Surangika Ranathunga

Massey University

Unclaimed · AI-structured profileORCID OpenAlex

Massey University NLP researcher: math visual generation for education and low-resource translation post-editing

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-17) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.

Research profile

Output—

Not yet scored: the snapshot lists only two cataloged arXiv papers from 2026 and no verified publication totals, so overall output volume cannot be measured. This axis would normally count papers across years and venues.

Impact—

Not yet scored: no citation counts, h-index, or citation-based metrics were provided for this profile, so research influence cannot be quantified. This axis would measure citations, downstream uptake, and influence within NLP and education technology.

Focus100/100

Scores 100 on focus: the bibliometric field distribution contains exactly one field — educational technology / natural language processing (text-to-image and SVG generation for math education) — indicating a single tightly defined research line. Both cataloged 2026 papers reinforce it: one generates pedagogically aligned visuals for lower-primary mathematics, the other diagnoses constrained decoding in low-resource automatic post-editing, both within applied NLP.

Collaboration44/100

Scores 44 on collaboration. The snapshot records 8 distinct co-authors, forming a recurring core group — Isuru Wijesiri, Nisansa de Silva, Kavindu Warnakulasuriya, Aloka Fernando, H. D. E. Maduranga, S. K. Munasinghe, and K. P. T. I. Weerasekara — suggesting repeated work within a small circle rather than broad external networks. Authorship positions on the two cataloged papers range from middle to last author.

Activity—

Not yet scored: the data provides only two papers, both dated 2026, with no multi-year publication history or timestamped records, so research tempo cannot be assessed. This axis would measure publication frequency and consistency across recent years.

Reproducibility20/100

Scores 20 on reproducibility, based on 1 reproducibility-assessed paper with a mean readiness of 20. The one reviewed paper, Math2Visual-X, is rated low: the Evolve review notes that key ablation and statistical details are missing from the LLM-routed SVG generation evaluation. Only one paper has been assessed so far, making this score provisional.

Research summary

Surangika Ranathunga is an applied natural language processing researcher affiliated with Massey University (profile resolved from public bibliographic data via ORCID 0000-0003-0701-0204). The available snapshot concentrates on two 2026 arXiv papers spanning two strands of applied NLP. The first, Math2Visual-X, is a modular framework for pedagogically aligned visual generation for lower-primary mathematics: Evolve's review found that LLM-routed symbolic SVG generation outperformed text-to-image baselines in teacher evaluation (4.82 versus 3.86), while flagging missing ablation and statistical details and rating reproducibility readiness as low. The second, 'Confident but Wrong: A Constrained Decoding Diagnostic for Low-Resource Automatic Post-Editing', extends the same applied orientation toward low-resource machine translation, with Ranathunga in the last author position; Evolve has not yet published a verdict for it. Bibliometrically, the profile is exceptionally focused: the field distribution contains exactly one field — educational technology / natural language processing (text-to-image and SVG generation for math education) — yielding a focus score of 100. Collaboration is moderate (score 44) with 8 distinct co-authors recorded, centering on a recurring group that includes Isuru Wijesiri and Nisansa de Silva. Reproducibility scores 20 from one assessed paper. Output, impact, and activity could not be scored because the snapshot lacks citation counts and multi-year publication history.

Collaboration pattern

Collaboration is moderate (radar score 44): 8 distinct co-authors appear in the snapshot, forming a recurring core group — Isuru Wijesiri, Nisansa de Silva, Kavindu Warnakulasuriya, Aloka Fernando, H. D. E. Maduranga, S. K. Munasinghe, and K. P. T. I. Weerasekara — with Nisansa de Silva appearing twice in the frequent-coauthor list, implying repeated collaboration. On the two cataloged 2026 papers Ranathunga held middle and last author positions and was not the corresponding author on either.

Research keywords

low-resource natural language processingautomatic post-editingconstrained decodingdiagnostic evaluationtext-to-image generationSVG generationLLM routingmath word problemsprimary mathematics educationsymbolic compositionteacher-based evaluation

Highlights

LLM-routed SVG generation beats text-to-image baselines in teacher evaluation
In Evolve's review of Math2Visual-X (2026), the modular framework's LLM-routed symbolic SVG pipeline scored 4.82 versus 3.86 for text-to-image baselines in teacher evaluation of lower-primary math visuals; the review still flags missing ablation and statistical details.
Reproducibility readiness rated low for Math2Visual-X
Evolve rates Math2Visual-X 'low' for reproducibility readiness (reproMean 20 from 1 assessed paper), because key ablation studies and statistical significance details are absent from the evaluation of the LLM-routed SVG generation approach.
Last-author diagnostic on constrained decoding for low-resource automatic post-editing
'Confident but Wrong: A Constrained Decoding Diagnostic for Low-Resource Automatic Post-Editing' (arXiv, 2026) lists Ranathunga as last author; it extends the profile's applied NLP focus toward low-resource machine translation. Evolve has not yet published a verdict, keywords, or reproducibility rating for it.

Timeline

2026Co-authored Math2Visual-X, a modular framework for pedagogically aligned lower-primary math visuals generation, posted on arXiv (middle author position).
2026Authored 'Confident but Wrong: A Constrained Decoding Diagnostic for Low-Resource Automatic Post-Editing' on arXiv in the last author position.
2026Evolve reviewed Math2Visual-X: LLM-routed symbolic SVG generation beat text-to-image baselines in teacher evaluation (4.82 vs 3.86).
2026Evolve rated Math2Visual-X low on reproducibility readiness, citing missing ablation and statistical details; the profile radar recorded focus 100 and 8 distinct co-authors.

Papers in the Evolve catalog

2026-09-18
Math2Visual-X: A Modular Framework for Pedagogically Aligned Lower-Primary Math Visuals Generation

Frequent co-authors

Isuru Wijesiri×1 Nisansa de Silva×1 Kavindu Warnakulasuriya×1 Aloka Fernando×1 H. D. E. Maduranga×1 S. K. Munasinghe×1 K. P. T. I. Weerasekara×1 Nisansa de Silva×1

About this profile

This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.

Radar scores are computed from a limited public bibliometric snapshot; output, impact, and activity axes lacked sufficient data to score.
Only two papers are covered in the Evolve catalog; the second paper has no review verdict, keywords, or reproducibility rating yet.
Reproducibility score rests on a single assessed paper (reproPapers = 1) and should be read as provisional.
Country and career details were not provided; the profile is name-resolved from public bibliographic data linked to ORCID 0000-0003-0701-0204.