Parts of Speech Tagging for Afaan Oromo
Parts of Speech Tagging for Afaan Oromo
|Getachew Mamo WegariInformation Technology DepartmentJimma Institute of TechnologyJimma, Ethiopia||Million Meshesha (PhD)Information Science DepartmentAddis Ababa UniversityJimma, Ethiopia|
(IJACSA) International Journal of Advanced Computer Science and Applications,Special Issue on Artificial Intelligence
Abstract—The main aim of this study is to develop part-of-speech tagger for Afaan Oromo language. After reviewing literatures on Afaan Oromo grammars and identifying tagset and word categories, the study adopted Hidden Markov Model (HMM) approach and has implemented unigram and bigram models of Viterbi algorithm. Unigram model is used to understand word ambiguity in the language, while bigram model is used to undertake contextual analysis of words.For training and testing purpose 159 sentences (with a total of 1621 words) that are manually annotated sample corpus are used. The corpus is collected from different public Afaan Oromo newspapers and bulletins to make the sample corpus balanced. A database of lexical probabilities and transitional probabilities are developed from the annotated corpus. These two probabilities are from which the tagger learn and tag sequence of words in sentences.The performance of the prototype, Afaan Oromo tagger is tested using tenfold cross validation mechanism. The result shows that in both unigram and bigram models 87.58% and 91.97% accuracy is obtained, respectively.
Keywords-Natural Language processing; parts of speech tagging; Hidden Markov Model; N-Gram; Afaan Oromo.
At the heart of any natural language processing (NLP) task, there is the issue of natural language understanding. However, the process of building computer programs that understand natural language is not straightforward. As explained in , natural languages give rise to lexical ambiguity that words may have different meanings, i.e. one word is in general connected with different readings in the lexicon. Homograph, the phenomenon that certain words showing different morpho-syntatic behavior are identically written. For instance, the word ‘Bank’ has different meanings; Bank (= financial institute), Bank (= seating accommodation), etc.
In other words, words match more than one lexical category depending on the context that they appear in sentences. For example, if we consider the word miilaa ‘leg’ in the following two sentences,
Lataan kubbaa miilaa xabata. ‘Lata plays football’.
Lataan miilaa eeraa qaba. ‘Lata has long leg’.
In the first sentence, miilaa ‘leg’ takes the position of adjective to describe the noun kubbaa ‘ball’. But in the second sentence, miilaa is a noun described by eeraa ‘long’.
|alter(A, B, prevtag(C))||Change A to B if preceding tag is C|
|alter(A, B, nexttag(C))||Change A to B if the following tag is C|
- the probability that one tag follows another (n-gram); for example, after a determiner tag an adjective tag or a noun tag is quite likely, but a verb tag is less likely. So in a sentence beginning with the run…, the word ‘run’ is more likely to be a noun than a verb base form.
- The probability of a word being assigned a particular tag from the list of all possible tags (most frequent tag); for example, the word ‘over’ could be a common noun in certain restricted contexts, but generally a preposition tag would be overwhelmingly the more likely one.
|P(word/tag )||*||P(tag/previous n tags)|
|Most frequent tag (likelihood)||N-gram (a prior)|
VII. AFAAN OROMO TAGSET AND CORPUS
|NN||A tag for all types of nouns that are not joined with other categories in sentences.|
|NP||A tag for all nouns that are not separated from postpositions.|
|NC||A tag for all nouns that are not separated from conjunctions.|
|PP||A tag for all pronouns that are not joined with other categories.|
|PS||A tag for all pronouns that are not separated from postpositions.|
|PC||A tag for all pronouns that are not separated from conjunctions.|
|VV||A tag for all main verbs in sentences.|
|AX||A tag for all auxiliary verbs.|
|JJ||A tag for all adjectives that are separated from other categories.|
|JC||A tag for adjectives that are not separated from conjunction.|
|JN||A tag for numeral adjectives|
|AD||A tag for all types of adverbs in the language.|
|PR||A tag for all preposition/postposition that are separated from other categories.|
|ON||A tag for ordinary numerals.|
|CC||A tag for all conjunctions that are separated from other categories.|
|II||A tag for all introjections in the language.|
|PN||A tag for all punctuations in the language.|
The collected corpus for the study was manually tagged by experts of linguists in the field. The tagging process is based on the identified tagset and corpus that is manually tagged, considering contextual position of words in a sentence. This tagged corpus is used for training the tagger and evaluates its performance. The total tagged corpus consists of 159 sentences (the total of 1621 tokens).
VIII. THE LEXICON
Lexicon was prepared from which the two probabilities are developed for the analysis of the data set.
TABLE III. SAMPLE OF LEXCON
A. Lexicon probability
The lexical probabilities have been estimated by computing the relative frequencies of every word per category from the training annotated corpus. All statistical information, that enables to develop probabilities, are derived automatically from a hand annotated corpus (the lexicon).
For instance, the lexical probability of the word Oromoon tagged with NN is calculated as:
C(Oromoon, NN) = 7
C(NN) = 334
|So, P(Oromoon/NN)||= C(Oromoon, NN)/C(NN)|
Where, C and P are count of and Probability, respectively.
TABLE IV. SAMPLE LEXICAL PROBABILITY
|Words with given lexical probability||Probability|
B. Transition Probability
In transitional probabilities, the information of one part-of-speech category preceded by other categories is developed from training lexicon corpus. For this study, bigram is used. Bigram considers the information of the category (t-1) preceded the target category (t).
That means, P(t/t-1), where t is – part-of-speech category.
For example, C($S) = 157
C(NN,$S) = 79P
(NN/$S) = C(NN, $S)/C($S)
TABLE V. SAMPLE TRANSITION PROBABILITY
A. Performance Analysis of the tagger
TABLE VI. AVERAGE TAGGER RESULTS
In the performance analysis, the tagger is repeatedly trained and tested following tenfold cross validation.
The algorithms of the tagger are tested with a corpus of 146 Afaan Oromo words in average in each test set and that is trained on the training set of 1315 words, and the result of each test are compared with a copy of the test set that is hand annotated. As a result, the results of the experiments for both bigram and unigram algorithms show an accuracy of 91.97% and 87.5% correctly tagged words in average respectively.
With this corpus, the distributions of accuracy performance in both models are not as far from each other. The maximum variation in the distribution of bigram and unigram models is 8.97 and 11.04 respectively. If the corpus is standardized, this variation will reduce since standardized corpus consist relatively complete representative of words for the language and fair distribution of words in training set and test are observed.
In bigram model, the statistical accuracy is performed more than unigram model. Bigram model uses probability of contextual information besides the highest probability of categories given a word in a sentence to tag the word. The difference accuracy rate from bigram to unigram is 4.39% with this dataset.
This indicates, contextual information (the position in which the word appear in sentence) affects the determination of word categories for Afaan Oromo language.
The support of Jimma University, Ethiopia, is greatly acknowledged. The authors would also like to acknowledge the help and support from Tigist Mazgabu, Dr. Mengesha Mamo and Mr. Fituma Tefera.
 Hermann Helbig. Knowledge representation and the semantics of natural language. Springer-Verlg Berlin Heidelberg, Germany, 2006.
 James Allen. Natural language Understanding. The Benjamin/Cummings Publishing company, Redwood City, Canada, 1995
 Gobinda G. Chowdhury. Natural Language Processing: Department of Computer and Information Sciences, University of Strathclyde, Glasgow G1 1XH, UK, http://www.cis.strath.ac.uk/cis/research/publications/ papers/strathcis_publication_320.pdf
 Ferran Pla and Antonio Molina. Natural Language Engineering: Improving part of- speech tagging using lexicalized HMMs_ 2004. Cambridge University Press, United Kingdom, 2004
 Mesfin Getachew. Automatic part-of-speech tagging for Amharic language an experiment using stochastic Hidden Markov Approach. MSc. Thesis. School of Graduate Studies, Addis Ababa University, 2001.
 Abara Nefa. Long Vowels in Afaan Oromo: A Generative Approach. M.A. Thesis. School of Graduate Studies, Addis Ababa University, 1988. Unpublished.
 Kula K. T., Vasudeva Varma and Prasad Pingali. Evaluation of Oromo-English Cross-Language Information Retrieval. In IJCAI 2007 Workshop on CLIA, Hyderabad (India), 2007.
 Morka Mekonnen. Text to speech system for Afaan Oromo. MSc Thesis. School of Graduate Studies, Addis Ababa University, 2001.Unpublished
 Diriba Magarsa. An automatic sentence parser for Oromo language. MSc Thesis. School of Graduate Studies, Addis Ababa University, 2000. Unpublished Assefa W/Mariam. Developing morphological analysis for Afaan Oromo text. MSc Thesis. School of Graduate Studies, Addis Ababa University, 2000. Unpublished
 Yenewondim Biadgo. Application of multilayer perception neural network for tagging part-of-speech for Amharic language. MSc Thesis. School of Graduate Studies, Addis Ababa University, 2005. Unpublished
 Gumii Qormaata Afaan Oromo. Caasluga Afaan Oromo. Komoshinii Aadaafi Tuurizimii Oromiyaa, 1996. Unpublished
 Pierre M. Nugues. An Introduction to Language Processing with Perl and Prolog. Springer-Verlag Berlin Heidelberg, Germany, 2006
 Daniel Jurafsky and James H. Martin. Speech and Language Processing: An Introduction to Natural Speech Recognition. Prentice-Hall, Inc., 2000.
 Sandipan Dand, Sudeshna Sarkar, Anupam Basu. Automatic Part-of-Speech Tagging for Bengali: An Approach for Morphologically Rich Languages in a Poor Resource Scenario. Department of Computer Science and Engineering Indian Institute of Technology Kharagpur, 2007
 Frank Van Eynde. Part-of-speech Tagging and Lemmatisation for the Spoken Dutch Corpus, Center for Computational Linguistics Maria-Theresiastraat 21 3000 Leuven, Belgium, 2000 Roger Garside and Nicholas Smith. A Hybrid Grammatical Tagger: CLAWS4, http://ucrel.lancs.ac.uk/papers/HybridTaggerGS97.pdf
 Simon STÅHL. Part-of-Speech Tagger for Swedish, Computer Science, Lund University, 2000
 Census report: Ethiopia’s population now 76 million. December 4th, 2008. http://ethiopolitics.com/news
 Yoshua Bengio and Yves Grandvalet. No Unbiased Estimator of the Variance of K-Fold Cross-Validation. Dept. IRO, Universit´e de Montr´eal C.P. 6128, Montreal, Qc, H3C 3J7, Canada, 2004 http://www.faqs.org/faqs/ai-faq/neural-nets/part3/section-12.html