Organizations: University of Toronto · Queen Mary University of London · Yale University · CoolWei AI Lab · University of Moratuwa · Ontario Tech University · Massey University · University of Essex
Sentence simplification aims to make complex text more accessible by reducing linguistic complexity while preserving the original meaning. However, progress in this area remains limited for mid-resource and low-resource languages due to the scarcity of high-quality data. To address this gap, we introduce the OasisSimp dataset, a multilingual dataset for sentence-level simplification covering five languages: English, Sinhala, Tamil, Pashto, and Thai. Among these, no prior sentence simplification datasets exist for Thai, Pashto, and Tamil, while limited data is available for Sinhala. Each language simplification dataset was created by trained annotators who followed detailed guidelines to simplify sentences while maintaining meaning, fluency, and grammatical correctness. We evaluate eight open-weight multilingual Large Language Models (LLMs) on the OasisSimp dataset and observe substantial performance disparities between high-resource and low-resource languages, highlighting the simplification challenges in multilingual settings. The OasisSimp dataset thus provides both a valuable multilingual resource and a challenging benchmark, revealing the limitations of current LLM-based simplification methods and paving the way for future research in low-resource sentence simplification. The dataset is available at https://OasisSimpDataset.github.io/.
Figures & tables
Lang
# Comp Sentences
Avg. Simp Sentences
Avg.Comp Length
Avg. Simp Length
Source Domain
English
2500
2.86
24.35
17.23
News
Sinhala
2500
5.00
30.12
28.78
Govt
Thai
1499
5.06
48.24
37.77
News
Tamil
520
4.66
23.22
17.65
Govt
Pashto
2500
3.00
28.81
20.31
Wiki
Table 1: Final statistics for each language in the OasisSimp dataset. Comp - complex, Simp - simplified.
Table 2: Sample Examples from the OasisSimp Dataset for each language.
Model
0 Shot
1 Shot
5 Shot
SARI Comp.
SARI
F ref
SARI Comp.
SARI
F ref
SARI Comp.
SARI
F ref
ADD
KEEP
DEL
ADD
KEEP
DEL
ADD
KEEP
DEL
Aya ✔
9.32
44.98
75.23
43.18
54.44
9.68
44.90
72.51
42.36
56.35
10.18
45.91
71.16
42.42
57.20
Cmd-R ✔
9.69
44.95
72.89
42.51
55.90
10.99
43.71
77.57
44.09
55.03
11.91
45.28
77.09
44.76
56.63
DeepSeek ✔
7.03
41.47
76.30
41.60
51.88
7.80
41.12
76.82
41.91
51.92
9.41
42.03
77.22
42.89
54.15
EuroLLM ✔
9.32
45.60
68.36
41.10
56.98
10.99
46.98
69.35
42.44
57.96
11.63
46.55
70.93
43.04
58.10
Table 3: Results on English ( OasisSimp-EN ) dataset : Zero, 1- and 5-shot Results. Highest value in bold and lowest underlined . ✔ : Language included in LLM training; ✔ : Likely included (Educated Guess); ✗ : Not included.
Model
0 Shot
1 Shot
5 Shot
SARI Comp.
SARI
F ref
SARI Comp.
SARI
F ref
SARI Comp.
SARI
F ref
ADD
KEEP
DEL
ADD
KEEP
DEL
ADD
KEEP
DEL
Aya ✗
0.47
23.11
68.49
30.69
54.94
0.40
26.01
67.39
31.27
54.22
0.21
27.62
65.75
31.20
44.78
Cmd-R ✗
0.18
29.95
65.05
31.73
58.32
0.19
30.24
65.08
31.84
55.94
0.13
28.87
65.18
31.40
45.34
DeepSeek ✗
0.16
16.92
68.71
28.60
-0.14
0.14
24.99
66.71
30.61
57.68
0.08
20.23
67.60
29.30
38.55
EuroLLM ✗
0.03
19.19
68.14
29.12
47.59
0.04
18.74
68.18
28.99
47.56
0.00
2.03
70.18
24.07
-60.47
Table 4: Results on Sinhala ( OasisSimp-SI ) dataset : Zero, 1- and 5-shot Results. Highest value in bold and lowest in underlined . ✔ : Language included in LLM training; ✔ : Likely included (Educated Guess); ✗ : Not included.
Model
0 Shot
1 Shot
5 Shot
SARI Comp.
SARI
F ref
SARI Comp.
SARI
F ref
SARI Comp.
SARI
F ref
ADD
KEEP
DEL
ADD
KEEP
DEL
ADD
KEEP
DEL
Aya ✗
0.33
21.01
84.30
35.22
56.49
0.38
24.45
83.53
36.12
59.01
0.43
25.98
82.59
36.33
60.66
Cmd-R ✗
0.47
29.02
82.19
37.23
62.17
0.51
26.88
81.85
36.41
61.84
0.45
28.89
81.86
37.07
62.26
DeepSeek ✗
0.12
15.27
84.34
33.24
27.98
0.18
15.80
83.95
33.31
51.68
0.14
15.77
84.09
33.34
40.95
EuroLLM ✗
0.15
26.35
80.43
35.64
52.25
0.23
26.26
80.23
35.57
60.26
0.16
28.05
77.71
35.31
62.57
Table 5: Results on Thai ( OasisSimp-TH ) dataset : Zero, 1- and 5-shot Results. Highest value in bold and lowest in underlined . ✔ : Language included in LLM training; ✔ : Likely included (Educated Guess); ✗ : Not included.
Model
0 Shot
1 Shot
5 Shot
SARI Comp.
SARI
F ref
SARI Comp.
SARI
F ref
SARI Comp.
SARI
F ref
ADD
KEEP
DEL
ADD
KEEP
DEL
ADD
KEEP
DEL
Aya ✗
2.42
18.48
70.60
30.50
72.42
2.25
27.49
68.73
32.82
74.40
2.00
34.27
66.62
34.30
76.24
Cmd-R ✗
2.78
26.40
69.28
32.82
74.14
2.07
30.78
66.61
33.15
74.97
1.44
34.40
66.15
34.00
76.08
DeepSeek ✗
0.33
11.88
70.83
27.68
-0.67
0.49
19.23
69.54
29.75
68.27
0.11
8.76
70.56
26.48
3.30
EuroLLM ✗
0.19
24.90
68.07
31.05
69.89
0.32
26.56
67.05
31.31
71.21
0.16
25.05
67.73
30.98
65.97
Table 6: Results on Tamil ( OasisSimp-TA ) dataset : Zero, 1- and 5-shot Results. Highest value in bold and lowest in underlined . ✔ : Language included in LLM training; ✔ : Likely included (Educated Guess); ✗ : Not included.
Model
0 Shot
1 Shot
5 Shot
SARI Comp.
SARI
F ref
SARI Comp.
SARI
F ref
SARI Comp.
SARI
F ref
ADD
KEEP
DEL
ADD
KEEP
DEL
ADD
KEEP
DEL
Aya ✗
0.62
23.98
67.47
30.69
49.17
1.08
45.60
58.62
35.10
60.83
1.77
53.81
47.17
34.25
68.25
Cmd-R ✗
0.75
50.82
51.73
34.44
61.91
0.93
54.41
44.44
33.26
67.84
0.70
56.53
35.62
30.95
70.52
DeepSeek ✗
0.52
41.19
60.71
34.14
38.65
0.90
48.83
54.59
34.78
63.65
0.91
50.16
52.51
34.53
66.26
EuroLLM ✗
0.50
54.28
44.40
33.06
67.55
0.65
54.87
43.28
32.93
69.72
0.78
55.37
42.09
32.75
70.42
Table 7: Results on Pashto ( OasisSimp-PS ) dataset : Zero, 1-, and 5-shot Results. Highest value in bold and lowest in underlined . ✔ : Language included in LLM training; ✔ : Likely included (Educated Guess); ✗ : Not included.
English
Model
0 Shot
1 Shot
5 Shot
Aya
18.38
20.59
21.72
Cmd-R
21.03
19.58
20.72
DeepSeek
17.19
17.27
18.91
EuroLLM
23.07
23.07
22.79
Gemma
15.48
16.00
18.92
Table 8: BLEU scores on English ( OasisSimp-EN ) dataset across different shot settings. Highest value in bold and lowest underlined .
Sinhala
Model
0 Shot
1 Shot
5 Shot
Aya
7.25
9.03
10.80
Cmd-R
11.86
11.90
11.63
DeepSeek
8.04
7.12
3.08
EuroLLM
2.24
2.19
0.00
Gemma
14.38
22.36
31.82
Table 9: BLEU scores on Sinhala( OasisSimp-SI ) dataset across different shot settings. Highest value in bold and lowest underlined .
Thai
Model
0 Shot
1 Shot
5 Shot
Aya
7.70
9.65
11.58
Cmd-R
13.08
12.42
13.84
DeepSeek
2.24
5.99
5.71
EuroLLM
11.76
13.94
16.73
Gemma
3.67
11.20
14.43
Table 10: BLEU scores on Thai( OasisSimp-TH ) dataset across different shot settings. Highest value in bold and lowest underlined .
Tamil
Model
0 Shot
1 Shot
5 Shot
Aya
11.26
20.69
29.97
Cmd-R
19.36
26.17
30.58
DeepSeek
3.67
10.11
1.48
EuroLLM
17.04
18.98
15.60
Gemma
9.62
21.36
36.71
Table 11: BLEU scores on Tamil( OasisSimp-TA ) dataset across different shot settings. Highest value in bold and lowest underlined .
Pashto
Model
0 Shot
1 Shot
5 Shot
Aya
7.42
22.04
32.73
Cmd-R
28.39
33.99
38.49
DeepSeek
18.37
25.15
26.92
EuroLLM
33.74
34.65
35.54
Gemma
7.22
12.84
22.65
Table 12: BLEU scores on Pashto( OasisSimp-PS ) dataset across different shot settings. Highest value in bold and lowest underlined .
Text simplification plays a crucial role in improving the accessibility and comprehensibility of written information for diverse audiences, including language learners and readers with limited literacy. Despite its importance, large-scale, high-quality datasets for training and evaluating text simplification models remain scarce for languages other than English. This paper reports an experimental study on the collection and processing of crowd-sourced simplification data from comparable corpora to construct a corpus suitable for both training and testing text simplification systems across multiple languages (Catalan, English, French, Italian and Spanish). We report mechanisms for sentence-level alignment from document-level data. The resulting dataset of the aligned sentence pairs is publicly available.
Kenji Hilasaca, Nouran Khallaf, Serge Sharoff
Centre for Translation, Localisation and Interpreting Studies School of Languages, Cultures and Societies University of Leeds, UK
Text simplification aims to make complex texts easier to understand while preserving their original meaning. Recent large language models can perform simplification through prompting, but it remains unclear whether adding explicit linguistic information about sentence complexity to the prompt improves their outputs. We investigate this question for multilingual sentence-level Easy-to-Read simplification in Catalan, Spanish, and Italian. Using Qwen3-8B, we compare a baseline prompt, a gold-cue prompt enriched with gold linguistic cues, and a predicted-cue prompt enriched with automatically predicted cues. We evaluate the outputs using SARI, BLEU, chrF, and BERTScore, and complement this evaluation with a manual qualitative analysis. Predicted-cue prompting obtains the best overall scores across all four metrics, although the gains over the baseline are small. Gold-cue prompting does not consistently improve over the baseline, and results vary across languages. These findings indicate that cue-based prompting can influence multilingual Easy-to-Read simplification, but its benefits are modest, metric-dependent, and language-dependent.
Text simplification supports second language (L2) learning by providing comprehensible input, consistent with the Input Hypothesis. However, constructing personalized parallel corpora is costly, while existing large language model (LLM)-based readability control methods rely on pre-labeled sentence corpora and primarily target English. We propose Re-RIGHT, a unified reinforcement learning framework for adaptive multilingual text simplification without parallel corpus supervision. We first show that prompting-based lexical simplification at target proficiency levels (CEFR, JLPT, TOPIK, and HSK) performs poorly at easier levels and for non-English languages, even with state-of-the-art LLMs such as GPT-5.2 and Gemini 2.5. To address this, we collect 43K vocabulary-level data across four languages (English, Japanese, Korean, and Chinese) and train a compact 4B policy model using Re-RIGHT, which integrates three reward modules: vocabulary coverage, semantic preservation, and coherence. Compared to the stronger LLM baselines, Re-RIGHT achieves higher lexical coverage at target proficiency levels while maintaining original meaning and fluency.