Facilitating Cross-Lingual Information Retrieval Evaluations for African Languages
Bibliographic record
Abstract
Web resources are becoming more available in various languages, increasing the importance of cross-lingual information retrieval (CLIR) in accessing information that is present in a different language. To support CLIR studies, test collections are actively curated in the information retrieval (IR) field for the evaluation of methods and systems. Resources which support the evaluation of CLIR for African languages exist, however, these resources are few and are mostly curated synthetically or through translation, making them biased towards certain retrieval methods or prone to “Translationese” issues. Current resources also have document collections collected from sources with scarce resources for African languages, potentially limiting the provision of documents relevant to a search query. To address these, we present CIRAL, a test collection covering retrieval between English and four African languages: Hausa, Somali, Swahili and Yoruba. With its corpora developed from African news and blogs, which are rich sources of textual data for these languages, CIRAL was formulated for the passage ranking task with queries in English and passages in the African languages. Native speakers of the African languages develop the queries and provide query-passage relevance assessment. As often done in IR to curate test collections and promote research participation in CLIR, CIRAL was hosted as a shared task at the Forum for Information Retrieval and Evaluation (FIRE) 2023, where pools were collected for a subset of the collection. In this thesis, we provide a detailed description of CIRAL as a body of work, covering its curation process and shared task. Additionally, we conduct retrieval and reranking experiments, evaluating the effectiveness of systems in CLIR for African languages and demonstrating the utility of CIRAL. These include BM25 baselines with query and document translations and dense retrieval baselines with multilingual dense passage retrievers. We also examine the zero-shot reranking capabilities of T5 cross-encoder models and Large Language Models (LLMs) such as GPT and Zephyr in CLIR for African languages. We hope CIRAL fosters CLIR evaluation and research in African languages, and hence the development of retrieval systems that are well-suited for such tasks.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.030 | 0.097 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.008 | 0.005 |
| Science and technology studies | 0.003 | 0.002 |
| Scholarly communication | 0.006 | 0.008 |
| Open science | 0.002 | 0.008 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.010 | 0.005 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".