Taiwan vs mainland Chinese technical terms: the numbers

How different are Taiwan vs mainland Chinese technical terms? We compared 6,126 term pairs from four official lists published by Taiwan's National Academy for Educational Research (NAER). In 51.4% of them the two sides use different words, even after both are written in the same character set; 37.3% differ only in Traditional versus Simplified characters; 11.3% are written the same. These lists were compiled to compare the two sides, so they may contain more differing terms than an average textbook. Read the shares as a description of these lists, not of every Chinese term.

How different are Taiwan vs mainland Chinese technical terms?

Each NAER list gives an English term, the term used in Taiwan and the term used in mainland China. We sorted every entry into one of three groups.

Computing has the most different words, which matches what students in Taiwan notice first: software, data and memory all have their own Taiwanese names. Medicine has the fewest. Physics sits in between, and many of its differences are the names of scientists: 都卜勒 and 多普勒 are both Doppler.

NAER's own research points the same way. A project it ran from 2010 to 2012 compared 24,841 academic term pairs and found 9,834 that were translated differently, about 39.6%. For school textbook terms it found 2,199 different out of 8,212, about 26.8% (NAER project page). That project used its own method, so its numbers and ours are not directly comparable.

What do the three groups mean?

• Written the same. Both sides use the same characters, for example 加速器 (accelerator).
• Characters only. The word is the same and only the script differs. 有效電阻 in Taiwan is 有效电阻 in mainland China (effective resistance). A Simplified to Traditional converter handles these well.
• Different words. The two sides chose different words. Taiwan says 軟體 for software; mainland China says 软件. Converting 软件 character by character gives 軟件, which is still not the Taiwan term.

The third group is the one that causes trouble. A student who learned 软件 from a Simplified word list can read 軟體 in a handout and not connect the two.

15 examples you will meet in class

All examples below come from the NAER lists. Where an entry lists several Taiwan forms, the table shows one of them and the note names the others. The lists are not perfect: the computing list gives 線, 穿線 and 引線 for "thread", while Taiwanese operating systems courses usually write 執行緒.

Computing (計算機)

The subject with the most different words: 66.1% of 1,532 entries.

Physics (物理學)

32.3% of 706 entries use different words. Names of scientists are a common source of differences.

Mechanical engineering (機械)

56.7% of 3,003 entries use different words, the largest list of the four.

Medicine (醫學)

The subject with the fewest different words: 23.3% of 885 entries.

Download the whole list:CSVAnki (TSV)

Why does this matter if you use translation tools?

Researchers who build language models for Taiwan say most models learn mainly from Simplified Chinese. The authors of Taiwan LLM, a model built for Taiwanese Mandarin, describe existing models as "predominantly trained on Simplified Chinese or English" (Lin and Chen, 2023). A 2025 study of 11 large language models found that "most LLMs disproportionately favored Simplified Chinese responses in the regional term choice task", where an item has different names in mainland China and Taiwan (Lyu et al., FAccT 2025).

In practice this shows up in three ways:

1. Translating English into Chinese. You may get mainland words written in Traditional characters, such as 軟件 or 信息, instead of 軟體 or 資訊. The text looks Taiwanese and the words are not.
2. Fixing it with a converter. OpenCC's phrase-aware converter, which swaps common mainland words for Taiwan ones, produced the Taiwan form for only 248 of the 3,150 entries that differ in wording (7.9%). It did better in computing (186 of 1,013, 18.4%) than in mechanical engineering (27 of 1,703, 1.6%) or medicine (3 of 206, 1.5%).
3. Studying from Simplified word lists. Many word lists and flashcard decks online follow mainland usage. If your course is in Taiwan, the exam uses the Taiwan term.

What helps:

• Take the term from your own handout, in the exact form your teacher writes it.
• Check unfamiliar terms in NAER's term database, which shows the Taiwan form next to the English.
• Keep one term list per course, so a word you met in week two is spelled the same in week ten. Our guide on studying in a foreign language shows how to build that list from lecture slides.
• If you use an AI tool on a PDF, read its output against the original page. Our answer to whether NotebookLM can translate a PDF explains what those tools return.

BanDoc builds that term list from your own handouts and keeps the original Chinese one tap away, so you review the word that will be on your exam. You can try it with one handout.

How we counted

1. We downloaded four NAER cross-strait term lists (computing, mechanical engineering, physics, medicine) on 2026-10-20.
2. Many entries give several renderings separated by semicolons or commas. We split them, removed part-of-speech tags such as 【动】, and treated text in brackets as optional, so 非晶[態固]體 matches both 非晶態固體 and 非晶體.
3. An entry is written the same if any Taiwan rendering equals any mainland rendering.
4. Otherwise it is characters only if they match after both are converted to Simplified characters with OpenCC's Taiwan to Simplified table (opencc-js 1.4.2, tw to cn). This step changes characters, never words.
5. Everything else is different words. For those entries we also converted the mainland form with OpenCC's phrase table (cn to twp) and counted how often the result equals a Taiwan rendering.

Limits to keep in mind:

• The lists compare the two sides, so they may over-represent terms that differ.
• Automatic character conversion can be wrong for a few characters.
• One matching rendering is enough to count as the same, which undercounts differences when Taiwan lists several forms.

The analysis script and the per-entry results are not published yet. If you want the classified CSV for research or teaching, write to support@bandocstudy.com.

Sources

• NAER cross-strait term lists (CSV): computing, mechanical engineering, physics, medicine.
• NAER, 兩岸學術名詞暨常用專業名詞翻譯之分析比較初探 (2010 to 2012).
• NAER term database: terms.naer.edu.tw.
• Lyu, Luo, Kang and Koenecke, Characterizing Bias: Benchmarking Large Language Models in Simplified versus Traditional Chinese, FAccT 2025.
• Lin and Chen, Taiwan LLM: Bridging the Linguistic Divide with a Culturally Aligned Language Model, 2023.
• OpenCC conversion tables, through opencc-js.

Data licence and attribution

The four term lists are published under Taiwan's Open Government Data License, version 1. The licence requires this notice for each dataset:

• 國家教育研究院 2026 國家教育研究院-兩岸對照名詞-計算機學術名詞。此開放資料依政府資料開放授權條款 (Open Government Data License) 進行公眾釋出，使用者於遵守本條款各項規定之前提下，得利用之。政府資料開放授權條款：https://data.gov.tw/license
• 國家教育研究院 2026 國家教育研究院-兩岸對照名詞-機械學術名詞。此開放資料依政府資料開放授權條款 (Open Government Data License) 進行公眾釋出，使用者於遵守本條款各項規定之前提下，得利用之。政府資料開放授權條款：https://data.gov.tw/license
• 國家教育研究院 2026 國家教育研究院-兩岸對照名詞-物理學學術名詞。此開放資料依政府資料開放授權條款 (Open Government Data License) 進行公眾釋出，使用者於遵守本條款各項規定之前提下，得利用之。政府資料開放授權條款：https://data.gov.tw/license
• 國家教育研究院 2026 國家教育研究院-兩岸對照名詞-醫學學術名詞。此開放資料依政府資料開放授權條款 (Open Government Data License) 進行公眾釋出，使用者於遵守本條款各項規定之前提下，得利用之。政府資料開放授權條款：https://data.gov.tw/license

The counts, percentages, pinyin and notes on this page are our own analysis, not official NAER figures.