---
title: "LXPER Index 2.0 — paper by Lee and Lee (2020)"
description: "This paper presents LXPER Index 2.0, an updated version of the authors' earlier model. 8 references, cited by 4 papers here."
authors: ["Bruce W. Lee", "Jason Lee"]
year: 2020
venue: "Proceedings of the 6th Workshop on Natural Language Processing Techniques for Educational Applications"
source: "https://aclanthology.org/2020.nlptea-1.3/"
doi: "10.18653/v1/2020.nlptea-1.3"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/lee-2020-lxper-index-2-0-improving/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/lee-2020-lxper-index-2-0-improving/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2020](https://md.research.elamur.ai/free/en/papers/2020/page/8/) / LXPER Index 2.0: Improving Text Readability Assessment Model for L2 English Students in Korea

# LXPER Index 2.0: Improving Text Readability Assessment Model for L2 English Students in Korea

**Authors:** [Bruce W. Lee](https://md.research.elamur.ai/free/en/authors/bruce-w-lee/) and [Jason Lee](https://md.research.elamur.ai/free/en/authors/jason-lee/) · **Year:** 2020 · **Venue:** [Proceedings of the 6th Workshop on Natural Language Processing Techniques for Educational Applications](https://md.research.elamur.ai/free/en/venues/nlptea/) ([Workshop on Natural Language Processing Techniques for Educational Applications 2020](https://md.research.elamur.ai/free/en/venues/nlptea/2020/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2020.nlptea-1.3/) · [PDF](https://aclanthology.org/2020.nlptea-1.3.pdf) · [DOI 10.18653/v1/2020.nlptea-1.3](https://doi.org/10.18653/v1/2020.nlptea-1.3) · [OpenAlex](https://openalex.org/W3120899979)

**Citations:** 7 citations (OpenAlex, 2026-10-14) · [4](https://md.research.elamur.ai/free/en/papers/lee-2020-lxper-index-2-0-improving/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
Standard readability formulas such as Flesch-Kincaid predict grades poorly for texts written for Korean learners of English. The reason is that L2 students acquire English almost entirely through a fixed national ELT curriculum, while native readers absorb it from many sources. This paper presents LXPER Index 2.0, an updated version of the authors' earlier model.

To fix the original CoKEC-text's heavy skew toward high-school grades, the authors added roughly 900 middle-school textbook texts and hand-cleaned passages that mixed in Korean characters. They tested 35 features in five families: traditional, POS-based, entity density, lexical chain, and curriculum word difficulty from CoKEC-word. Features were kept or dropped by Pearson correlation and redundancy, and a logistic regression model replaced the earlier linear one.

On a held-out set spanning K7–K12, the model's predictions rise steadily with grade, and its average error is 0.34 grade levels, against 1.05 for LXPER 1.0 and 2.10 for Flesch-Kincaid. The work shows that a carefully labeled, curriculum-specific corpus of fewer than 4000 texts can support accurate L2 readability assessment.

## Key points
- **Training corpus:** The expanded CoKEC-text has 3700 Korean MOE-sanctioned texts from K7 to K12.5, where K12.5 is the CSAT college entrance exam. The original version had 2760 texts, with only 17 at K7 and 80 at K9.
- **Test set:** 540 texts, 90 per grade. K10–K12 come from KICE mock exams (2017–2020) and K7–K9 from government-approved middle-school textbooks.
- **Average error (in grade levels):** LX 2.0 scored 0.34, LX 1.0 scored 1.05, Flesch-Kincaid 2.10 and Dale-Chall 3.04. Lexile and Coh-Metrix L2 could not be compared meaningfully because their scales differ and they showed no consistent trend across grades.
- **Feature findings:** The strongest single feature was the count of level-D (K9–11) words from CoKEC-word, with r = 0.532. CoKEC-word classifies 30,608 curriculum words into 6 levels. Lexical chain features correlated only weakly with grade.
- **Feature selection and tools:** Features were kept if Pearson correlation exceeded 0.07 and dropped if highly correlated with another feature. Parsing and counting combined spaCy (sentence splitting), NLTK, Gensim (lexical chains) and the Berkeley Neural Parser.

## Interesting ideas
- Measuring vocabulary difficulty with word lists tied to the learner's actual curriculum outperformed generic syllable or length measures, so "difficulty" is best defined relative to what the target reader has been taught.
- Treating grade levels as classes with logistic regression gave much better results than fitting a straight line, since difficulty may not grow linearly with grade.
- The test set deliberately mixed two text sources and judged success partly by whether predictions climb steadily from grade to grade, not only by average error.
- Stray foreign-script fragments, such as Korean glosses inside English texts, counted as tokens or even sentences and quietly distorted per-sentence averages, so the authors removed them by hand.
- Features were screened with a simple significance threshold, and then one member of each strongly correlated pair was dropped, keeping the model small and interpretable.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (answer): School-course readers could tag each word with the curriculum level at which it is introduced, which would explain why a passage is hard for a given grade.
- **I2** (learn): When predicting ordinal labels such as levels, ranks or confidence tiers, try classification against regression rather than assuming linearity.
- **I3** (learn): Evaluations could add source-diverse splits and monotonicity checks as sanity controls, catching models that score well on average but rank inputs inconsistently.
- **I4** (tokens): The tokenizer should detect and isolate non-English spans before counting or parsing, so per-sentence statistics and UD trees are not polluted.
- **I5** (learn): A cheap filter like this could prune candidate features or repair-rule triggers, keeping only those that are informative and non-redundant, which supports explainability.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

📚 Shelves (6): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [💭 Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/) · 1,552 works · [💭 Developmental and Educational Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/developmental-and-educational-psychology/) · 499 works

🧭 Topics (1): [English](https://md.research.elamur.ai/free/en/topics/english/) · 2,263 works

## Cite

Cite as: Bruce W. Lee and Jason Lee. 2020. LXPER Index 2.0: Improving Text Readability Assessment Model for L2 English Students in Korea. Proceedings of the 6th Workshop on Natural Language Processing Techniques for Educational Applications.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{lee2020lxper,
  title = {{LXPER Index 2.0: Improving Text Readability Assessment Model for L2 English Students in Korea}},
  author = {Bruce W. Lee and Jason Lee},
  booktitle = {Proceedings of the 6th Workshop on Natural Language Processing Techniques for Educational Applications},
  year = {2020},
  doi = {10.18653/v1/2020.nlptea-1.3},
  url = {https://doi.org/10.18653/v1/2020.nlptea-1.3}
}
```

- [← previous Is it Great or Terrible? Preserving Sentiment in Neural Machine Translation of Arabic Reviews](https://md.research.elamur.ai/free/en/papers/saadany-2020-is-it-great-or-terrible/)
- [next → Neural Mention Detection](https://md.research.elamur.ai/free/en/papers/yu-2020-neural-mention-detection/)
- [↑ back to papers of 2020](https://md.research.elamur.ai/free/en/papers/2020/page/8/)

## References

8 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Rebekah G. Benjamin. 2012. Reconstructing readability: Recent developments and recommendations in the analysis of text difficulty. Educational Psychology Review, 24(21). [↗](https://doi.org/10.1007/s10648-011-9181-8)
2. Menglin Xia, Ekaterina Kochmar, and Ted Briscoe. 2016. [Text readability assessment for second language learners](https://md.research.elamur.ai/free/en/papers/xia-2016-text-readability-assessment-second-language/). In Proceedings of the 11th Workshop on Innovative Use of NLP for Building Educational Applications, pages 12–22.
3. Steven Bird, Edward Loper, and Ewan Klein. 2009. Natural Language Processing with Python. O’Reilly Media Inc. [find](https://scholar.google.com/scholar?q=Steven+Bird%2C+Edward+Loper%2C+and+Ewan+Klein%2E+2009%2E+Natural+Language+Processing+with+Python%2E+O%E2%80%99Reilly+Media+Inc%2E)
4. Radim Řehůřek and Petr Sojka. 2010. Software framework for topic modelling with large corpora. In Proceedings of LREC Workshop on New Challenges for NLP Frameworks, pages 45–50. [find](https://scholar.google.com/scholar?q=Radim+R%CC%8Cehu%CC%8Ar%CC%8Cek+and+Petr+Sojka%2E+2010%2E+Software+framework+for+topic+modelling+with+large+corpora%2E+In+Proceedings+of+LREC+Workshop+on+New+Challenges+for+NLP+Frameworks%2C+pages+45%E2%80%9350%2E)
5. Scott A. Crossley, Jerry Greenfield, and Danielle S. Mc-Namara. 2008. Assessing text readability using cognitively based indices. TESOL Quarterly, 42(3). [↗](https://doi.org/10.1002/j.1545-7249.2008.tb00142.x)
6. Edgar Dale and Jeanne S. Chall. 1949. The concept of readability. Elementary English, 26(23). [find](https://scholar.google.com/scholar?q=Edgar+Dale+and+Jeanne+S%2E+Chall%2E+1949%2E+The+concept+of+readability%2E+Elementary+English%2C+26%2823%29%2E)
7. Lijun Feng, Martin Jansche, Matt Huenerfauth, and Noemie Elhadad. 2010. [A comparison of features for automatic readability assessment](https://md.research.elamur.ai/free/en/papers/feng-2010-comparison-features-automatic-readability-assessment/). In Proceedings of the 23rd International Conference on Computational Linguistics, pages 276–284.
8. Michael Heilman, Kevyn Collins-Thompson, Jamie Callan, and Maxine Eskenazi. 2007. Combining lexical and grammatical features to improve readability measures for first and second language text. In [find](https://scholar.google.com/scholar?q=Michael+Heilman%2C+Kevyn+Collins%2DThompson%2C+Jamie+Callan%2C+and+Maxine+Eskenazi%2E+2007%2E+Combining+lexical+and+grammatical+features+to+improve+readability+measures+for+first+and+second+language+text%2E+In)

*Updated 1 November 2026*