Detecting Clitics Related Orthographic Errors in Turkish

19806

Authors

nan

Abstract

AbstractFor the spell correction task, vocabulary based methods have been replaced with methods that take morphological and grammar rules into account. However, such tools are fairly immature, and, worse, non-existent for many low resource languages. Checking only if a word is well-formed with respect to the morphological rules of a language may produce false negatives due to the ambiguity resulting from the presence of numerous homophonic words. In this work, we propose an approach to detect and correct the “de/da” clitic errors in Turkish text. Our model is a neural sequence tagger trained with a synthetically constructed dataset consisting of positive and negative samples. The model’s performance with this dataset is presented according to different word embedding configurations. The model achieved an F1 score of 86.67% on a synthetically constructed dataset. We also compared the model’s performance on a manually curated dataset of challenging samples that proved superior to other spelling correctors with 71% accuracy compared to the second-best (Google Docs) with and accuracy of 34%.

Keywords

nan

Full Content

Proceedings of Recent Advances in Natural Language Processing, pages 71–76, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_009 71 Detecting Clitics Related Orthographic Errors in Turkish U˘gurcan Arıkan Onur Güngör Department of Computer Engineering, Bogazici University, Istanbul, Turkey {ugurcan.arikan,onurgu,suzan.uskudarli}@boun.edu.tr Suzan Uskudarli Abstract For the spell correction task, vocabulary based methods have been replaced with methods that take morphological and gram- mar rules into account. However, such tools are fairly immature, and, worse, non- existent for many low resource languages. Checking only if a word is well-formed with respect to the morphological rules of a language may produce false negatives due to the ambiguity resulting from the pres- ence of numerous homophonic words. In this work, we propose an approach to de- tect and correct the “de/da” clitic errors in Turkish text. Our model is a neural se- quence tagger trained with a synthetically constructed dataset consisting of positive and negative samples. The model’s perfor- mance with this dataset is presented accord- ing to different word embedding configu- rations. The model achieved an F1 score of 86.67% on a synthetically constructed dataset. We also compared the model’s per- formance on a manually curated dataset of challenging samples that proved superior to other spelling correctors with 71% accu- racy compared to the second best (Google Docs) with 34% accuracy. 1 Introduction Misspellings can change the meanings of words and, consequently, of sentences, which can lead to major miscommunication and frustration. This paper focuses on a common spelling error in Turk- ish, namely the spelling of the “de/da” clitic. Its written form (“de” and “da”) depends on the vowel harmony rule that is based on the last vowel of the word previous to the conjunction. When the final vowel of the prior word is in {e,i,ö,ü} the clitic is written as “de”, otherwise (in {a,ı,o,u}) it is written as “da”. For example, in the sentence “Selin de burada” meaning “Selin is also here”, the last word before the clitic (“de”) is “Selin” whose final vowel is “i”. Thus, the clitic is written as “de”. Whereas, in the sentence “Fatma da burada” meaning “Fatma is also here”, the last word before the clitic (“da”) is “Fatma” whose final vowel is “a”, causing the clitic to be written as “da”. The “de/da” clitic in Turkish is a conjunction when it is written separately and has the same mean- ing as "as well", "too", and "also" in English. In addition to being a conjunction, the “de” and “da” homonyms may be used as locative suffixes mean- ing “at” or “in”. For example, the word “araba” (car) with the suffix “-da” (“arabada”) means “in the car”. Although the “de/da” clitic in the meaning of conjunction must always be written separately, it is commonly confused with the locative suffix "de/da" and incorrectly written concatenated to the previous word. The misspelling of the "de/da" clitics alter the meaning of a sentence, and possibly render it mean- ingless. For example, when the clitic in the sen- tence "Araba da gördüm" is misspelled …

Content truncated to 500 words

Labeled Text

AbstractFor the spell correction task, vocabulary based methods have been replaced with methods that take morphological and grammar rules into account. However, such tools are fairly immature, and, worse, non-existent for many low resource languages. Checking only if a word is well-formed with respect to the morphological rules of a language may produce false negatives due to the ambiguity resulting from the presence of numerous homophonic words. In this work, we propose an approach to detect and correct the “de/da” clitic errors in Turkish text. Our model is a neural sequence tagger trained with a synthetically constructed dataset consisting of positive and negative samples. The model’s performance with this dataset is presented according to different word embedding configurations. The model achieved an F1 score of 86.67% on a synthetically constructed dataset. We also compared the model’s performance on a manually curated dataset of challenging samples that proved superior to other spelling correctors with 71% accuracy compared to the second-best (Google Docs) with and accuracy of 34%.

Publication Details

Paper ID:
19806
Year:
2019
Publication:
TIPSTER
Publication Type:
conf
DOI:
nan