Financial sentiment beyond English
Investor sentiment and sentiment analysis appear on 490 recent papers across the AI and finance journals, but only 22 of them carry a multilingual, cross-lingual or non-English language keyword, and financial sentiment analysis on 6. Most of the world's markets are discussed in languages the standard tools were not built for. How well do the tools work there, and do they move markets?
What prompted this#
Since October 2025, 490 papers carry a sentiment keyword (206 in AI journals, 193 in finance, 50 in economics): "investor sentiment" on 186, "sentiment analysis" on 127, "multimodal sentiment analysis" on 74, "news sentiment" on 19, "financial sentiment analysis" on 6. Only 22 of the 490 also carry a multilingual, cross-lingual, low-resource or specific non-English language keyword, led by "low-resource languages" on 9 and "multilingual sentiment analysis" on 6.
Questions#
- Do financial sentiment models fine-tuned on English lose accuracy on Japanese, Korean, Arabic or Indonesian financial news, and do translation layers recover it?
- Does news sentiment predict returns in markets with fewer analysts and more retail investors, more or less than in US markets?
- Do the sentiment measures used in published papers on emerging markets reflect the languages their investors read?
Data that could help#
Financial news and social media in several languages, local market returns, and multilingual financial sentiment datasets, which are few.
Next step#
Build a labelled financial sentiment set in three non-English languages and compare the accuracy of English-trained models, multilingual models and translation pipelines.