Source-linked AI summary
Semantic Enrichment of Nigerian Pidgin English for Contextual Sentiment Classification
Wuraola Fisayo Oyewusi, Olubayo Adekanmbi, Olalekan Akinsande
TL;DR
Nigerian Pidgin’s meanings have evolved, creating a challenge for sentiment analysis based on English lexicons. The study updates VADER with 300 Pidgin tokens, applies it to about 14,000 tweets, and reports better labels than the original English lexicon.
Problem
Nigerian Pidgin’s evolving meanings and code-mixed use create a sentiment-analysis gap for English-based lexicons and models.
Source: §1 · pp. 1–2Method
The study updates VADER with 300 Nigerian Pidgin tokens, scores about 14,000 tweets, and compares its labels with expert-speaker labels.
Source: §2 · p. 2, §3 · p. 2, §A.1 · p. 4Results
Updated VADER produced better sentiment labels than the original English lexicon, while human annotators captured nuances that rule-based labeling missed.
Source: §4 · p. 3Takeaways & Limitations
The conclusion identifies increasing the dataset’s number of instances as further work.
Source: §4 · p. 3The authors note that more work could increase the number of instances in the dataset.
Source: §4 · p. 3
Abstract
from arXiv · showhide
Nigerian English adaptation, Pidgin, has evolved over the years through multi-language code switching, code mixing and linguistic adaptation. While Pidgin preserves many of the words in the normal English language corpus, both in spelling and pronunciation, the fundamental meaning of these words have changed significantly. For example,'ginger' is not a plant but an expression of motivation and 'tank' is not a container but an expression of gratitude. The implication is that the current approach of using direct English sentiment analysis of social media text from Nigeria is sub-optimal, as it will not be able to capture the semantic variation and contextual evolution in the contemporary meaning of these words. In practice, while many words in Nigerian Pidgin adaptation are the same as the standard English, the full English language based sentiment analysis models are not designed to capture the full intent of the Nigerian pidgin when used alone or code-mixed. By augmenting scarce human labelled code-changed text with ample synthetic code-reformatted text and meaning, we achieve significant improvements in sentiment scoring. Our research explores how to understand sentiment in an intrasentential code mixing and switching context where there has been significant word localization.This work presents a 300 VADER lexicon compatible Nigerian Pidgin sentiment tokens and their scores and a 14,000 gold standard Nigerian Pidgin tweets and their sentiments labels.
1 BACKGROUND
Nigerian Pidgin has evolved through code-switching and social change, altering words’ lexical and grammatical contexts and their meanings. This complicates sentiment analysis because English lexicons may not capture Pidgin usage, while prior work has explored code-mixed analysis and synthetic text.
1 BACKGROUND
- Intrasentential code-switching alternates languages within a sentence and can change lexical and grammatical context for situational, stylistic, and sociolinguistic reasons.The section describes Nigerian Pidgin as shaped by intertextuality and urban youth culture.Source: §1 · p. 1
- Pidgin serves daily communication for an estimated 3–5 million people and is a second language for up to 75 million in Nigeria.The section links its changing meanings to intertextuality and urban youth subculture.Source: §1 · p. 1
- VADER is a social-media-oriented lexicon and rule-based sentiment tool with about 9,000 tokens, including sentiment-bearing slang and emoticons.The section also notes prior code-mixed sentiment work and label transfer to synthetic code-switched text, with reported accuracy gains of 1.5%, 5.11%, and 7.20% across three language pairs (Samanta et al., 2019).Source: §1 · p. 2
2 METHOD
The study applies original and updated VADER to about 14,000 Nigerian Pidgin tweets, updating the lexicon with 300 Pidgin tokens and comparing its labels with expert speakers.
2 METHOD
- The updated VADER lexicon adds 300 Nigerian Pidgin tokens and their sentiment scores to the original lexicon.The study uses both versions to calculate compound sentiment scores for about 14,000 tweets.Source: §2 · p. 2
- Sentiment labels from updated VADER are compared with labels assigned by expert Pidgin English speakers.The passage describes this comparison but does not report its outcome.Source: §2 · p. 2
- Figure 1 is identified as the study’s methodology for semantic enrichment of Nigerian Pidgin English for contextual sentiment classification.The caption does not specify further methodological steps.Source: Figure 1 · p. 2
3 RESULTS
The results describe a 300-token translation from VADER’s English lexicon and note challenges in mapping between one-word Pidgin and English expressions.
3 RESULTS
- 300 Nigerian Pidgin tokens were translated from VADER’s English lexicon, but direct translations often produced phrases rather than single words.Some Pidgin words also corresponded to multiple English words.Source: §3 · p. 2
- The results section links to the Nigerian Pidgin tweet-and-sentiment resource and the 300-token lexicon with sentiment scores.The passages identify the resources but provide no further description of their contents.Source: §3 · p. 2
- Table 1 is captioned as presenting Nigerian Pidgin tweets with different sentiment labels.The caption does not specify the labels or their distribution.Source: Table 1 · p. 3
4 CONCLUSION
The authors report better sentiment labels from updated VADER than from the original English lexicon, while noting that human annotators capture nuances rule-based labels miss.
4 CONCLUSION
- Updated VADER generated better sentiment labels than the original VADER English lexicon.The conclusion does not quantify the improvement.Source: §4 · p. 3
- Human annotators captured nuances that rule-based sentiment labeling could not capture.The conclusion contrasts the two labeling approaches without specifying the kinds of nuances.Source: §4 · p. 3
- The authors state that more work could increase the number of instances in the dataset.The conclusion does not specify a target dataset size.Source: §4 · p. 3
A APPENDIX
The appendix identifies Table 2 as reporting average sentiment scores for Nigerian Pidgin sentiments with multiple English meanings.
A APPENDIX
- Table 2 reports average sentiment scores for Nigerian Pidgin sentiments with multiple English meanings.Source: Table 2 · p. 4
A.1 SELECTION OF DATA LABELLERS
Three people indigenous to or living in Nigeria’s South South region were briefed on word sentiments, and each data point was checked by at least one other person after initial labeling.
A.1 SELECTION OF DATA LABELLERS
- Three people indigenous to or living in Nigeria’s South South region were briefed on word sentiments, and each data point was verified by at least one other person after initial labeling.Source: §A.1 · p. 4