronouns [7]. Example ቀይሕ/red, ዏቢይ/big. In Geez,
an adjective encounter in different state. Like in gender, case and number [42]. Most of The
Geez adjective can be inflected and derivate to plural number by prefixing “እሇ at the
beginning and suffixing “ን ፣ያን ፣ ት. Show the following table. Table 3.2

Prefixing Suffix Geez Singular Amharic Geez Inflected Amharic

እሇ መኑ ማን እሇመኑ
plural Word እነማን
ን Word
ፍንው Singular
የተሊከ ፍንዋን Plural
የተሊኩ
ያን ሰማያዊ ሰማያዊ ሰማያውያን ሰማያውያን
ት ቅዴስት የተመሰገነች ቅደሳት የተመሰገኑ
Tatble3. 3 Example of Geez Amharic adjective suffix and prefix to change in to plural
form
Verb

Verb is the word that used to indicate action or state of the sentence. According to [23] Geez
verb (ግስ) begin by five alphabets (fidel). These are ግዕዜ /Geez/ (first order), ራብዕ /rabi/
(fourth-order), ኃምስ /hamɨs/ (fifth-order), ሳዴስ /sadɨs/ (sixth-order), and ሳብዕ /sabɨ/ (seventh-
order) of alphabets and finish in two alphabets. These alphabets are ግዕዜ /Geez/ (first order),
and ኃምስ /hamɨs/ (fifth-order). The number of alphabet in Geez verb (ግስ) is between 2-7 [23].

According to the Ethiopian scholars, the types of verbs described as perfective, indicative,
subjunctive and imperative verbs are called ዓብይት አናቅፅ /abeyt anaqtsi/ .This verbs can close
the idea of a sentence independently without seeking a help of other verbs. In addition to these
verbs, the scholars identify ንዐሳን አናቅጽ /niusan anaqtsi/ which cannot close a sentence unless
other verbs are add to them.

39
 As shown in Table 3.4, the tense-mood identified by the Ethiopian scholars are the same as
that of those identified by the foreign scholars except that the Ethiopians identified extra
moods which are categorized as ንዐሳን አናቅጽ.

Tense-mood Identified by Ethiopian Scholars By foreign scholar

No Tense-mood Category
1 ቀዲማይ /ኃሊፊ አንቀጽ Perfective Perfective
ካሌአይ
(perfective)/ትንቢት አንቀጽ Indicative Imperfective
ዏቢይ አንቅጽ(abiy anqets)
ሣሌሣይ /ትእዚዜ (subjunctive)
(indicative) Subjunctive
የቅርብ ትእዚዜ አንቀጽ Imperative
ምክኒያታዊ አንቀጽ (Jussive)
(Imperative)
ቦዜ አንቀጽ (Gerundive)
አርእስት አንቀጽ (Infinitive)
ንዐሳን አንቅጽ /nius
ሳቢ ዗ር አንቀጽ (sabizər
ቅጽሌ አንቀጽ (ḱɨtsil anqetsi) anqets/
anqetsi)
Tatble3. 4 Tense-mood Identified by Ethiopian Scholars [38]
The inflection and derivation rule of verb follows the rule of their root verb. There are 8 type
of root (modal) verb [43].The difference of each root (modal) verb based on the following
criteria [22].

1. Number alphabet(የሆያት ብዚት)

2. Shape of alphabet(የሆያት አወቃቀር)
3. And way of pronouncing(አነባበብ)
Geez root verb
No Root verbs No of alphabet(የሆያት Shape of alphabet Pronouncing (አነባበብ)
(ግስ አርእስት) ብዚት) (የሆያት አወቃቀር)

1 ቀተሇ 3 ግዕዜ ግዕዜ ግዕዜ /ḱətələ/

2 ቀዯሰ 3 ግዕዜ ግዕዜ ግዕዜ /ḱəddəsə/

3 ገብረ 3 ግዕዜ ሳዴስ ግዕዜ /gəbɨrə/

40
4 አእመረ 4 ግዕዜ ሳዴስ ግዕዜ ግዕዜ /ɁəɁmərə/

5 ባረከ 3 ራብዕ ግዕዜ ግዕዜ /barəkə/

6 ሤመ 2 ኃምስ ግዕዜ /śemə/

7 ብህሇ 3 ሳዴስ ሳዴስ ግዕዜ /bɨhɨlə/

8 ቆመ 2 ሳብዕ ግእዜ /ḱomə/

Tatble3. 5 Geez root verb

A root verb is used as a base for another verb. Geez verbs follow one root verb inflection and
derivation rule. So to be clear if a verb similar to the root verb ቀተሇ/qetele, then the inflection
and derivation rule similar to ቀተሇ. Look at the following derivation and inflection verb ቀተሇ
(qetele).

ቀተሇ -- ገዯሇ (ቀዲማይ አንቀጽ )(perfective)

ይቀትሌ -- ይገዴሊሌ(ካሌኣይ አንቀጽ/ትንቢት ) (indicative) ዏቢይ አንቀጽ

ይቅትሌ -- ይግዯሌ/ ይገዴሌ(subjunctive)

ይቅትሌ---- ይገዴሌ ዗ንዴ(Jussive)

ቀቲሌ (ቀቲልት) -- መግዯሌ (infinitive)

ቀታሉ -- የገዯሇ …..Adjective

ቀታሉያን -- የገዯለ …adjective for plural male ንዐስ አንቀጽ

ቀታሉት -- የገዯሇች……..adjective for singular female

41
ቀታሌያት -- የገዯለ….adjective for plural femal ንዐስ አንቀጽ

ቅቱሌ -- የተገዯሇ…adjective

ቅትሇት -- አገዲዯሌ…… sabizer

ቀቲል…. ገዴል....... Gerundive

ቀታሉ -- ገዲይ……..Noun

ቀትሌ -- ውጊያ… noun(ጥሬ ዗ር) ስም ሲሆን

If some one knows how to inflect the verb ቀተሇ/ qetele, all verbs similar to qetele follow the
same rule.So the verb like ነበረ ፣ ወረዯ follows the ቀተሇ inflation and derivation rule.

Adverb (ተውሳከ ግስ)

Adverb is a word that modifies verbs for such categories as time, manner, place or direction
[43]. A word gives additional meaning to verbs. Like how, why, where, etc [44]. In Geez there
are eight main type of adverb [43] such as:

 Adverb of manner (ኩነታዊ) for example ፍጡነ -- በፍጥነት (quickly)

 Adverb of place (መካናዊ) for example ዜየ -- እዙህ (here)
 Adverb of time (ጊዚያዊ) for example ጌሠም -- ነገ (tomorrow)
 Adverb of frequency (የዯጊመ ጊዛ) for example ወትረ _ _ በማ዗ውተር (always)
 Adverb of certainty (ርግጠኝነትን የሚገላጽ) for example እሙነ --በርግጠኝነት (surely)
 Adverb of degree (የከፍታን ገሉጭ) for example ጽዴቀ --በአግባቡ (fairly)
 Interrogative (መጠይቃዊ) for example ማዕዛ -- መቼ (when)
 Relative (ተዚማጅ) for example በጊዛ

Conjunction (አያያዢ)

Conjunction is a word that links clauses or sentences or to coordinate words in the same
clause. ከመ፣ አምሳሇ፣ በ዗፣ በእንተ፣ ህየንተ፣ በይነ ፣አመ are some example of conjunction in Geez
[45].

42
Preposition

According to [44]Geez proposition can divide in to two.

1. Prepositions that fall onto nouns to indicate start and end, direction, comparison, time,
place .like እም (ከ)፣ ኀበ (መንገሇ፣ ወዯ)፣ ከመ (እንዯ)፣ በ (በቁሙ)፣ሇ (በቁሙ)፣ አሜሃ (ጊዛ) ፣ ውስተ
(ውስጥ)፣ ኦፍኣ (ውጪ).
2. The second Prepositions having (of „s) meaning. Like ዗ ፣ እንተ ፣ እሇ -- የ. Some preposition
used as conjunction [44]

Prepositions can be freestanding or prefixed to another POS [42]. For instance: በ /be/,
ሇ/le/, ወ/we/,዗/ze/ e.t.c are prefixed to another POS. ኀበ/hebe/፣መንገሇ/Mengele/፣ ከመ/keme/፤
አሜሃ/ameha/ ፣ ውስተ/wiste/፣ ኦፍኣ/afia/ e.t.c are free-standing preposition. The meaning of a
preposition can vary depending on the case of its object.

Interjection

Interjection is words that express strong feeling, emotion, or surprise [42]. They are often
capable of standing on their own.

Examples

 ሐዊሳ/hawisa/፣ሐዌሳ /hawisa/‟sign of joy.‟

 ሇይሌየ/leylye/፣ላሌየ/leylye/‟woe is me.‟
 ሶ/so/‟sign of request please!‟.

3.2.1. Affixation in Geez word

Affixation indicates the addition of affixes on the root/stem words to cause word variations.
Geez support all types of the affixes namely prefixes, suffixes, infixes and circumfixes.

43
 3.2.1.1. Prefixes

Prefixes are a morpheme that attached at the beginning of a lexical item or base-morpheme.
As described in [38] [46] Geez prefixes used to indicate Verbal-stem-marker, person-marker
prefixes, negative marker and basic out breeding phones (አስራው ፊዯሊት) [5].

Verbal-Stem-Marker Prefixes
This type of prefixes are attached to the front of the base-stem verb to form four more derived
stems namely causative, reflexive, reciprocal, and causative-reciprocal. For example the word
ቀተሇ/perfective, አቅተሇ/causative, ተቀትሇ/reflective, ተቃተሇ/reciprocal,አስተቃተሇ/causative-
reciprocal.
Look the following table.
No The Five Stem Types
Tense mode ገቢር አገብሮ ተገብሮ ተጋብሮ አስተጋብሮ(Causative
(Perfectve)) (Causative) (Reflexive) (Reciprocal) Reciprocal)

Perfective ቀተሇ አቅተሇ ተቀትሇ ተቃተሇ አስተቃተሇ

Indicative ይቀትሌ ያቀትሌ ይትቀተሌ ይትቃተሌ ያስተቃትሌ

Subjunctive ይቅትሌ ያቅትሌ ይትቀተሌ ይትቃተሌ ያስተቃትሌ
Imperative ቅትሌ አቅትሌ ተቀተሌ ተቃተሌ አስተቃትሌ
Jussive ይቅትሌ ያቅትሌ ይትቀተሌ ይትቃተሌ ያስተቃትሌ
Gerundive ቀቲል አቅቲል ተቀቲል ተቃቲል አስተቃቲል
Infinitive(a) ቀቲሌ አቅትል አስተቃትል ተቀትል ተቃትል
Infinitive(b) ቀቲልት አቅትልት አስተቃትልት ተቀትልት ተቃትልት
Tatble3. 6 five stem verb with tense mode [46]
Person Marker Prefixes
These are prefixes attached in front of the stem verb, to show the indicative, subjunctive and
jussive mood verbs to indicate the subject (doer) of the action of the verb. This prefixes are አ፤
ነ፤ተ፤ የ.Not only for this alphabet but also the ካዕብ /kaɁb/ (second-order), ራብዕ /rabɨɁ/ (fourth-
order) and ሳዴስ /sadsɨɁ/ (six-order) of the above alphabet is used as person marker prefixes.

44
For example, the verb ፈቀዯ (he liked) has its indicative, subjunctive and jussive forms as ይፈቅዴ
(he will like), ይፍቅዴ (he must like) and ይፍቅዴ (for him to like). The alphabet „ይ‟ the ሳዴስ
/sadsɨɁ/ (six-order) of የ. In both case „ይ‟ indicates the subject of the action. So the subject
is „He‟.

Negation makers (አለታ)

Negation marker prefix used to show the negativity of verbs and nouns. Those alphabets are
አሌ and ኢ. Only negative verbs can have the possibility of having two prefixes. The negation
marker of the verb always present at the beginning of the verb. Example the word ቦ /present
አሌቦ/absent and ተፈቀዯ/he is liked by) ኢተፈቀዯ/he is not liked by).

Basic out of breeding phones (አስራው ፊዯሊት)

Basic out breeding phones are prefixes that used for making the verbs to be (future tense and
past participle).This prefixes are /ይ፣ ት፣ ን፣ እ/

Basic out breeding phones (አሥራው ፊዯሊት(ይ፣ ት፣ ን፣ እ)

Past Future Past participle

ቀዯሰ (he sanctified) ይቄዴስ (he will sanctify) ይቀዴስ (he has sanctified)

ቀዯሰት(she sanctified) ትቄዴስ (she will sanctify) ትቀዴስ (she has sanctified)

ቀዯስነ (we sanctified) ንቄዴስ (we will sanctify) ንቀዴስ (we have blessed)

ቀዯስኩ (I sanctified) እቄዴስ (I will sanctify) እቀዴስ (I have sanctified)

Tatble3. 7 Basic out breeding phones

Not only this but also the ግዕዜ /geez/ (first order) and ራብዕ /rabɨɁ/(second order) of the above
alphabet(ይ፣ ት፣ ን፣ እ) is used as basic out breeding phones for some exceptional verbs.
Example1 ሐተተ to የሐትት/ Example2 አቀመ to ያቀውም. There for the alphabet “የ “and “ያ” the
ግዕዜ /geez/ (first order) and ራብዕ/rabɨɁ/(second order) of “ይ” respectively.

45
 3.2.1.2. Suffixes

Those morphemes added at the end of the verbs. In Geez this morpheme shows Gender
(Masculine or feminine), Number (Singular or Plural), nearness or farness of the verb. The
suffix of the verb in Geez language indicates either subjectivity or objectivity. Suffixing
morphemes at the end of a verb to show only the subject Gender, Number and nearness or
farness called subjective marker suffixes [7]. In addition to this suffixes at the end of the verb
to show the object‟s Number Gender nearness or farness known as Objective marker
suffixes.
Subject markers (ባሇቤት አመሌካቾች/ [ክሙ፣ ክን፣ ት፣ ነ፣ ከ፣ ኩ፣ ኪ/kimu፣ kini፣ ti፣ ne፣ ke፣ ku፣ kī] are
used for indicating the subject on verbs [41]. Example the word ቀተሌኩ/I kill / the suffixes
ኩ/ku indicates the subject I. For detail understand look the following table.

Geez
Pronoun Verb/ግሱ Suffix
አነ ቀተሌኩ ኩ
አንተ ቀተሌከ ከ
አንቲ ቀተሌኪ ኪ
ውእቱ ቀተሇ ግዕዜ ዴምጽ(ኧ)
ይእቲ ቀተሇት ት
ንሕነ ቀተሌነ ነ
አንትሙ ቀተሌክሙ ክሙ
አንትን ቀተሌክን ክን
ውእቶሙ ቀተለ ካዕብ ዴምጽ(ኡ)
ውእቶን ቀተሊ ራብዕ ዴምጽ(ኣ)

Tatble3. 8 Geez subjective suffix adapted from

Object markers suffixes ( ተሳቢ አመሌካቾች[ሁ፣ ሃ፣ ሙ፣ ሆሙ፣ ሆን፣ ን፣ ዋ፣ ዎ፣ ዎሙ፣ ዎን፣ ካ፣ ኮ፣
ቶ፣ ያ፣ ዮ፣ ዮሙ፣ ዮን/hu፣ ha፣ mu፣ homu፣ honi፣ ni፣ wa፣ wo፣ womu፣ woni፣ ka፣ ko፣ to፣ ya፣
yo፣ yomu፣ yoni] indicate the object of the verb [41] [5].

46
 Pronoun Geez verb Subjective suffix Objective suffix

አንተ ቀተሌኩከ -ኩ -ከ

አንቲ ቀተሌኩኪ -ኩ -ኪ

ውእቱ ቀተሌክዎ -ክ -ዎ

ይእቲ ቀተሌክዋ -ክ -ዋ

አንትሙ ቀተሌኩክሙ -ኩ -ክሙ

አንትን ቀተሌኩክን -ኩ -ክን

ውእቶሙ ቀተሌክዎሙ -ክ -ዎሙ

ውእቶን ቀተሌክዎን -ክ -ዎን

Tatble3. 9 Geez objective suffix

47
 CHAPTER FOUR

Design of Geez Spelling Checker

4. Overview

In the previous chapters, the researcher describes some of the related works of spelling error
detecting and correcting mechanisms for different languages. The main goal of this study is to
design and implement interactive spelling checker for Geez language. In this chapter, the
researcher has discussed the detail architecture of the system, the algorithm for Geez spelling
error detection and correction and the rule definition structure.

4.1. Morphological analyzer for Geez Language

Morphology is a branch of linguistic that deals the word formation and internal structure of the
word in the language [8].The smallest unit of word, which has meaning or grammatical
function in the language known as morpheme. So to be clear the word “አቡኪ” break in to two
morphemes “ኡኪ” is morpheme and “አብ” is another morpheme. Morpheme can divide in two
types. These are free morpheme and bound morpheme.
Free morphemes/lexical: this is a morpheme with complete meaning and they can stand
alone as an independent word in a sentence. This morpheme does not need to attach to other
words to give meaning. Example the word አብ፣እም this represents father, Mather respectively.
Bounded morphemes/affix: this is a type of morphemes cannot appear as a word by itself
and cannot stand alone, but when connected to another morpheme, gives meaning. For
example, the word አቡኪ/your father/ the morpheme አብ free morpheme were as the morpheme
ኡኪ is bound morpheme.
As mentioned above Geez have complex morphology.so creating a dictionary having all the
words of the language is very difficult task considering the large variations of affix
combination in the language. To solve this problem, we use morphological features and
dictionary of only the stem words and dictionary of affixes that are going to be attached to the
stem word based on the provided rule. For example the stem word ቀተሇ(he killed) ቀተለ፤ቀተሊ፣

48
ቀተሇ፣ቀተሇቶ፣ቀተሌኪ፤ቀተሌከ፤ቀተሌኩ፤ቀተሌኩከ e.t.c. From this example only the stem word ቀተሇ will
have kept in the dictionary and the other words inflected based on the morphological feature.
If the input word not found in the dictionary, the morphological analyzer split the word in to
probable stem word and probable affix.
So far there is no full developed morphological analyzer for Geez language. Due to this
reason, the researcher adopts the Hunspell dictionary and affix file format. Hunspell is an
open source spell checker. It has been designed especially for language with rich morphology
[24]. Hunspell requires two files to define the way of language being spell checked. These
files are dictionary file(*.dic) and affix(*.aff) file.

The dictionary compiled from [23] book. The dictionary contains verb, adjective, adverb, noun
and preposition. The researchers use this dictionary as base dictionary and add some country
name and spiritual common name like ኢትዮጵያ/Ethiopia/፣ሚካኤሌ/Mikael/፣ገብርኤሌ/Gebriael/፣
አብርሐም/Abrham/፤ ኢየሩሳላም/Eyerusalem. The word is listed one word per line followed by
forward slash (“/”), then 0 or more flags, which represents the word attributes, for example
affixes. Look the following example
4
SET UTF-8
ቀተሇ/itLkeiu PFX WB Y 4

አ዗዗/kitzei PFX WB 0 ወበ

PFX WB 0 እም
ኢየሩሳላም/WB
PFX WB 0 ዗
ፈተሇ/itLkeiu
PFX WB 0 ሇ

SFX it Y 2
SFX it 0 ት [^ተት]

SFX it 0 ቶ

Figure4. 1 (a and b) Morphological feature (affix and rule) and Dictionary

49
Figure 4.1a to shows how to build a dictionary. The first entry 4 indicates the number of
estimate word in dictionary. Geez word start at the second line flowed by slash and flags. The
slash ( “/” ) indicates the end of dictionary word.ቀተሇ/qetel/፣አ዗዗/azeze፣ኢየሩሳላም/iyerusaliem
and ፈተሇ/fetele are dictionary words. WB,it,k,ei,u and z are flags that points to the rule of
affix in morphological feature. For instance, WB points ዗፣ወበ፣ሇ፣እም. (See fig4.1b).

Geez language supports all affixes such as prefix, suffix, infix and circumfix. An affix file in
hunspell may contain many attributes. For example, SET used for setting the character
encoding of affixes and stem files. PFX and SFX defines prefix and suffix classes respectively
named with affix flags.

At shown in the above fig4.1b ወበ፤዗፣ሇ and እም attached at the beginning of the word (i.e.,
prefix). ት and ቶ attached at the end of the word(i.e., Suffix).as shown in fig 4.1a all
accepted words for the word “ኢየሩሳላም/WB” is ዗ኢየሩሳላም፤ወበኢየሩሳላም፤ሇኢየሩሳላም፤
እምኢየሩሳላም.

CIRCUMFIX X

PFX h Y 1

PFX h 0 ይ /X

SFX u Y 1

SFX u ተሇ ትሌ/uh [ተሇ]

Figure4. 2 Geez circumfix

In addition to prefix and suffix, Geez also support infix and circumfix. Look in fig 4.2 shows
if the word ends with ተሇ and ይ attached at the beginning of the word, and then change the
alphabet ተሇ to ትሌ. For example from the fig 4.1a the word ቀተሇ, ፈተሇ found in the dictionary,
then ይፈትሌ፣ ይቀትሌ are valid word.

50
 4.2. System Architecture for Geez Spelling Checker

The proposed system has three main components. These are preprocessing component, error
detection component and error correction component. Preprocessing component are
responsible for accept a block of Geez text from the user, then tokenize each word by using
word delimiter, remove punctuation marks, remove duplicate words and passes the candidate
words to the error detection component. The error detection component is responsible to
accept the candidate words from preprocessing component and check whether the candidate
word is valid or not based on the Geez dictionary and morphological feature. The error
correction component is responsible for providing suggestion for misspelled word by using
replacement rule and LED. Then after provide probable word suggestion, the user will select
correct list from the list.

The proposed system has two back end database. These are dictionary and morphological
feature. The dictionary database contains Geez language dictionary words, this includes noun,
verb, adjective and preposition as mentioned above in this chapter. Morphological feature
database also contains affix of Geez and defined morphological rule. Morphological rule used
to show how to attach each affix to the stem word to generate valid word (see fig4.1a and b).
General system architecture of Geez Language spelling checker are provided in the figure 4.3

51
 Block of Geez text

Word tokenization
Word preprocessor component

Not found Morphological

Morphologic
Dictionary analyzer
al feature
(Affix and
Found rule)

Error detection component

Morphological generation

Provide ranking suggestion

Error correction component

Correct text

Figure4. 3 General system architecture

52
 4.2.1. Preprocessing Component

The first component of this study is preprocessing component. According to modern

informational retrieval book [47] there are five types of document preprocessing. These are
tokenization, elimination of stop word, stemming, index term selection and thesauri. Form this
document preprocessing, only tokenization used for this study.

Tokenization

Tokenization is responsible to split a block of text in to individual tokens. These tokens may
be paragraphs, phrases, sentences, punctuation marks, digits or individual words [26]. The text
is broken based on some boundary delimiter such as white space and punctuation marks. The
Geez language words are separate or ended by blank space or punctuation. After the splitting
text is done, all punctuation marks are removed.

In this study tokenization, consider at word level. First read block of Geez texts and split in to
word, Then after, the candidate word pass to the dictionary to check either the candidate word
is found in the dictionary or not.
Input: Block of Geez Text

Output: List of Tokenize Geez words

1. Empty _list=[ ]
2. Read a character
3. For each character in a block of text up to end of document is reached
4. If a character is punctuation mark or space
5. Skip
6. Else append the read character to empty_list
7. End if
8. End for

Algorithm4. 1 for Tokenization process

The algorithm for tokenization to split a block of text starts working by reading block of text.
First, read a character (line 2-4) up to punctuation mark or space is reached. When the process

53
reached to the space or punctuation mark (line 4), the algorithm will skip (line 5) and store the
word in to empty_list (line 6).

4.2.2. Error Detection Component

The error detection components detect whether the individual word is valid or not. This
component contains stem word in dictionary database and affix with rule in another database.
To achieve this component the system performs two main activities. These are dictionary
lookup and apply morphological analyzer.

A. Dictionary lookup

Accept the candidate word from preprocessing component and check whether the word is
found in the dictionary or not. If the word founds in the dictionary, the candidate word will be
marked as valid and no need further process. However, the candidate word not found in the
dictionary, it needs further process on morphological feature. Then pass to morphological
analyzer. Look at the following algorithm to show dictionary lookup.

Input: tokenize words from preprocessing component

Output: correct or incorrect

1. Read tokenize words

2. For word in tokenize word
3. If the word present in the dictionary
4. Return correct
5. Else
6. Return incorrect

Algorithm4. 2 Dictionary lookup

The algorithm search the word from dictionary (line1-3).If the word found in the dictionary.
The word consider as correct word (line 4).But if the word is not found in the dictionary, it
consider as the word is incorrect(line 5-6).The word not found in dictionary, pass to
morphological analyzer for further processing.

54
 C. Morphological analyzer

The task of morphological analyzer in this study is to accept words from dictionary and
decompose these words in to stem word and affix morphemes. Then send the stem word to
dictionary and the affix morpheme to affix file to check the valid stem word and valid affix.
The researcher develop knowledge base morphological analyzer algorithm based on hunspell
dictionary and affix file format. Look at the following algorithm.

Input: word I_word from dictionary

Output: list of affix and stem words

1. Scan input word from right to left and left to right to look for valid suffix and
prefix
For each valid suffix in I_Word strip them and store result in a buffer
For each valid prefix in I_Word strip them and store result in a buffer
//pass list of affix and stems to the error detection module
Return stem and affix
//If there is no valid suffix and prefix
2. Scan input word from left to right and right to left and then look for a possible
stem
For each valid roots in I_Word strip the root and store result in a buffer
//pass list of possible stem to the error detection module
Return stem and affix //valid stem and invalid affix
Algorithm4. 3 Algorithm for morphological analyzer [3]
This algorithm makes use of rules stored in the knowledge base to strip a given word into its
stem words and affix. In this process, each individual word have scanned from right to left and
left to right in order to get affix and valid stem. The exact affix stripping is possible only for a
correctly spelled word. In the case of misspelled word, as there is an ambiguity as to whether
the error exists in the stem or affix, only probable affix stripping can be done. To illustrate
how the error detection and morphological analyzer works, consider the unknown word
መጽአት/she came/. The error detection module first check if መጽአት is found in the dictionary or

55
not. Since it is not found in the dictionary, the system cannot automatically say that it is
misspelled, for it may be inflected or derivate. Then the word passes to the morphological
analyzer. The morphological analysis algorithm will first scan from the right to left and left to
right to search for a valid suffix and prefix. Since /-ት/ is a valid suffix in Geez. Now the
System checks the unknown morpheme መጽአ for its presence in the dictionary. Hence, it is
found in the dictionary. Finally, to determine if the unknown word መጽአት is an acceptable
word, all the rules required to append suffix /-ት/will be checked in the affix file. Then, the
system will recognize መጽአት as a valid word, and no further process is needed. The same
processes have done for a misspelled word.

4.2.3. Error Correction Component

The main function for this component is to provide suggestion for misspelled words. In order
to achieve this component, the system perform morphological generation, suggestion and
ranking.

A. Morphological generation

Morphological generator is the process of rebuilding of decomposed word that come from
error detection module. It is the task of producing the appropriate inflected form of word
according to a given morphological features, first accepts a Pairs of morphemes obtained after
affix stripping in morphological analyzer. Then classify the error in one of the following
classes.

I. Correct stem and correct affix

First, the algorithm scan from left to right and right to left to get stem word and affix. Then the
stem word is searched in the dictionary and affix is searched in morphological features, if both
the stem word and affixes are found in the dictionary and the combination satisfies the rule,
then the word is considered as correct word and no further processing.

II. Incorrect stem and correct affix

If the stem word is incorrect, it is difficult to determine the exact affix that match with the
stem word. However, after strip the word, we get a list of possible pairs of stripped stem word

56
and valid affix. Then, the valid affix shows a specific category of the probable stem word.
After getting the possible valid affix, then possible stem word is checked for a word match. If
a match found, the possible suggestion are generated for the misspelled word by using the
replacement rule and LED. Rules in the knowledge base and LED techniques had used for
error correction.

III. Correct stem and incorrect affix

Similarly, in this combination again the word is scanned form left to right and right to left,
then the word stripped into a stem word and possible affix. If the stem is valid and the affix is
invalid, the valid stem shows a specific category of the possible affix. After that, as mentioned
above use replacement rule and LED to get probable correction.

IV. Incorrect stem and incorrect affix

For the case of misspelled word that is occur due to the miss combination of invalid stem and
invalid affix, the morphological analyzer tries to split the word into stem and affix, but the
stripped stem word and affix does not exist in the provided class and the word is recognized as
misspelled word. Some nearest word are displayed as suggestions for that misspelled word by
taking nearest stem word and affix that match to this stem word class. Look at the following
algorithm.

Input: morphemes from error detection (morphological features and dictionary) component

Output: list of words for suggestion

1. For each error word given from error detection

2. Identify stem word class and affix class of the error word
3. If match found
4. Append affix to stem word
5. 4.4 for morphological
Algorithm Generate list of words
generato 1
6. End if
Algorithm 4.4 for morphological generator
7. Return list of words
8. End for

Algorithm4. 4 for morphological generator

57
 D. Provide Ranking and Suggestion

After the morphological generator was done, the next step is providing suggestion for
misspelled word and ranking for the suggestions. Replacement rule and LED are used for
suggestion and ranking. LED is the most efficient and known algorithm for many language to
rank suggestion [3] [26]. For providing suggestion, we have uses maximum suggestion 5 as
the threshold value.

58
 CHAPTER FIVE

5. Experiment

5.1. Overview

In this chapter, we have discussed how to collect the test data, detailed implementation of
spelling checker and the prototype of the GLSC system. In addition to this, measure the
performance of the GLSC system and discussed the challenge and limitation to implement
GLSC.

5.2. Testing Data Preparation

Geez language does not have electronic error free text for spelling checker and any other NLP
applications. Even though, some of EOTC books available in online and in android mobile
applications. Such as EOTC bible, mezigebe haymanot, Wudasie Maryam, Teamre
Maryam,Sinksar , But this electronic source have expose to error. Especially homophone
alphabet interchangeably error, almost all [4].Due to this fact, we do not use electronic source
to build dictionary. Because of preparing error-free Geez dictionary helps to maximizing the
performance of spelling checker system. As mentioned in chapter 4, build lexicon of the
language from [23] book and identify the morphological feature of each word by linguistics
experts.

To evaluate the system, we have taken 3000 words directly from EOTC bible from online.
After removing repeated word, we got 1318 unique words. The linguistic expert identify the
misspelled words manually, got 185 misspell words. From these misspelled words, the
majority errors are encounter on (substitution, deletion and insertion). Look at the following
table.

Total Number of Cause of error Total number of error Percentage in (%)

Non word error

Substitution 70 37.8

59
 185 Deletion 32 17.2

Insertion 26 14.05

Other type 57 30
error(transposition
,split word , combined
word and mixed)

Table5. 1 Result of cause of pattern error

5.3. Prototype of the System

The prototype of the system developed to show how to implement the GLSC system
architecture that presented in Chapter 4. The system developed by using python programming
language. The reason for selecting python as development environment is, python has known
for its efficiency in the implementation of data processing applications and it can be integrate
with hunspell dictionary format. The prototype uses two reference files stored in hunspell
dictionary format (i.e., dictionary of the language and morphological feature (affix and rule)).

The prototype worked by accepting input word from the user. After accepting the word from
the user, tokenize the sentence in word level and remove punctuation marks by preprocessing
module. Then detect the misspelled word. If the word is misspelled, highlight by red color.
Look at the following figure to show the prototype of the system for error detection .

60
Figure5. 1 sample example of error detection
Fig 5.1 shows, how Geez language spelling checker system detects the misspelled words. The
user writes Geez text in the provided space and click check button. At that time the system
identify the misspelled word and if misspelled is found shade in red color.

61
Figure5. 2 sample prototype to rank suggestion
Fig 5.2 shows, how to suggest misspelled word by the system. As mentioned in chapter 2 the
first phase of spelling checker is identify misspelled words and the second phase of spelling
checker is providing suggestion for misspelled word.

After writing a block of text in provided space, click check button. If the word is misspelled
words, highlight in red color. Then in order to get suggestion for misspelled word, Double-
click a highlighted word to select it. Then right click on the selected word to pop the
suggestion .Then choose one candidate correction word.

62
 5.4. Performance Evaluation

To see the functionality and performance of the system, we conduct an experiment on GLSC
system. Functionality indicates what the GLSC should do (i.e., error detection and error
correction). Performance shows to what extent the functionality of GLSC system achieve the
objective. In this work, we conduct two experiments. The first experiment conducted on error
detection component and the second experiment conducted on error correction component.

5.4.1. Performance of Spelling Error Detection

The experiment has conducted to measure the performance of the GLSC.As we clearly
explain in chapter 4 sub section 4.3.2, the Geez language spelling error detected by error
detection component (dictionary lookup and morphological analyzer). So the first experiment
conducted on spelling error detection to see the performance of error detection component.

As mentioned above in this chapter sub section 5.2, we have collected 1318 unique words
from bible. From these words, linguistics expert identifies 185 words as misspelled words. In
order to evaluate the proposed system, we use the following parameters.

 True positive (TP) indicates valid words recognized by the spelling checker as
valid
 True Negative (TN) indicates invalid words are detected by the spelling checker
as invalid.
 False Positive (FP) indicates invalid words not detected by the spelling checker as
invalid.
 False negative (FN) indicates valid words not recognized by the spelling checker
as valid

According to this parameter, we can calculate the lexical recall, error recall, lexical precision
and error precision. The total number of correct and incorrect word identified by linguistics
expert is 1133 and 185 respectively. But the GLSC system detected only 155 words from 185
misspelled identify by linguistics expert, 94 correct words detected by GLSC as misspelled
and 30 misspelled words have not detected by GLSC. Look at the following table

63
 Result
Lexical Error Lexical Error
recall Recall Precision precision Accuracy

True
positive(TP) 1069 =91.9% =83.7% =97.2% =62.2% 90.07%
True
negative(TN) 155
False
negative(FN) 94
False
positive(FP) 30
Table5. 2 Error detection result
As shown in table 5.2 to measure the overall performance (accuracy) of error detection
component, uses the following formula

P= (TP+TN)/(TP+TN+FN+FP )=(1069+155)/(1069+155+94+30)=90.07%

5.4.2. Performance of Spelling Error Correction

The Spelling error corrector component of the proposed system is responsible to provide an
alternative word(s) for misspelled word(s) as we mention in chapter four sub sections 4.3.3.
In this section, we measure how well the spelling checker can suggest corrections for all true
negatives (i.e. incorrect words flagged by the GLSC). To measure the spelling error suggestion
we use suggestion adequacy (SA).SA refers to a spelling checker‟s ability to display the end-
user with relevant suggestions for all true negatives.

Although SA often not used to measure spelling checker evaluations, some researchers have
already attempted to express the SA of spelling checkers as a metric [48].

SA= ……… (5.1)

64
Look at the following sample example to show how to calculate SA.

Only correct Suggestion

Invalid word flagged by Total No of suggestion flagged flagged by linguistic
GLSC by GLSC expert
በንተ 4 3
ሇእግዙኢነ 1 1
዗ተጸውአ 1 1
ወተፊሌጠ 2 2
በሉሰን 2 2
ሇትምተ 5 3
መሊኪተ 3 3
ነጉረ 4 4
በአፈነቢያቲሁ 2 1
Table5. 3 sample example to calculate SA
SA determined by summing all the scores (where S is a score for a suggestion), and dividing it
by the total number (N) of true negatives (Tn):

∑
SA= ……………………………………………….(5. )

Table 5.3 shows, the GLSC flagged 155 true negative words. In order to correct them, the
GLSC suggest an alternative word. As we can see from the results, it is impossible to say that
all the words suggested by GLSC are correct words.So we used linguistic expert to identify
which word is correct and which one is in correct. For instance, from table 5.3, the first row
the word “በንተ” is misspelled word. GLSC suggest 4 words for this word. From these
suggestion, only 3 are (በእንተ/beente፤በነት/benete፤በቤተ/bebiete) correct suggestions and one is
(ብንተ/binte) incorrect word .There for We use SA to measure the extent to which spelling
correction of performance is achieved and got correct suggestion 87.5%.

5.5. Result and Discussion

The experimental results show value of lexical recall 91.9%, error recall 83.7%, lexical
precision 97.2%, error precision 62.2% and correct suggestions provided by GLSC 87.5%.
The overall performance of the system is 90.05%. This shows that the GLSC system perform

65
well in flagging words as valid/invalid and suggestion. However, needs some improvement to
increase the performance of both error detection and suggestion generation. We have
identified some of the reason that reduces the performance of the GLSC system. The first
reason that out of vocabulary problem. As researcher knowledge, there is no readymade
dictionary for Geez language so far. Due to this reason, our dictionary not contains all Geez
words. Additionally, the testing data randomly have taken from bible. In this case, some of
personal name and place name may include in testing data. For example, the word ቂሳር/qisar
and ቄሬኔዎስ/qierenewos are spiritually recognizable names. However, The GLSC detect as an
error. For this reason, some of recognizable personal names or place names flagged as invalid.
Therefore, increase the size of the dictionary will increase the performance of the GLSC
system.

The second reasons are affix rule mismatching. Geez verb inflected and derivate based on
their root verb as mentioned in chapter 2 section 2.1.3.But some verb are follow exceptional
rules. For example the root verb ቀተሇ(he killed), to indicates future time add prefix ይ at the
first and the last two character change in to ሳዴስ(sadis)(six order). There for it will be
ይቀትሌ(he will kill).So the Geez word found under root verb ቀተሇ follow the same rule. For
example, ነበረ/nebere፤ነገረ/negere፤ ፈተሇ/fetele follows this rule. The word ሐተተ(he examine) also
found in the root verb ቀተሇ. However, it follows exceptional rule to inflect and derivate. To
indicate future time add the prefix የ instead of ይ. There for it will be የሐትት (he will
examine).if we add ይ at the prefix of the word ሐተተ we will get invalid (ይሐትት).There for
developed well organized affix rule increase the performance of GLSC.

5.6. Challenge and Limitation

The first challenge we face is Geez have complex morphology and some word follow
exceptional rule. For example to inflect the verb ቀተሇ/qetele/ to show the subject „I‟ as the
doer of the action, we change the last alphabet to ሳዴስ /sadis and add prefix ኩ/ku to get
ቀተሌኩ(I killed).Most words follow the same rule of changing the order of their last alphabet to
ሳዴስ /sadis and add prefix ኩ/ku. However, there is an exception for verbs ending in ቀ, ገ, ከ.
For example applying the same derivation to the verbs „አጥመቀ‟/atimeqe/, „ኀዯገ/hedege‟

66
„ነሰከ/neseke‟ results in „አጥመቅኩ‟ „ኀዯግኩ‟ „ነስክኩ‟ respectively. These are invalid words. These
and other exceptions in Geez language make the system flagged invalid word as correct and
valid word as incorrect.

The second challenge we faced is, lack of error free corpus to build lexicon of the language.
Manually reviewing the collected Geez data from book is very tiring and time-consuming task.

67
 CHAPTER SIX

6. Conclusion and Recommendation

6.1 Conclusion

Spelling checker is an NLP application, used to detect misspelled word(s) in a sentence or a

paragraph and provide an alternative for misspelled word(s). The spelling checker is language
dependent. (i.e., a spelling checker for one language cannot work for another language). There
are different techniques of for spelling error detection and correction. For error detection,
dictionary lookup and N-gram analysis are common techniques. Rule-based, N-gram,
similarity key, edit distance, neural network, probabilistic and noisy channel are some of the
techniques used to provide suggestions.

In this research, the researcher applied morphology-based approach for the implementation of
Geez spelling checker, particularly to detect and correct non-word error. The error was
detected using dictionary lookup and morphological analyzer and corrected by using
replacement rule and LED. We adopt the Hunspell dictionary and affix file format to design a
lexicon (i.e. the knowledge base component) and hashing algorithm for searching. Hunspell is
an open source spelling checker tool. It has designed especially for languages that have
complex morphology. Hunspell needs two files. These files are dictionary and morphological
feature (affix) with defined rule. The dictionary saved as Geez.dic and morphological feature
(affix) saved as Geez.aff. A dictionary built from 6115 words, and 955 rules have defined
from affix file.

The developed GLSC was implemented using python programming language. A user can
provide a block of text by typing in space provided. Finally, testing has carried out in order to
prove the spelling checker system using sample data of Geez language words collected from
bible. To measure the performance of the system, we have used information retrieval
evaluation metrics of recall and precision for error detection and suggestion adequacy for
suggestion. For testing, 1318 unique words randomly have taken from bible to conduct
experiment.

68
From the conducted experiment, we got the result of lexical recall 91.9%, error recall 83.7%,
lexical precision 97.2%, error precision 62.2% and correct suggestions provided by GLSC
87.5%. The overall performance of the system is 90.05%.

Generally, we concluded that the performance of the Geez spelling checker is well with some
limitations. Increase the size of the dictionary and develop well organized rule will increase
the overall performance of the GLSC system.

Contribution of this study

This study has the following contribution: -

 This experimental study shows that a morphology-based approach for spelling error
detection and correction for the Geez Language with an experimental performance
result of 90.05% overall error detection and 87.5% error suggestion is a good candidate
for development of spelling checker for the language.
 This study shows the lexicon and affix of Geez language for error detection and
correction
 We developed a system that detects error words for Geez language
 We developed a system that provides a list of words a suggestion for correcting
misspelled words for Geez language

6.2. Recommendation

We would like to recommend the following key points for further research that helps to
improve the performance and include some functionality.

 In this study, we tried to solve Non-word error in Geez language, one can extend it to
work on spelling checker that can detect and correct real-word errors by considering
Semitics and grammar of the language to make spelling checker system more
interactive.
 One of the challenges we faced in this thesis was finding standard corpus and free from
homophone alphabet error of Geez language. Some of the corpus that found in online

69
 faced for homophone error. For instance instead of ሐመር write ሀመር or ኀመር is an error.
Therefore, future study may need to give attention to the development of standard
Geez language corpus that can motivate the researchers to work on any NLP
application not only on spelling checker application and minimize time to collect
corpus in order to evaluate their system performance by the researchers.
 We develop morphology-based spelling checker for Geez language in limited
dictionary and test in limited data. In order to increase the performance of the system,
one can increase the size of dictionary and develop well organize rule
 For future research will by including part of speech tagging to overcome the real word
error detection and correction problems will increase the performance of the system.
 Integrating the full GLSC with word processor and search engine is a future research.

70
Reference

[1] M. Abeje, "DESIGNING AUTOMATIC PRONUNCIATION DETECTION FOR GEEZ LANGUAGE," Bahir
Dar University(Msc Thesis), Bahir Dar, 2018.

[2] f. Gizachewu, "Developing part of speech tagger for guragginga language," Bahir Dar
University(Msc thesis), Bahirdar , 2017.

[3] B. Hadis, "SPELL CHECKER FOR TIGRIGNA LANGUAGE USING RULE BASED MORPHOLOGICAL
ANALYZER AND UNSUPERVISED APPROACH," Bahir Dar university (thesis Msc), Bahir Dar, 2019.

[4] A. Tesfie, "Context-Dependent Spelling Error Detection and Correction for Geez language,"
Gondar University, Gondar University Department of Computer Scienc Msc thesis, 2022.

[5] W. Manaye, "DESIGNING GEEZ NEXT WORD PREDICTION MODEL USING STATISTICAL
APPROACH," Bahir Dar University (MSc thesis), Bahir Dar, 2020.

[6] G. H., "Enhanced Morphological Analayzer for geez verb," international Journal of Science &
Technoledge, china, 2020.

[7] T. Kass, "Morpheme-Based Bi-directional Ge’ez -Amharic Machine Translation," Addis Ababa,,
2018.

[8] A. Belay, "DESIGNING A STEMMER FOR GE’EZ TEXT USING RULE BASED APPROACH," Ababa
University(Msc Thesis), Addis ababa, 2010.

[9] G. Olani, "Design And Implementation Of Morphology Based Spell Checker," International Journal
of Scientific & Technology Research, vol. 3, no. 12, 2014.

[10] M. Tilahun, "Automatic Speeling Checker In amharic languge," Bahir Dar University(Msc Thesis),
Bahir Dar, 2019.

[11] Rajashekara Murthy S,Vadiraj Madi,Ramakanth Kumar P, "A NON-WORD KANNADA SPELL
CHECKER USING MORPHOLOGICAL ANALYZER AND DICTIONARY LOOKUP method," International
Journal of Engineering Sciences & Emerging Technologies, vol. 2, no. 2, 2012.

[12] AmanjotKaur, Dr. ParamjeetSingh,Dr. ShavetaRani, "Spellchecking and Error Correcting System
for text paragraphs written in Punjabi Language using Hybrid approach," International Journal of
Advanced Research in Science, Engineering and Technology, vol. 2, no. 11, 2015.

71
[13] Baljeet kaur ,Harsharndeep Singh, "Design and Implementation of HINSPELL -Hindi Spell Checker
using hybrid approch," International Journal of scientific research and management (IJSRM) , vol.
3, no. 2, 2015.

[14] M. Shimelis, "Amharic spelling detection and correction system morphology based approch,"
Addis Ababa University(MSc thesis), Addis ababa, 2020.

[15] B. H, "Spell Checker for tigringa language using rule based using morphological analayzer and
unsupervised approch," Bahir Dar University(MSc thesis) , Bahir Dar, 2019.

[16] W. Barud, "Multilingual Spelling Checker for Selected Ethiopian Languages," International Journal
of Advanced Science and Technology, vol. 29, 2020.

[17] W. Barud, "Dictionary Based Spelling Corrector System the Case of Six Ethiopian Languages,"
International Journal of Aquatic Science, vol. 12, no. 02, 2021.

[18] G. Asefa, "AUTOMATIC AMHARIC SPELLING ERROR DETECTION AND CORRECTION USING HYBRID
APPROACH," Journal of Emerging Technologies and Innovative Research, vol. 5, no. 6, 2018.

[19] S. TEGEN, "OPTICAL CHARACTER RECOGNITION FOR GE’EZ SCRIPTS WRITTEN ON THE VELLUM,"
UNIVERSITY OF GONDAR, UNIVERSITY OF GONDAR Msc Thesis, 2017.

[20] መ. አ. ተክላ, መጽሐፈ ታሪክ ወግስ, አዱስ አበባ, 1997.

[21] Veena Dixit, Satish Dethe, Rushikesh K. Joshi, "Design and Implementation of a Morphology-
based Spellchecker for Marathi,an Indian Language," Mumbai, Indian Institute of Technology
Bombay.

[22] H. Haile, "ግእዜ መማሪያ," dirzon.com, [Online]. Available: www.dirzon.com/Doc/Details/Habtamu

Haile%3Aየግእዜ መማሪያ.pdf.

[23] ሉ. ሉ. ያ. ሽፈራው, መጽሐፈ ግስ ወሰዋሰው መርኆ መጻሕፍት.

[24] January 2022. [Online]. Available: http://hunspell.github.io/.

[25] S. M. El Atawy,A. Abd ElGhany, "Automatic Spelling Correction based on n-Gram Model," vol.
182, August 2018.

[26] F. T, "Morphology Based Spell Checker for Kafi Noonoo Language," Addis Ababa University (MSc
thesis), Addis Ababa , 2018.

[27] K. KUKICH, "Techniques for Automatically Correcting Words in Text," ACM Computing Surveys,

72
 vol. 24, 1992.

[28] A. Kedir, "Designing Dictionary Based Spelling Checker for Afaan Oromo," IMMA, 2019.

[29] T. KUMERA, "CONTEXT BASED SPELLCHECKER FOR AFAN OROMO WRITING," JIMMA UNIVERSITY
SCHOOL OF GRADUATE STUDIES DEPARTMENT OF INFORMATION SCIENCE Msc thesis , JIMMA ,
2018.

[30] H. L. Liang, "SPELL CHECKERS AND CORRECTORS:A UNIFIED TREATMENT," UNIVERSITY OF

PRETORIA,, SOUTH AFRICA , 2008.

[31] Er. Sumreet Kaur Randhawa Er.Charanjiv Singh Saroa, "STUDY OF SPELL CHECKING TECHNIQUES
AND AVAILABLE SPELL CHECKERS IN REGIONAL LANGUAGES: A SURVEY," International Journal
For Technological Research In Engineering, vol. 2, no. 3, 2014.

[32] Neha Gupta Pratistha Mathur, "Spell Checking Techniques in NLP: A Survey," International
Journal of Advanced Research in, vol. 2, no. 12, 2012.

[33] Ritika Mishra,Navjot Kaur, "A Survey of Spelling Error Detection and Correction Techniques,"
International Journal of Computer Trends and Technology, vol. 4, no. 3, 2013.

[34] "www.geeksforgeeks.org/hamming-distance-two-strings/," geeksforgeeks. [Online]. [Accessed

January 2022].

[35] M. S. Nonghuloo, "Spell Checker for Khasi Language," International Journal of Software
Engineering, vol. 8, 2017.

[36] Gerhard B. van Huyssteen*, Roald Eiselen & Martin Puttkammer, "Evaluating Evaluation Metrics
for Spelling Checker Evaluations," North-West University, Potchefstroom, .

[37] MOHAMMED ATTIA1,2, PAVEL PECINA3,YOUNES SAMIH4, KHALED SHAALAN2,and J O S E F V A N

G E N A B I T H1, "Arabic spelling error detection and correction," 2015.

[38] Y. A, "Morphological Analysis of Ge’ez Verbs Using Memory Based Learning," Addis Ababa
University(MSc thesis), addis ababa, 2014.

[39] ድ. ሇ. ብርሃኑ, የግእዜ መማሪያ, ዋሽንግተን ዱሲ, 2006.

[40] ኪ. ወ. ክፍላ, መጽሐፈ ሰዋስው ወግስ, አዱስ አበባ , 1909.

[41] T. ABEBE, "DESIGNING AUTOMATIC SPEECH RECOGNITION FOR GE’EZ LANGUAGE," BAHIR DAR,
ETHIOPIA , 2018.

73
[42] M. Kebede, "Development of Part of Speech Tagger for Ge’ez Languag," Addis Ababa, Ethiopia,
2017.

[43] መ. ዯ. ቀሇብ, ትንሣኤ ግእዜ, አዱስ አበባ: በኢትዮጵያ ኦርቶድክስ ተዋህድ ቤተክርስቲያን ማኅበረ ቅደሳን , 1996.

[44] B. Abe, "Geez to Amharic Machine Translation," Addis Ababa, 2018.

[45] W. Wolde, "INFORMATION EXTRACTION MODEL FROM GE’EZ TEXT," BAHIR DAR, 2021.

[46] D. Berihu, "DESIGN AND IMPLEMENTATION OF AUTOMATIC MORPHOLOGICAL ANALYZER FOR

GE’EZ VERBS," ADDIS ABABA UNIVERSITY SCHOOL OF GRADUATE STUDIES DEPARTMENT OF
COMPUTER SCIENCE, 2010.

[47] Ricardo Baeza Yates Berthier Ribeiro-Neto, Modern information retrival, Nework: ACM press.

[48] Menno M van Zaanen and d Gerhard B van Huyssteen, "Various Uses of a Spelling Checker
Project: Practical Experiences, Teaching learning," southern African Linguistics and Applied
Language Studies, 2009.

[49] ዲ. ዯ. ቀሇብ, ግእዜ ሕያው ሌሳን በቀሊሌ ዗ዳ, አዱስ አበባ, 1993 .

[50] N. Ytayal, "Context based spell checker for amharic," Jimma Univsrsity(Msc thesis), Jimma, 2016.

[51] Andargachew Mekonnen Gezmu, Andreas Nürnberger, Binyam Ephrem Seyoum, "Portable
Spelling Corrector for a Less-Resourced Language: Amharic," in Otto-von-Guericke-Universität
Magdeburg, Addis Ababa University, Otto-von-Guericke-Universität Magdeburg, Addis Ababa
University, 2018.

[52] Atakilti Brhanu Kiros1,Petros Ukbagergis Aray2, "Tigrigna language spellchecker and correction
system for mobile phone device," vol. 11, 2021.

[53] Itziar Aduriz, Iñaki Alegria, Xabier Artola, Nerea Ezeiza, Kepa Sarasola and Miriam Urkia, "A
spelling corrector for Basque based on morphology," Literary and lingustics computing .

[54] M. Asadullah, "FINITE STATE RECOGNIZER AND STRING SIMILARITY BASED SPELLING CHECKER
FOR BANGLA," BRAC University Department of Computer Science and Engineering Msc thesis,
Dhaka, 2007.

[55] Bakkali Hamza,Gueddah Hicham ,Yousfi Abdellah,Belkasmi Mostaf, "For an Independent Spell-
Checking System from the Arabic Language Vocabulary," International Journal of Advanced
Computer Science and Applications,, vol. 5, 2014.

74
 Appendix A: Geez Alphabet and Number

Adapted from [5] with little modification

75
 Appendix B: Sample Affix file Rule in Hun spell
#prefix

PFX ei Y 5

PFX ei 0 ኢ

PFX ei 0 ዗ኢ

PFX ei 0 ወ዗ኢ

PFX ei 0 በ዗ኢ

PFX ei 0 ወኢ

PFX Ys Y 8

PFX Ys 0 ይ

PFX Ys 0 ት

PFX Ys 0 እ

PFX Ys 0 ዗ይ

PFX Ys 0 ኢይት

PFX Ys 0 ኢተ

PFX Ys 0 ኢይ

PFX Ys 0 ተ

PFX bz Y 4

PFX bz 0 በ዗

PFX bz 0 ዗በ

PFX bz 0 ሇ዗

76
PFX bz 0 ዗ሇ

PFX WB Y 7

PFX WB 0 ወበ

PFX WB 0 እም

PFX WB 0 ወ

PFX WB 0 ሇ

PFX WB 0 ዗

PFX WB 0 በ

PFX WB 0 ወኢ

#suffix

SFX pi Y 8

SFX pi 0 ነ

SFX pi 0 ኪ

SFX pi 0 ክሙ

SFX pi 0 ከ

SFX pi 0 የ

SFX pi 0 ት

SFX pi 0 ሁ

SFX pi 0 ሆሙ

SFX Li Y 10

SFX Li ሇ ላሁ ሇ
77
SFX Li ሇ ላሃ ሇ

SFX Li ሇ ላክን ሇ

SFX Li ሇ ላሆሙ ሇ

SFX Li ሇ ላሆን ሇ

SFX Li ሇ ላክሙ ሇ

SFX Li ሇ ላኪ ሇ

SFX Li ሇ ላከ ሇ

SFX Li ሇ ላየ ሇ

SFX Li ሇ ላነ ሇ

#infix

PFX IN Y 11

PFX IN ከዏ ተክዕ ከዏ

PFX IN ም እም ም

PFX IN ከሠ ክሡ ከሠ

PFX IN በጽ ይብጽ በጽ

PFX IN ዯመ ዴሙ ዯመ

PFX IN በዜ ብዘ በዜ

PFX IN ፈሇ አፍሇ ፈሇ

PFX IN ተሰ አሰ ተሰ

PFX IN ነበ አንበ ነበ

PFX IN አንቀ አናቅ አንቀ

78
PFX IN ጽን አጽና ጽን

PFX ts Y 6

PFX ts ጽሕ ጻሕ ጽሕ

PFX ts ጸን አጽ ጸን

PFX ts ጸን ጽኑ ጸን

PFX ts ተን አን ተን

PFX ts ነበ ነቢ ነበ

PFX ts ዴን ዯና ዯን

79
 Appendix C: Sample Corpus

እምጳውልስ ገብሩ ሇእግዙኢነ ኢየሱስ ክርስቶስ ወሐዋርያ ዗ተጸውአ ወተፈሌጠ ሇትምህርተ ወንጌለ እግዙአብሔር
዗አቅዯመ ነጉረ በአፈ ነቢያቲሁ ወመጻሕፍቲሁ ቅደሳት በእንተ ወሌደ ዗ተወሌዯ ወመጽአ እም዗ርዏ ዲዊት በሥጋ ሰብእ
ግብረ ወአርአየ ከመ ወሌዯ እግዙአብሔር ውእቱ በኀይለ ወበመንፈሱ ቅደስ ከመ ተንሥአ እሙታን ኢየሱስ ክርስቶስ
እግዙእነ ዗ቦቱ ረከብነ ጸጋ ሐዋርያተ ከመ ናስምዖሙ ሇአሕዚብ በስሙ ከመ አንትሙኒ ይእዛ ኮንክሙ ጽዉዓነ በኢየሱስ
ክርስቶስ።

በእንተ ጽዉዓን ወበእንተ ሃይማኖሙ

ሇኵልሙ እሇ ሀሇዉ ብሔረ ሮሜ ፍቁራኒሁ ሇእግዙአብሔር ወኅሩያኒሁ ወቅደሳኒሁ ሰሊም ሇክሙ ወጸጋ እግዙአብሔር
አቡነ ወእግዙእነ ኢየሱስ ክርስቶስ አቀዴም አእኵቶቶ ሇእግዙአብሔር በኢየሱስ ክርስቶስ በእንተ ኵሌክሙ እስመ ተሰምዏት
ሃይማኖትክሙ ውስተ ኵለ ዓሇም ወእግዙአብሔር ሰማዕትየ ዗ኪያሁ አመሌክ በኵሌ መንፈስየ ወበትምህርተ ወሌደ ከመ
እዛከረክሙ በጸልትየ ዗ሌፈ ወእስእሌ ከመ ይሠርሐኒ እግዙአብሔር በፈቃደ እምጻእ ኀቤክሙ እስመ እፍቅዴ ወከመ
ትርከቡ ጸጋሁ ሇመንፈስ ቅደስ በእንተዜ ከመ ይትፈሣሕ ሌብክሙ እስመ ኀበርክሙ አሚነ ምስላየ።

በእንተ ጻሕቀ መምህራን ወዕሴቶሙ ወእፈቅዴ ባሕቱ ታእምሩ አኀዊነ ከመ ዗ሌፈ እፈቅዴ እምጻእ ኀቤክሙ ወይሰአነኒ
እስከ ይእዛ ወሇእመኒ ቦ ከመ እርከብ ዕስትየ በሊዕላክሙ ከመ በሊዕሇ አሕዚብኒ ወበሊዕሇ አረሚኒ ወበሊዕሇ ሐቃሌኒ
ወሇጠቢባንኒ ወሇአብዲንኒ እስመ ሇኵለ ሰብእ እምሂር ወዓዱ ፈዴፋዯ እጽሕቅ ሇክሙ ሇእሇ ብሔረ ሮሜ እምሀርክሙ
እስመ ኢየኀፍር ምህሮ ወንጌለ እስመ ኀይለ ሇእግዙአብሔር ውእቱ ሇኵልሙ እሇ የአምኑ ቦቱ ሇአይሁዲዊ ቀዲማዊ
ወሇአረማዊኒ ዯኃራዊ ወቦቱ ያስተርኢ ጽዴቀ እግዙአብሔር ወርትዐ እስመ ያጸዴቆሙ ሇእሇ የአምኑ ቦቱ በአሚን እስመ
ከማሁ ይብሌ መጽሐፍ በአሚን የሐዩ።

በእንተ መቅሠፍት ዗ይመጽእ ሊዕሇ ኃጥኣን

ወይመጽእ መቅሠፍተ እግዙአብሔር እምሰማይ ሊዕሇ ኵለ ሰብእ ኃጥእ ወዏማፂ እሇ ሇጽዴቅ ወይመይጥዋ በዓመፃሆሙ
እስመ አእምሮ እግዙአብሔር ክሡት በኀቤሆሙ ወአርአየ እግዙአብሔር ሊዕላሆሙ ወ዗ኢያስተርኢ እግዙአብሔር
እምፍጥረተ ዓሇም ይትዏወቅ በፍጥረቱ በኀሌዮ ወበአእምሮ ወከመዜ ይትአመር ኀይለ ወመሇኮቱ ዗ሇዓሇም ከመ ኢይርከቡ
እስመ እን዗ የአምርዎ ሇእግዙአብሔር አኮ ከመ እግዙአብሔር ዗አእኯትዎ ወሰብሕዎ ዗እንበሇ ዗ሐሰውዎ ወረኵሱ
በኅሉናሆሙ ወተጸሇሇ ሌቦሙ ወእን዗ ይፈቅደ ይጥብቡ አብደ ዕስመ ወሇጡ ስብሐቲሁ ሇእግዙአብሔር ዗ኢይመውት
ወአምሳሇ ርእየ ሰብእ መዋቲ ረሰዩ ወከመ እንስሳ ወከመ አራዊት ወከመ አዕዋፍ ወበእንተዜ አግብኦሙ ወኀዯጎሙ ከመ
ያርኵሱ ርእሶሙ ሇሉሆሙ ነፍስክሙ እስመ ሐሰተ ረሰይዎ ሇጽዴቀ እግዙአብሔር ወአምሇኩ ወተፀአፅኡ ተግባሮ
ወኀዯግዎ ሇፈጣሬ ኵለ ዗ውእቱ አምሊክ ቡሩክ ሇዓሇመ ዓሇም አሜን ወበእንትዜ ወሀቦሙ እግዙአብሔር መቅሠፍተ እኩየ
ወአንስቲያሆሙኒ ኀዯጋ ፍጥረቶን ወተመሰሊ በ዗ኢኮነ ፍጥረቶን ወከማሁ ዕዯዊሆሙኒ ኀዯጉ አንስቲያሆሙ ወነደ
በፍትወቶሙ ወገብኡ በበይናቲሆሙ ብእሲ ሊዕሇ ብእስ ኀሣሮሙ ገብሩ ወባሕቱ ፍዲሆሙ ወይገብእ ጌጋዮሙ ዱበ

80
ርእሶሙ ወበከመ ኢኀሇይዎ ሇእግዙአብሔር በሌቦሙ ከማሁ እግዙአብሔርኒ ወሀቦሙ ሌበ እበዴ ከመ ይግበሩ ዗ንተ
዗ኢይዯለ እ዗ እሙንቱ ጽጉባነ ኵለ ዓመፃ ወእከይ ወጹግ ወትዕግሌት ወጽጉባነ ቅንአት ሐማምያን ቀታሌያን ጕሕሊውያን
መስተመይናን ዜኁራን እኩያነ ሌማዴ ወግዕዜ ሐማይያን መስተሣሌቃን መስተሐብባን ዕቡያን ዜለፋን ዏሊውያን ሐሳውያን
ጸሊእያነ እግዙእ ሥሑጻን አብዲን ወዜንጉዓን ወመስተኃሥሣን ሇእከይ ወአሌቦሙ ምሕረት እን዗ ሇሉሆሙ የአምሩ ኵነኔሁ
ሇእግዙአብሔር ከመ ይዯሌዎ ሞት ሇ዗ገብረ ዗ንተ ከመዜ አኮ ባሕቲቶሙ ዗ይገብርዎ ዓዴ ይዌሕክዎ ያግብእዎ።

ኀበ ሰብአ ሮሜ

በእንተ ዗ይግዕዜ ቢጾ ዕጓሇ እመሕያው በምንትኑ ሇመኯንነ ጽዴቅ ሶበ አንተ ዜኯ ዗ትግዕዜ በሊዕሇ ባዕዴ ሶበ ሇሌከ
ትገብር ዜኯ ዗ትጸሌእ በሊዕሇ ቢጽከ አኮኑ ርእሰከ ትግዕዜ እስመ ሇሉከ ትግበር ከመ ጽዴቅ ኵነኔሁ ሇእግዙአብሔር
዗ያመጽእ መቅሠፍተ ሊዕሇ እሇ ይገብሩ ዗ንተ ከመዜ ኀሌዮ እስኩ ዕጓሇ እመሕያው ሇዜንቱ ሇእመ ብከ ኀበ ታመሥጥ
እምኵነኔሁ ሇእግዙአብሔር ሶበ ሇሉከ ትገብር ሇዜኩ ዗ትግዕዜ በሊዕሇ ባዕዴ ወ዗ትጸሌእ ትትሔ዗ብኑ ታስተዏብድ
ሇእግዙአብሔር ምሕረቱ ወበትዕግሥቱ ወበኦሆ ብሂልቱ ሊዕላከ ከመ ምሕረቱ ሇእግዙአብሔር ኪያከ ያገርር ኀበ ንስሓ
ወባሕቱ በአምጣነ ታጸንዕ ሌበከ ወኢትኔስሕ ት዗ግብ ሇከ መቅሠፍተ ሇአመ ይበጽሐከ ኵነኔሁ ሇእግዙአብሔር እስመ
ውእቱ ይፈዴዮ ሇኵለ በከመ ምግባሩ በኵነኔ ጽዴቁ ሇባዕዴን።

በእንተ ተዏጋሥያን ወከሓዴያን

ወሇእሇሰ ተዏገሡ በምግባር ሠናይ ወየኀሥሡ ክብረ ወስብሐተ ውእቱኒ ይሁቦሙ ሕይወተ ዗ሇዓሇም ወሇእሇሰ ከሓዴያን
ወዓሊውያነ ጽዴቅ ወመፍቀርያነ ዏመፃ ፍዲሆሙ መቅሠፍት ወመንሱት ወቍጥዓ ወገዓር ወምንዲቤ ሇነፍሰ ኵለ ሰብእ
዗እኩይ ምግባሩ እመኒ አይሁዲዊ ወእመኒ አረማዊ ክብር ወስብሐት ወሰሊም ሇኵለ ዗ሠናይ ምግባሩ እመኒ አይሁዲዊ
ወእመኒ አረማዊ እስመ ኢያዯለ እግዙአብሔር ሇገጸ ሰብእ እስመ ኵሙ እሇ ፈዴፈዯ እምሕግ አበሆሙ ፈዴፈዯ እምሕግ
ኵነኔሆሙ ወኵልሙ እሇ ዗እንበሇ ሕግ አበሳሆሙ ዗እንበሇ ሕግ ኵነኔሆሙ።

በእንተ እሇ ይፈቅደ ይጽዯቁ በአጽምዖ መጻሕፍት

ቦኑ በአጽምዖ መጻሕፍት ይጸዴቁ በቅዴመ እግዙአብሔር አኮኑ ዲእሙ ይጸዴቁ አሕዚብኒ እሇ አሌቦሙ ሕግ ይሠርዐ
ልሙ ሕገ ልሙ ሇሉሆሙ ወይገብሩ ዗በሕጎሙ ገቢረ ሕግ እን዗ ጽሑፍ ውእቱ ውስተ ሌቦሙ እምግባሮሙ
ወያርሰሐስሖሙ ሌቦሙ ወይቀሌዮሙ ወየምሩ ከመ አሌቦ ዗ይብለ ወ዗ይወቅሱ አመ እግዙአብኄር ሇዕጓሇ እመሕያው
዗የኀብኡ ወ዗ይከብቱ ውስተ ሌቦሙ በከመ መሀርኩ በእንተ ኢየሱስ ክርስቶስ።

በእንተ ዗ሇባዕዴ ይሜህር ወሇሉሁ ኢይገብር

ወሶበ አንተ አይሁዲዊ ዗ተዏርፍ በኦሪትከ ወትትሜካሕ በእግዙአብሔር ወተአምር ፈቃድ ወትፈሌጥ ዗ይኄይስ ወምሁር
አንተ ኦሪተ ወትትአመን ከመ አንተ መርሖሙ ሇዕዉራን ወብርሃኖሙ ሇእሇ ውስተ ጽሌመት ወመጥብቢሆሙ ሇአብዲን
ወመምህሮሙ ሇሕፃናት ወትትሜሰሌ ጻዴቀ ወተአምር ሕገ ኦሪት በ዗ትጸዴቅ ወእፎኑ እንከ ዗ኢትሜህር ርእሰከ ዗ሇባዕዴ
81
ትሜህር ኢትስርቁ ትብሌዋ ወሇሉከ ትሰርቅ ኢት዗ምዉ ትብሌ ወሇሉከ ትዛሙ ወተሐውር ኀበ ብእሲተ ብእሲ ጣዖተ
ወሇሉከ ትሰርቅ ቤተ መቅዯስ ወትትሜካሕ በኦሪት ወሇሉከ ዏሊዊሃ ሇኦሪት ሇእግዙአብሔር።

በእንተ እሇ ይፀርፉ ሊዕሇ ስመ እግዙአብሔር

ወናሁ በእንቲኣክሙ ይፀርፉ አሕዚብ ሊዕሇ ስመ እግዙአብሔር በከመ ጽሑፍ ግዜረትሰ ትበቍዏከ ሇእመ ገበርከ ኦሪት
ወእመሰ ኢገበርከ ኦሪት ግዜረትከ ቍሌፈተ ትከውነከ ወሇእመሰ እን዗ ቇሊፍ አንተ ወዏቀብከ ኦሪት ቍሌፈትከ ግዜረተ
ትከውነከ።

82
 Appendix D: Sample Suggestion

invalid word flagged by GLSC Total No of suggestion by GLSC Only correct Suggestion
በንተ 4 3
ሇእግዙኢነ 1 1
እየሱስ 3 1
዗ተጸውአ 1 1
ወተፊሌጠ 2 2
በሉሰን 2 2
ሇትምተ 5 4
መሊኪተ 3 3
ነጉረ 4 4
በአፈነቢያቲሁ 2 1
ወሉደ 2 2
ሰብ 1 1
ወበመ ንፈሱ 1 1
አትሙኒ 1 1
አም 5 5
ሃይማኖሙ 1 1
እሰማይ 2 2
እሇሇ 5 5
ኩሌክሙ 2 2
በኵሌ 2 1
መፈስየ 2 2
዗ሌፈፈ 2 2
ትርቡ 2 2
በንተዜ 1 1
እስከይእዛ 3 3
ዕስትየ 1 1
አሕዚቡኒ 3 3
እምሂር 1 1
ፈዯፋዯ 2 2
ወንዯለ 2 2
ይትወወቅ 1 0
ወመሇከቱ 2 2
ዕስመ 3 3
ስብሀቲሁ 1 1
ሇግዙአብሔር 2 2
ንስሳ 1 1
አግብሙ 1 1

83
ወኀዯጎኦሙ 2 2
ወሀሙ 1 1
ኀዯጋኣ 2 2
በበእይናቲሆሙ 2 2
ብእስ 2 2
ርይእሶሙ 1 1
ከሚሁ 4 4
እ዗ 4 4
ወጽጉባኣነ 2 2
በጾ 4 2
ስጽዴቅ 3 3
ዝበ 2 2
ሇሌከ 5 5
ሊሇ 3 1
ትብር 5 5
ትትኤሔ዗ብኑ 1 1
ምህሕረቱ 1 1
ወበወሆ 2 2
ብይልቱ 3 2
ትእ዗ግብ 2 2
ሇባዴን 2 2
ክ ብረ 5 5
አበሆሙ 2 2
዗ንበሇ 3 3
እግዙአብኄር 2 2
ርህሰከ 1 1
ጣወተ 2 2
ወትትሜካ 2 2
ሇኦወሪት 3 3
ሕገኦሪት 3 2
መጻሐፍ 1 1
በሐበ 3 3
ቦእሇ 5 5
ነቢከ 3 3
ወሰገእዯ 2 2
ወኅበር 5 5
ወአወሰበ 1 1
ከመጥንተ 3 3
ባህቲ 2 2
ሇበሌ 5 5
አበሁ 4 3

84