implementation of speech tagging for teaching …

African Journal of Pure and Applied Science Education Volume 18, Number 1, pp 73 – 86, July 2020 www.ajopase.com; [email protected] ISSN 11187670

Tijani, R. A, Akiode, J. I AJOPASE, Vol. 18, No. 1, July 2020

IMPLEMENTATION OF SPEECH TAGGING FOR TEACHING NIGERIA ENDANGERED LANGUAGE:

CASE OF YORUBA LANGUAGE

Tijani, R. A and Akiode, J. I Computer Science Department, Federal College of Education, Osiele Abeokuta

[email protected], [email protected] Abstract

Part of Speech tagging (POS) is usually a foundation stage for many Natural Language Processing (NLP) applications. Yoruba language is possibly going into extinction in few decades to come if proper measure is not taken. Therefore, this research paper proposes a stochastic learning approach, Hidden Markov Model (HMM), an existing model to be modified and implemented on Yoruba language (corpus) to ease its teaching and learning to younger generations. Yoruba literatures on grammar and morphology were reviewed to understand the syntax of the language and also to identify possible tagsets. 14 distinct tagsets were identified from Penn Treebank tagset and 500 randomly selected sentences from Lagos Nigerian Women Union (NWU) speech corpus were tagged for training when 100 sentences were set aside for testing purpose. Since there is no readymade standard annotated corpus, the manual tagging process to prepare corpus for this work was adopted. Raw Yoruba text was tagged as training data for HMM to learn the sentence structure of the language which it uses for tagging subsequently. Experiment was conducted on the model with 100 sentences to evaluate its accuracy using Tag-wise precision method; a precision of 0.71 and recall of 0.76 was achieved. The application was made platform independent.

Key words: Tagging, POS, Tagset, Corpus, NLP

Introduction

The importance of language in aiding human existence cannot be glossed over; it allows individual to

communicate meaningfully and also distinguished human from other animals. The only way Avogadro was

able to tell us that equal volume of all gases under identical temperature and pressure contains the same

number of molecules makes language unconventional to people which can be both an individual property

i.e. as knowledge and a social property i.e. when it performs its functions (Haruna, 2017).

NLP is an application and research area that enables machine to learn, read and understand human

languages (Abiola, Adetunmbi, & Oguntimilehin, 2013). Speech tagging, an aspect of NLP known as part

of speech tagging (POS tagging) is important in aspects such as Speech Synthesis, Information Retrieval,

Speech Recognition and machine translation (Francis, 2014).

http://www.ajopase.com/

mailto:[email protected]

mailto:[email protected]



POS Tagset is a list of symbols (tags) that indicate grammatical part of speech for a particular corpus

(language) or set of tags from which taggers chooses the appropriate option for a particular word. This is

a requirement for providing lexical resources, specifically dictionary entries and grammar rules. The tag

sets and the terminology for tagging are defined by different projects such as Brown Corpus Tag-Set, Penn

Treebank Tag-Set etc.

Table 1: Penn Treebank Tag Set

S/No Tag Description S/No Tag Description

1 CC Coordinating Conjuction 16 POS Possesive ending

2 CD Cardinal number 17 VBG Verb, gerund/present participle

3 DT Determiner 18 VBN Verb, past participle

4 FW Foreign word 19 VBP Verb, non-3rd ps. sing. Present

5 IN Preposition 20 VBZ Verb, 3rd ps. sing. present

6 JJ Adjective 21 WDT wh-determiner

7 MD Modal 22 WP wh-pronoun

8 NN Noun, singular or mass 23 WP Possessive wh-pronoun

9 NNS Noun, plural 24 WR Bwh-adverb

10 NNP Proper noun, singular 25 RBR Adverb, comparative

11 NNPS Proper noun, plural 26 RBS Adverb, superlative

12 PP Possesive pronoun 27 RP Particle

13 PRP Personal pronoun 28 UH Interjection

14 RB Adverb 29 VBD Verb, past tense

15 TO To

Part of speech tag is usually place after each word in a text e.g. The/DT boy/NN ate/VB yam/NN (using

the above tag set)

In Nigeria, a multi-cultural country, there are predominantly three dialects which include Hausa, Igbo and

Yoruba. Communities are linguistically complex and due to fear of comprehension, people hardly use their

indigenous languages. English language plays the role of dominant language which to a large extent could

be said to be one of the causes of language regression in Nigeria (Ayenbi, 2019).

Yoruba Language

Yoruba is the third most spoken language in West Africa with over fifty Million speakers (Eze-Uzoamaka

& Oloidi, 2017). This language first appeared in writing in the 19th century with the first publication

produced by John Raban (Adedjouma, Aoga, & Igue, 2013). Yoruba language has twenty five (25) distinct

alphabets: twelve (7 oral and 5 nasal) vowels; eighteen consonants. It’s a toner language with three tones:




high, low and middle. The various Yorùbá dialects in Nigeria can be classified into three major dialect areas

viz:

i. North-West Yorùbá NW which include Ibadan, Oyo, Ogun and Lagos (Éko areas),

ii. Central Yoruba(CY) which include Igbomina, Yagba, Ife, Ekiti, Akure and Ijebu areas and

iii. South- East Yoruba (SEY) which include Okitipupa, Ondo, Owo, Sagamu and some parts of Ijebu

(Abiola, Adetunmbi & Oguntimilehin, 2015).

Morphology of Yoruba Language

Morphology in relation to linguistics simply means the study of the forms and structures of word in a

particular language. Morphological analysis is a fundamental piece of NLP; it is pivotal to language

comprehension and computerization as it represents word development in dialects.

Yoruba language is semi– agglutinative language which implies a language where words are comprised of

linearly successive morphemes with components representing a meaningful morpheme. As a rule, the

word order is Subject-Object-Action-word (Verb), though other word order applies since the language is

verse (Enikuomehin, 2015).

The language order can be

1. Sentence = Verb + Subject (Noun/Pronoun) + Object(Preposition/Noun)

Fig. 1.2: Tree diagram for Verb-Subject-Object (VSO)

Sentence

Verb Subject Object

Noun Pronoun Preposition

Noun

Yoruba Language

(25 Alphabets)

18 Consonants 12 Vowels

7 Oral 5 Nasal

Fig. 1.1: Yoruba Language Alphabets




2. Sentence = Subject + Verb(Noun/Pronoun) + Object(Adjective/Adverb)

Fig. 1.3: Tree diagram for Subject-Verb-Object (SVO)

Endangered Language

An endangered language is one on the path to extinction or a language that is at risk of falling out of use

as speakers die out i.e. speakers fail to pass it on from one generation to the next (Wamalwa & Maris,

2013). Research has it that the 7 billion inhabitants of the world speak only 3 per cent of the world’s 6,000

plus languages and more than half of the world’s population speaks English, Russian, Mandarin, Hindu or

Spanish. UNESCO decries the state that even languages with thousands of speakers are no longer being

acquired by children. It is estimated that about 90 per cent of the languages may be replaced by dominant

languages by the end of the 21st century. The threat posed by these five most spoken languages to the

third world countries where majority of their languages are minority is real and huge (Lee, 2016).

The identification of Yoruba language as endangered started a long time ago. Fabunmi (2005) and Fakoya

(2008), identified Yorùbá language as endangered and listed recommendations to help the language. Also,

in 2013, the same warning of extinction was made when research was done on its usage by secondary

school students in South Western part of the Country. He highlighted some recommendations which

include awareness, encouragement and making Yoruba an official language in its region (Balogun, 2013).

Oparinde (2018) could not agree less but also established Yoruba language as endangered and drew

attention to the eminent danger of extinction. He went further to list extinct languages such as Latin, Old

Dutch, Middle Dutch even Old English and Middle English among others. This paper upheld the position

that Yoruba language is endangered and proffers solution using POS tagging for easy

presentation/teaching of Yoruba language.

Model/Algorithms for POS Tagging

Existing Part of Speech tagging algorithms include Brill tagger, hidden Markov Model, the Baum-Weltch

algorithm e. t. c.

Sentence

Subject Verb Object

Noun Pronoun Adjective Adverb




i. Brill Tagger

Brill tagger is a machine learning method basically used for classification problems, where the goal is to

assign classifications to a set of samples. It does not use hand-crafted rules, language specific knowledge

or external lexicon but rather used rules learnt from detecting error (Larsson, 1995).

ii. Hidden Markov Model

The Hidden Markov Model (HMM) is a statistical tool for generating probabilistic sequence models

commonly used for POS-tagging which is an upgrade of Andrei Markov’s mathematical theory of Markov

process of the early 20’s. The tag sequence probability is determines by the product of transition

probability and emission probability (Garrette & Baldridge, 2012).

a. Viterbi Algorithm

This is a dynamic programming algorithm and the most common decoding algorithm for Hidden Markov

Models (HMMs). Viterbi algorithm is a search algorithm that avoids the polynomial expansion of a breadth

first search by trimming the search tree at each level. In a sentence or word sequence, HMM taggers

choose the tag sequence that maximizes as in formula:

𝑃 (𝑤𝑜𝑟𝑑| 𝑡𝑎𝑔) 𝑋 𝑃(𝑡𝑎𝑔| 𝑝𝑟𝑒𝑣𝑖𝑜𝑢𝑠 𝑁𝑡𝑎𝑔𝑠) (Antony, 2011).

b. Baum-Weltch / Expectation-Maximization (EM) algorithm

Baum-Weltch algorithm is one of the standard training algorithms for HMM. It is used where manually

annotated corpus is not available for training the Model. It works by iteratively computing an initial

estimate of the probabilities and use it to compute a better estimate thereby learn from each iteration.

Choice of POS Tagging Algorithm/Model

In 2003, HMM with Viterbi algorithm recorded overall highest accuracy and succeeded best in annotation

of both known and unknown words (Beata, 2003). In 2011, POS tagging on Malay using Trigram Hidden

Markov Model (THMM) recorded 67.9% accuracy (Mohamed, Omar, & Ab Aziz, 2011). Also in 2017, POS

Tagging of Bahasa Indonesia text achieved accuracy of 62.69 using Hidden Markov (Amrullah, Hartanto,

& Mustika, 2017). Hence, HMM is modified to accommodate Yoruba language since it has been used on

many other rich languages such as Malay, Swedish, Indonesia and so on.

Methodology

Word classes and sentence structure development of Yoruba language are evaluated from literature;

Penn Treebank tagset was adopted and discussions were made with Yoruba language experts to set




tagset. Also HMM was modified to accommodate Yoruba language. Finally, module from Natural

Language Toolkit (NLT) which is a suite of program modules, data sets and tutorials supporting research,

teaching in computational linguistics and natural language processing was adopted for the

implementation phase.

Tagset Preparation

We considered Penn Treebank tagset as a reference point for our tagset design diverging where

necessary. The Penn Treebank tagging guidelines for English proposed a set of 36 tags, which is believed

to be one of the standard tagset for English. However, the number and types of tags needed for POS

tagging differ from language to language. In the context of Yoruba language, we did not know of many

works on tagset design when we started the work. Thus, a Tagset of 14 distinctive tags was coined out of

the Penn Treebank tagset for the purpose of this study.

Table 2: Tagset used for Yoruba Language Tagging

Tag Description Tag Descriptio

n

Tag Description Tag Descriptio

NN Noun, singular or

Mass

RB Adverb CC Coordinating

Conjunction

MD Modal Verb

NNP Noun, plural DT Determina

nt

PRP Pronoun IN Preposition

NNPS Proper Noun

singular

TO To FW Foreign word

JJ Adjective VB Verb CD Cardinal number

Corpus Preparation

LAGOS-NWU (Lagos Nigeria Women Union) Yoruba Speech Corpus was our choice of digital resources due

to its arrangement and accuracy in words spelling. This is an open source material which may be used free

of licensing charges. The texts form this corpus required no cleaning; it has an arrangement of one

sentence per line which made it suitable for the study. LAGOS-NWU Yoruba Speech has 21,728 words

containing around 8000 distinct words. This may not be as large as some English corpus which may contain




millions of words but it is the best we could get as at the time of this research. 500 randomly selected

sentences where manually tagged as there was no readily available tagged corpus of Yoruba language

available. The manual tagging is done by the researcher with the aid of English-to-Yoruba Yoruba-to-

English Dictionary. Due to the cumbersome nature of manual tagging and time constraint, approximately

five hundred (500) randomly selected sentences were tagged from the main corpus. These tagged

sentences were used for training of the model (HMM).

Tagging Process

A training corpus consisting of manually annotated sentences (i.e. 500 sentences approximately 2,000

words) from the main corpus was prepared; this is used in the training of HMM model. Adopting the

supervised learning approach, the manually tagged corpus served as input which allows the model to learn

the rules of the language. The corpus reader reads the contents of the corpus and passes it to the

Tokenizer which breaks down the sentence to word level (tokens) using space character. The tagset

analyzer extracts the tags from the words and stores them in the database. The decoding algorithm

(Viterbi algorithm) computes the Part of Speech (POS) tag probabilities which are important for finding

the sequence of words in the input sentence. These tagged texts were used for training the POS tagger;

the trained tagger takes untagged text as an input and tags the words based on the knowledge that it has

acquired during the training to produce tagged text as output. Randomly 100 sentences distinctive of the

training data were selected for testing the model. This process can be represented using the diagram

below:

Fig. 2: Training process of HMM

Database

Training Corpus

Tokenizer

Tagset Tagset Analyzer

POS Tag

Probabilities

Tagger Model (HMM)




HMM Training Algorithm

# Input data format is “natural_JJ language_NN …”

make a map emit, transition, context

for each line in file

previous = “<s>” # Make the sentence start

context[previous]++

split line into wordtags with “ “

for each wordtag in wordtags

split wordtag into word, tag with “_”

transition[previous+“ “+tag]++ # Count the transition

context[tag]++ # Count the context

emit[tag+“ “+word]++ # Count the emission

previous = tag

transition[previous+” </s>”]++

# Print the transition probabilities

for each key, value in transition

split key into previous, word with “ “

print “T”, key, value/context[previous]

# Do the same thing for emission probabilities with “E”

Decoding with Viterbi Algorithm

Viterbi algorithm reduces the complexity of the HMM core issues ranging from finding the best part of

speech tag sequence for a given sequence of words in the input sentence to the polynomial time. For any

input sentence of length n, there are |K|n possible tag sequences. The exponential growth with respect to

the length 𝑛 means that for any reasonable length sentence, this search will not be tractable. The

probability of a label sequence given a set of observations is defined in terms of the transition probability

and the emission probability which is mathematically represented as:

𝑝(𝑥1 … 𝑥𝑛, 𝑦1 … 𝑦𝑛+1) = ∏ 𝑞(𝑦𝑖|𝑦1−2, 𝑦𝑖−1)

𝑛+1

1=1

∏ 𝑒(𝑥𝑖|𝑦𝑖)

𝑛

𝑖=1

… 𝑒𝑞 1.1




𝑟(𝑦1 … 𝑦𝑘) = ∏ 𝑞(𝑦𝑖|𝑦𝑖−2, 𝑦𝑖−1) ∏ 𝑒(𝑥𝑖|𝑦𝑖)

𝑘

𝑖=1

𝑘

𝑖=1

… 𝑒𝑞 1.1.1

Such that ‘r’ is considering the first k terms where 𝑘 ∈ {1. . 𝑛} and for any label sequence y1…yk. Also, is set

𝑆(𝑘, 𝑢, 𝑣) which is simply the set of all labeled sequences of length k that ends with the bigram(𝑢, 𝑣). We

have set of sequences 𝑦1 … 𝑦𝑘 such that 𝑦𝑘−1 = 𝑢, 𝑦𝑘 = 𝑣, 𝜋(𝑘, 𝑢, 𝑣) which is simply the sequence with

the maximum probability and can finally be defined as

𝜋(𝑘, 𝑢, 𝑣) = max⟨𝑦1…𝑦𝑘⟩ ∈ 𝑆(𝑘,𝑢,𝑣)

𝑟(𝑦1 … 𝑦𝑘) … 𝑒𝑞 1.2

𝜋(𝑘, 𝑢, 𝑣) = (𝜋(𝑘 − 1, 𝑤, 𝑢) × 𝑞(𝑣|𝑤, 𝑢) × 𝑒(𝑥𝑘|𝑣)) … 𝑒𝑞 1.3

Finally, we have

max

(𝑦1…𝑦𝑛+1)

𝑝(𝑥1 … 𝑥𝑛, 𝑦1 … 𝑦𝑛+1) = max𝑢∈𝑘,∈𝑣∈𝑘

(𝜋(𝑛. 𝑢, 𝑣) × 𝑞(𝑆𝑇𝑂𝑃|𝑢, 𝑣)) 𝑒𝑞 1.4

The pseudo code for Viterbi Algorithm

Input: a sentence 𝑥1 … 𝑥𝑛, parameters 𝑞(𝑠|𝑢, 𝑣) and 𝑒(𝑥|𝑠)

Initialization: Set 𝜋(0,∗,∗) = 1, and 𝜋(0, 𝑢, 𝑣) = 0 for all (𝑢, 𝑣) such that 𝑢 ≠∗ or 𝑣 ≠∗

Algorithm:

For 𝑘 = 1 … 𝑛,

- For 𝑢 ∈ 𝑘, 𝑣 ∈ 𝑘.

𝜋(𝑘, 𝑢, 𝑣) = max𝑤∈𝑘

(𝜋(𝑘 − 1, 𝑤, 𝑢) × 𝑞(𝑣|𝑤, 𝑢) × 𝑒(𝑥𝑘|𝑣))

𝑏𝑝(𝑘, 𝑢, 𝑣) = 𝑎𝑟𝑔 max𝑤∈𝑘

(𝜋(𝑘 − 1, 𝑤, 𝑢) × 𝑞(𝑣|𝑤, 𝑢) × 𝑒(𝑥𝑘|𝑣))

Set (𝑦𝑛−1, 𝑦𝑛) = max𝑢,𝑣

(𝜋(𝑛, 𝑢, 𝑣) × 𝑞(𝑆𝑇𝑂𝑃|𝑢, 𝑣))

For𝑘 = (𝑛 − 2) … 1,

𝑦𝑘 = 𝑏𝑝(𝑘 + 2, 𝑦𝑘+1, 𝑦𝑘+2)

Return the tag sequence 𝑦1 … 𝑦𝑛

The above algorithm has execution time of O(n|K|3), hence it is linear in the length of the sequence, and

cubic in the number of tags.

Implementation and Result

Software Specification




Natural Language Toolkit (NLTK) and Python programming Language Python with its Libraries are used.

PyCharm by jetbrain is adopted as the IDE and other Python libraries utilized for the purpose of this study

may include

● Scikit-learn, a Python module for machine learning which is built on SciPy. It runs on Python

version 3.5 or higher; NumPy version 1.11.0 or higher; SciPy version 0.17.0 or higher.

● Python NumPy which is often referred to as python alternative to MATLAB and often used with

SciPy module.

Flow Chat of Tagging Process

Tokenizer

Start

Input Yoruba Corpus (One sentence per

line)

Split sentence into words using space delimiter (“ ”)

Trained Model

Manually

Annotated

Corpus

Calculate Probable POS Tag

using Viterbi Algorithm

Assign POS Tag to each token

Display Tagged Corpus

Stop

Tokens

Fig. 3: Tagging Process




Result

Evaluation of the Model

Tag-wise precision recall is used to evaluate the model.

Precision = or

Recall = or

f-measure = or

Accuracy =

Where, TP= True positive, FP= False positive, TN= True negative and FN=False negative.

Tables 4.2 and 4.3 showed the precision and recall as obtained from the test as well as the F-measure

which is calculated from the relationship between recall and precision when β is 0.5.

Recall Precision

No of correct POS tag in gold data

No of correct tagged POS in tagged data

No of total POS tag in tagged data

No of correct tagged POS in tagged data

β (precision) + (Recall)

(β + 1.0) (Precision)(Recall)

TP

TP + FN

β x 1 + (1 - β) x 1 1

TP + TN

TP + TN+FP+FN

TP

TP + FP

HMM output- Notepad

Fig. 4.1: Output




S/No Tags Precision Recall F-measure

1 CC 0.73 0.76 0.74

2 CD 0.61 0.70 0.65

3 DT 0.68 0.71 0.69

4 FW 0.71 0.75 0.73

5 IN 0.63 0.69 0.66

6 JJ 0.59 0.61 0.60

7 MD 0.71 0.73 0.72

8 NN 0.61 0.69 0.65

9 NNS 0.71 0.71 0.71

10 NNPS 0.80 0.82 0.81

11 PRP 0.67 0.70 0.68

12 RB 0.78 0.80 0.79

13 TO 0.63 0.65 0.64

14 VB 0.71 0.72 0.71

Average 0.68 0.72 0.70

Discussion

The result of the experiment is satisfactory and it demonstrated the effectiveness of HMM as a POS

tagging algorithm using very small amount of training data set. However, this research might not record

a very high accuracy and precision like English language taggers but it is a ground breaking result for a

Language like Yoruba that has a few NLP tools such as tagset, corpus and so on. Also, the choice of

stochastic learning approach, HMM is a good choice known to improve in accuracy as size of training data

increases.

S/No Tags Precision Recall F-measure

1 CC 0.83 0.82 0.82

2 CD 0.64 0.72 0.68

3 DT 0.71 0.79 0.75

4 FW 0.69 0.73 0.71

5 IN 0.70 0.72 0.71

6 JJ 0.63 0.69 0.66

7 MD 0.70 0.74 0.72

8 NN 0.65 0.70 0.67

9 NNS 0.72 0.76 0.74

10 NNPS 0.83 0.87 0.85

11 PRP 0.65 0.73 0.69

12 RB 0.63 0.70 0.66

13 TO 0.75 0.80 0.77

14 VB 0.76 0.83 0.79

Average 0.71 0.76 0.73

Table 4.3: 100 Sentences

Fig.4.2.1: Accuracy graph for 50 sentences

Table 4.2: 50 Sentences

Fig.4.3.1: Accuracy graph for 100 sentences




Conclusion

This paper is focused on creating a platform that can break down Yoruba corpus into appropriate part of

speech for basic understanding of words, function and its uses. This platform will simplify the teaching of

the language to younger generation and also build their interest to want to learn more about the

language. Such modern digital resources will be attractive to younger learners, providing a fun and

painless route to serious learning.

References

Adedjouma, S. R, Aoga, J. O, & Igue, M. (2013). Part-of-Speech tagging of Yoruba Standard , Language of

Niger-Congo family, 1(1), 2–5.

Abiola, O. B., Adetunmbi, A. O., & Oguntimilehin, A. (2013). A Computational Model of English To Yoruba

Noun-Phrases Translation System, 2013(1), 34–43.

Abiola, O. B., Adetunmbi, A. O., & Oguntimilehin, A. (2015). Using Hybrid Approach for English- To-

Yoruba Text to Text Machine Translation System ( PROPOSED ), 4(8), 308–313.

Amrullah, A. Z, Hartanto, R & Mustika, I. W. (2017). A comparison of different part-of-speech tagging

technique for text in Bahasa Indonesia. Proceedings - 2017 7th International Annual Engineering

Seminar, InAES 2017. https://doi.org/10.1109/INAES.2017.8068538

Ayenbi, O. F. (2019). Language regression in Nigeria, (April), 51–64. https://doi.org/10.4000/esp.136

Balogun, T. A. (2013). African Nebula , Issue 6 , 2013 An Endangered Nigerian Indigenous Language : The

Case of Yorùbá Language Temitope Abiodun Balogun Balogun , An Endangered Nigerian Language,

(6). Retrieved from

https://pdfs.semanticscholar.org/8136/84d9e7b182ea85a0cdb53cef342e44122e1e.pdf?_ga=2.213

446038.326713198.1579613088-1127722245.1554325916

Beata, M. (2003). Comparing Data riven Algorithm for POS Tagging of Swedish. Centre for Speech

Technology,Department of Speech, Music and Hearing, Royal Insitute of Technology, Sweden, SE-

100, 44–52.

Enikuomehin Oluwatoyin A. (2015). A Computerized Identification System for Verb Sorting and

Arrangement in a Natural Language : Case Study of the Nigerian, 3(1), 43–52.

Eze-Uzoamaka, P., & Oloidi, J. (2017). International Journal of Arts and Humanities ( IJAH ). International

Journal of Arts and Humanities (IJAH) Bahir Dar- Ethiopia., 6(21), 2006–2017.




https://doi.org/10.1111/MEC.14437

Fabunmi, F. A. (2005). Is Yorùbá an Endangered Language ? ∗, 14(August 2004), 391–408.

Fakoya, A. A. (2008). Endangerment scenario : The case of Yorùbá. California Linguistic, XXXIII(1), 1–23.

Francis, M. (2014). a Comprehensive Survey on Parts of Speech Tagging Approaches in Dravidian

Languages, (August), 7–10.

Garrette, D., & Baldridge, J. (2012). Type-Supervised Hidden Markov Models for Part-of-Speech Tagging

with Incomplete Tag Dictionaries. Emnlp, (July), 821–831.

Haruna, H. H. (2017). Linguistic Diversity and Language Endangerment: Towards the Revitalisation of

Bole Language In Nigeria. International Jornal for Innovative Research in Multidisciplinary Field,

3(10), 108–112. Retrieved from

https://www.researchgate.net/publication/324528843_Linguistic_Diversity_and_Language_Endan

gernment_Towards_the_Revitalisation_of_Bole_Langauge_in_Nigeria

Larsson, M. (1995). Part-of-Speech Tagging Using the Brill Method. Technology.

Lee, N. H. (2016). Assessing levels of endangerment in the Catalogue of Endangered Languages ( ELCat )

using the Language Endangerment Index. Language in Society, 2(45), 271–292.

https://doi.org/10.1017/S0047404515000962

Mohamed, H., Omar, N., & Ab Aziz, M. J. (2011). Statistical malay part-of-speech (POS) tagger using

Hidden Markov approach. 2011 International Conference on Semantic Technology and Information

Retrieval, STAIR 2011, (June), 231–236. https://doi.org/10.1109/STAIR.2011.5995794

Oparinde, K. M. (2018). Language Endangerment : The English Language Creating a Concern for the

Possible Extinction of Yorùbá Language Language Endangerment : The English Language Creating a

Concern for the Possible Extinction of Yorùbá Language *. Journal of Communication, 8(2), 112–

119. https://doi.org/10.1080/0976691X.2017.1396006

Antony, P. J (2011). Parts Of Speech Tagging for Indian Languages: A Literature Survey. International

Journal of Computer Applications, 34(8), 975–8887.

Wamalwa, E. W., & Maris, S. (2013). Language Endangerment and Language Maintenance : Can

Endangered Indigenous Languages of Kenya Be Electronically Preserved ? Language Endangerment

and Language Maintenance : Can Endangered Indigenous Languages of Kenya Be Electronically

Preserved ? Nternational Journal of Humanities and Social Science, 3(7), 258–266.


implementation of speech tagging for teaching …

Documents