In this work, we analyze shortcomings in cross-lingual knowledge transfer in large, modern reasoning LLMs. We demonstrate that the perceived gap in knowledge transfer is primarily a script barrier. First, we conduct an observational data analysis on the performance of thinking models on two datasets with local knowledge from around the world, ECLeKTic and MultiLoKo. Our regression analysis shows that script match - not language or family - is the primary predictor of knowledge transfer failure once model capability and question difficulty are accounted for. We further this finding by providing the LLMs with the key entities of the questions in their source language and find that this disproportionately improves cross-script questions. We then posit that these LLMs could be reasoning better at test-time. To evaluate this, we develop a synthetic generation pipeline to design SFT samples to encourage the model to better reason about transliteration ambiguities when trying to fetch parametric knowledge at inference-time. We show that teaching two models to reason better reduces the cross-script transfer gap. As a result, we conclude that there is potential to improve cross-lingual parametric knowledge transfer during post-training.
Figures & tables
Variable
ECleKTic
MultiLoKo
OLS Intercept
− 0.350
− 0.439
Belebele Score
0.999
0.937
Script Match
0.121
0.133
Language Family
n.s.
n.s.
In-Language
n.s.
n.s.
Model Indicators (vs. Qwen3-30B)
Table 1: OLS Coefficients ( β ) when trying to predict the binary score of a question, with no scaling . n.s. denotes statistically insignificant ( β ); p>0.05 .
Condition
All
Non- Latin
Latin
Script =
Script =
GPT-OSS-120B
Original
20.6
16.3
23.6
18.5
24.6
Prefix, Control
21.2
17.9
23.5
19.3
24.8
Prefix w/ src Ent.
22.2
19.4
24.2
20.7
25.1
Δ
+1.0
+1.5
+0.7
+1.4
+0.2
Qwen3-235B-A22B
Table 2: Impact of our think-prefixing on ECLeKTic accuracy ↑ . Script = denotes cross-script; Script = denotes same-script.
ECLeKTic (12 languages)
MultiLoKo (31 languages)
Latin
Resource
Script
Latin
Resource
Script
Model
Run
All
No
Latin
Low
High
=
=
All
No
Latin
Low
High
=
=
GPT-OSS-120B
Base
20.6
16.3
23.6
12.6
23.9
18.5
24.6
24.4
18.0
31.2
17.0
27.8
21.6
27.3
LoRA
21.6
18.6
23.8
15.3
24.0
20.1
24.7
25.0
18.7
32.0
17.7
28.3
22.3
27.4
Qwen3-30B-A3B
Base
18.1
15.1
20.1
12.2
20.0
16.9
20.3
25.6
21.4
30.1
19.6
27.1
21.1
28.0
LoRA
18.9
16.3
20.8
12.8
18.2
18.0
20.6
26.0
21.9
30.4
20.2
27.4
21.7
28.2
Table 3: Comprehensive evaluation of Base vs. SFT across two datasets. We compare performance across Latinity (Non-Latin vs. Latin), Resource levels (Low vs. High), and Script matching ( = vs. = ).
Appendix figures & tables2 assets
Supplementary material from the paper’s appendix.
Appendix
Variable
ECleKTic
MultiLoKo
OLS Intercept
0.452
0.569
Belebele Score
0.088
0.100
Script Match
0.056
0.064
Language Family
n.s.
n.s.
In-Language
n.s.
n.s.
Model Indicators (vs. Qwen3-30B)
Appendix
Table 4: OLS Coefficients ( β ) when trying to predict the binary score of a question, after features are standard-scaled . n.s. denotes statistically insignificant coefficients; p>0.05 . The values represent the expected increase in probability from an increase in one standard deviation of the feature. It allows for comparison between continuous features (e.g. Belebele score) and the rest. The value is essentially a metric for how much variance the feature encapsulates.
Hyperparam.
GPT-OSS-120B
Qwen3-30B-A3B
Num. Samples
599
387
Target Modules
All Attn
All Attn
LoRA Rank ( r )
32
32
LoRA Dropout
0
0.05
Learning Rate
2×10−5
6×10−5
Epochs
8
8
Appendix
Table 5: Hyperparameters for the LoRA training runs for which results are reported in Table 3 .