Source-linked AI summary
Development of a General Purpose Sentiment Lexicon for Igbo Language
Emeka Ogbuju, Moses Onyesolu
TL;DR
General-purpose sentiment lexicons are scarce for low-resource languages, including Igbo. This paper builds IgboSentilex from Liu’s lexicon with translation and native-speaker additions, then reports 95.75% average polarity agreement in comparisons with Liu and NRC.
Problem
General-purpose sentiment lexicons are scarce in low-resource languages, limiting their availability for sentiment analysis without translation into English.
Source: §1 · p. 1Method
Results
95.75% average polarity agreement was reported for IgboSentilex in comparisons with Liu and NRC on BBC Igbo news corpora.
Source: §3 · pp. 2–3Takeaways & Limitations
The authors present IgboSentilex as a building block for Igbo technology, with potential applications including document classification.
Source: §4 · p. 3
Abstract
from arXiv · showhide
There are publicly available general purpose sentiment lexicons in some high resource languages but very few exist in the low resource languages. This makes it difficult to directly perform sentiment analysis tasks in such languages. The objective of this work is to create a general purpose sentiment lexicon for the Igbo language that can determine the sentiment of documents written in the Igbo language without having to translate it to the English language. The material used was an automatically translated lexicon by Liu and the manual addition of Igbo native words. The result of this work is a general purpose lexicon called IgboSentilex. The performance was tested on the BBC Igbo news channel. It returned an average polarity agreement of 95.75 percent with other general purpose sentiment lexicons.
1. Introduction
General-purpose sentiment lexicons are common in English but scarce for low-resource languages, and Igbo was absent from a broad multilingual lexicon effort. The paper motivates an Igbo lexicon as a resource for language technology and preservation.
1. Introduction
- Low-resource languages, including Igbo, have little representation in lexicon development; Igbo was also omitted from a 136-language lexicon effort.The paper frames Igbo lexicon development as supporting NLP tools for an endangered language.Source: §1 · p. 1
- The paper aims to create a general-purpose sentiment lexicon for Igbo, a language identified as one of the world’s 2,488 endangered languages.The introduction links this objective to developing NLP tools for language preservation.Source: §1 · p. 1
- English general-purpose lexicons include manually compiled unigrams and automatically compiled N-grams.Language-specific sentiment lexicons had also been developed for Turkish, Bengali, Telugu, Irish, Indonesian, Spanish, and Dutch.Source: §1 · pp. 1–2
2. Materials and Methods
The lexicon was built from Liu’s opinion lexicon, using automatic translation and native-speaker review for untranslated entries. Its development pipeline includes aggregation, translation and polarity assignment, followed by preprocessing.
2. Materials and Methods
- Liu’s lexicon supplied seed words, most of which were automatically translated; native speakers interpreted and classified 230 positive and 738 negative untranslated words.Figure 1 marks the manually translated terms in red.Source: §2 · p. 2, Figure 1 · p. 2
- The development pipeline progresses from data collection and aggregation through recursive translation and polarity determination to preprocessing.Figure 2 presents the stages of Igbo sentiment corpora development.Source: §2 · p. 2, Figure 2 · p. 2
- Preprocessing removes noise and diacritics or accents to normalize and tokenize the lexicons; development was carried out using R/RStudio.The passage describes removal of diacritics or accents as part of tokenization.Source: §2 · p. 2
3. Results and Discussion
IgboSentilex contains 7,000 words, and its evaluation on BBC Igbo news corpora produced 95.75% average polarity agreement against Liu and NRC. The paper also reports one corpus-level disagreement involving IgboSentilex.
3. Results and Discussion
- IgboSentilex contains 7,000 words: 2,100 positive and 4,900 negative, extending Liu’s lexicon.An additional 200 words came from the Igbo Bible corpus and received intuitive sentiment ratings from native translators.Source: §3 · p. 2
- The evaluation used eight BBC Igbo news corpora spanning categories such as entertainment, trending news, and movies, with document polarity based on which polarity’s words were more numerous.Figure 3 shows a sample BBC news corpus.Source: §3 · p. 2, Figure 3 · p. 3
- 95.75% average polarity agreement was reported in comparisons with Liu and NRC; the entire corpus agreed except ID 03, where Liu and NRC agreed but IgboSentilex did not.Table 1 is identified as the system’s performance comparison.Source: §3 · p. 3, Table 1 · p. 3
4. Conclusion
The conclusion presents IgboSentilex as a building block for Igbo-language technology and suggests uses including document classification. It proposes future work on more direct translations and evaluation beyond news items.
4. Conclusion
- The authors position IgboSentilex as a building block for Igbo technology, with potential uses including security, relief interventions, and document classification.They intend the work to inspire lexicon translation for other low-resource languages in Nigeria and NLP task design.Source: §4 · p. 3
- Future work will create more direct lexicons from single Liu root words and retest performance on Igbo corpora beyond news items.Source: §4 · p. 3