Why GenAI for Chinese Studies?

Kwok-leong Tang

Who Am I?

  • Kwok-leong Tang 鄧國亮
  • Historian by training (Late Imperial China)
  • Lecturer, EALC
  • Managing Director, Digital China Initiative

Schedules

Tuesday 11 February 2025

  • Why GenAI for Chinese studies?
  • Use Cases
  • Google NotebookLM

Tuesday 18 February 2025

  • How to use GenAI to build a data collection
  • How to use API to process data

Questions

Please enter your responses to the questions below in the chatbox.

  • How often do you use chatbots? (e.g., every day, a few times a week, never)
  • Which GenAI tools have you used? (e.g., ChatGPT, Deepseek, Copilot)
  • What are the main tasks you assign to GenAI tools? (e.g., translation, writing improvement, text summarization)

Why GenAI for Chinese Stduies?

Since its release on November 30, 2022, ChatGPT (and other Large Language Models, LLMs) has brought generative AI into the spotlight of our daily lives. What is the significance of generative AI for us, as grduate students of Chinese Studies?

Communicating with Machines in Natural Languages

  • Large language models (LLMs) and machine learning were not new concepts before November 2022.
  • The revolution brought by ChatGPT and generative AI, in general, lies in the interaction experience between humans and machines.
  • We can now communicate with machines in natural languages, whether in English, Vietnamese, Mandarin, Japanese, or Korean.

GenAI and students in Chinese Studies

  • It simplifies technological complexities, making it easier to adopt digital tools in research processes. This means less coding and less training (for both humans and machines).
  • In digital research workflows, domain expertise has become more significant due to the necessity of evaluation.

Please Remember

  • You will get the most when you can evaluate the results!
  • Use GenAI to support your research process but not your conclusion!

Magic 101

  • Magic Power: Large Language Models
  • Magic Wand/Staff: Chatbot Interfaces
  • Spell/Chanting: Crafting Inspiring Prompts

Use Cases (Spells)

Citation Styles

Formatting citations and references to align with the style guidelines of publishers is a task that often consumes a significant amount of our time. In this particular scenario, we will be tailoring our citations to adhere to the style guide of JAS.

Journal of Asian Studies Style Guide: https://dukeupress.edu/Assets/Downloads/JAS_style_guide.pdf

Citation Styles

Please help me to prepare a list of references. The general rules are based on *The Chicago Manual of Style, 17th ed.*. For references to works published in Chinese, Japanese, Korean, or any other
Asian script, please provide titles in the following format: Transcription <original characters> [English translation]. Translations of journal or periodical names are not required. Here are two examples

Huang Shang 黃裳. 1947. *Guanyu Meiguo bing* 關於美國兵 [About American soldiers]. Shanghai: Shanghai chuban gongsi.
Peng Su 彭蘇. 2006. “Beida ‘fanMei’ nüsheng Ma Nan de zhenshi jinkuang 北大“反美”女生馬楠的真實近況 [Updates on Ma Nan, “anti-American” female student from
Peking University]. *Nanfang renwu zhoukan* 南方人物週刊 9:38–39.

Please confirm your understanding of the task. Once confirmed, I will proceed to provide you with the references for formatting.

Citation Styles

Once your chatbot confirm that they understand the task, you can paste the following texts.

石崎又造：《近世における支那俗語文学史》（東京：清水弘文堂書房，1967年） 
林俊宏：《朱舜水在日本的活動及其貢獻研究》（臺北：秀威資訊科技股份有限公司，2004年） 
湯沢質幸：《近世儒学韻学と唐音 —訓読の中の唐音直読の軌跡》（東京：勉誠出版，2014年） 
劉元卿：《劉聘君全集》，《四庫全書存目叢書》（濟南：齊魯書社，1997年）影印清咸豐二年（1852）重刻本 
余英時：《朱熹的歷史世界》（北京：生活 • 讀書 • 新知三聯書店，2011年）

What happened?

  • Few-shot prompting: Provide several examples and instruct the AI to generate responses following the same style.
  • Application to other use cases: Leverage this approach in similar scenarios.

Parallel Reading

The two paragraphs below are the Chinese and English versions of the same text. Please compare them sentence by sentence in a table, with each row corresponding to one sentence.

Paragraph A
觀自在菩薩。行深般若波羅蜜多時。照見五蘊皆空。度一切苦厄。舍利子。色不異空。空不異色。色即是空。空即是色。
受想行識。亦復如是。舍利子。是諸法空相。不生不滅。不垢不淨。不增不減。是故空中無色。無受想行識。無眼耳鼻舌身意。無色聲香味觸法。無眼界。乃至無意識界。無無明。亦無無明盡。乃至無老死。亦無老死盡。無苦集滅道。無智亦無得。以無所得故。菩提薩埵。依般若波羅蜜多故。心無罣礙。無罣礙故。無有恐怖。遠離顛倒夢想。究竟涅槃。三世諸佛。依般若波羅蜜多故。得阿耨多羅三藐三菩提。故知般若波羅蜜多。是大神咒。是大明咒。是無上咒。是無等等咒。能除一切苦。真實不虛。故說般若波羅蜜多咒。即說咒曰。揭諦揭諦　波羅揭諦　波羅僧揭諦 菩提薩婆訶。

Paragraph B
Om Homage to the Perfection of Wisdom the Lovely, the Holy! Avalokita, the Holy Lord and Bodhisattva, was moving in the deep course of the Wisdom which has gone beyond. He looked down from on high, He beheld but five heaps, and He saw that in their own-being they were empty. Here, O Sariputra, form is emptiness and the very emptiness is form; emptiness does not differ from form, form does not differ from emptiness, whatever is emptiness, that is form, the same is true of feelings, perceptions, impulses, and consciousness. Here, O Sariputra, all dharmas are marked with emptiness; they are not produced or stopped, not defiled or immaculate, not deficient or complete. Therefore, O Sariputra, in emptiness there is no form nor feeling, nor perception, nor impulse, nor consciousness; No eye, ear, nose, tongue, body, mind ; No forms, sounds, smells, tastes, touchables or objects of mind ; No sight-organ element, and so forth, until we come to: No mind-consciousness element ; There is no ignorance, no extinction of ignorance, and so forth, until we come to : There is no decay and death, no extinction of decay and death. There is no suffering, no origination, no stopping, no path. There is no cognition, no attainment and no non-attainment. Therefore, O Sariputra, it is because of his non-attainmentness that a Bodhisattva, through having relied on the Perfection of Wisdom, dwells without thought-coverings. In the absence of thought-coverings he has not been made to tremble, he has overcome what can upset, and in the end he attains to Nirvana. All those who appear as Buddhas in the three periods of time fully awake to the utmost, right and perfect Enlightenment because they have relied on the Perfection of Wisdom. Therefore one should know the prajnaparamita as the great spell, the spell of great knowledge, the utmost spell, the unequalled spell, allayer of all suffering, in truth -- for what could go wrong? By the prajnaparamita has this spell been delivered. It runs like thi: gate gate paragate parasamgate bodhi svaha. This completes the Heart of perfect Wisdom.

Text Comparison

Instruction:
Compare the two classical Chinese texts. Punctuate both texts. Output a table that includes the following information for each sentence:
- The punctuated sentence in Text A
- The corresponding punctuated sentence in Text B
- Any differences between the two sentences
- An analysis of those differences

Output a table in this format:
The sentence in Text A |The sentence in Text B|Difference|Analysis

Input:
    - Text A: 古之王者世有史官君舉必書所以慎言行昭法式也左史記言右史記事事為春秋言為尚書帝王靡不同之周室既微載籍殘缺仲尼思存前聖之業乃稱曰夏禮吾能言之杞不足徵也殷禮吾能言之宋不足徵也文獻不足故也足則吾能徵之矣以魯周公之國禮文備物史官有法故與左丘明觀其史記據行事仍人道因興以立功就敗以成罰假日月以定曆數藉朝聘以正禮樂有所褒諱貶損不可書見口授弟子弟子退而異言丘明恐弟子各安其意以失其真故論本事而作傳明夫子不以空言說經也 
    - Text B:古之王者世有史官君舉必書所以慎言行昭法式也左史記言右史記事其事為春秋言為尚書帝王靡不同之周室既微載籍殘缺仲尼思存前聖之業以魯周公之國禮文備物史官有法故與左丘明觀其史記據行事仍人道因興以立功就敗以成罰假日月以定歷數藉朝聘以正禮樂有所褒諱貶損不可書見口授弟子弟子退而異言丘明恐失其真故論本意而作傳明夫子不以空言也

Data Extraction (Named Entity Recognition)

You will extract information from deeds. You will be given a few examples, then you should follow the same output format to extract the information from other deeds. The output is in form with five columns:
Title (標題): the title of the deed;
People/Organizations (人名/團體): People and organizations, and their roles, mentioned in the deed. 
Contractor (立契者)
Date of Contract (立契時間)

Here are the examples:
Deed A:
立承佃鄭復禮，今在李德榮處承出根靣全民田壹號，坐產洋下洋，圡名大頭坵，現丈弍畝伍分弍厘捌毛正，承來耕作，靣约遞年認纳乾凈上籠租谷壹仟觔正，早六晚四送還理清，不得包箂，不得少欠，如有少欠，係保代賠，其田即听李家召佃別耕，且鄭家復禮不得阻留之理。今欲有慿，立承佃字壹帋為照。
光緒叁拾年弍月初七日立承佃字鄭復禮
保佃：胞兄鄭復奎
信字
Result:
| 標題 | 人名/團體 | 地點 | 立契者 | 立契時間 |
| -- | ----- | -- | --- | ---- |
| 立承佃鄭復禮 | 鄭復禮(立據人);李德榮(給墾人);鄭復奎(保人) | 洋下洋;大頭坵 | 鄭復禮 | 光緒三十年二月七日 |

Deed B:
立承佃高仁旋，今在嶺頂村李德荣處承出民田一號，坐產洋下洋，圡名三夷，經丈叁畝弍分零，面約每年納出上籠乾淨租谷壹仟壹百觔止，早六晚四理还清椘，不敢少欠，如有少欠，係保代賠，恐口無慿，立承佃為據。
光緒叁拾叁年弍月初三日立承佃高仁旋
保佃：胞姪高居鏘
信字
Result:
| 標題 | 人名/團體 | 地點 | 立契者 | 立契時間 |
| -- | ----- | -- | --- | ---- |
| 光緒三十三年二月三日立承佃高仁旋 | 高仁旋(立據人);李德荣(給墾人);高居鏘(保人) | 嶺頂村;洋下洋;三夷 | 高仁旋 | 光緒三十三年二月三日 |

You do not generate deeds yourself. Once you understand the task, you can tell me to give you the deeds for extraction.

Testing

Deed A

立據給出山林埔地契字人業主金協福，即胡協成，即胡桂森兄弟，有承祖父當官，向金雞翁四柱耆番明給出暖暖山後什份藔等處山林埔地壹所。其東西南北四至界址，俱各登載在大墾內明白。今因乏銀公費，願將墾內抽出山林埔地壹所，小土名石碇仔，東至溪為堺、西至楠仔根大崙分水、南至大崙分水與王明月觀，毗連為堺、北至大崙分水與胡寔觀毗連透白與石下有應公崙明踏為堺；又連過溪壹小段，東至大崙水流落為堺、西至溪為堺、南至與王家毗連為堺、北至石鼻崙水流落為堺，四至堺址俱各踏明。要出賣於人，先問房親人等，不欲承受，外托中引就與胡金錠兄弟身上出首承買，時仝中言議值價銀柒拾大員正，銀即日仝中交收足訖。其山林埔地隨即踏明堺址，俱付銀主，前去蓋屋居住、栽種竹木，以及招佃耕作、開闢水田、收租納課，併帶大小坑水，通流灌溉，上流下接，各不阻當。年配納番口糧銀壹員五角正，任從開築坡、圳路，墾成水田，永為己業，不敢阻當滋事。若是開墾成田，依例按甲配納金協福大租，不得異言。保此業係是森承祖父當官向番明給之業，與別人無干，亦無重張典掛他人財物，以及來歷交加不明為礙。如有不明，森兄弟一力出首抵當，不干銀主之事，此係二比甘願，各無反悔。口恐無憑，今欲有憑。立據給出山林埔地契字壹紙，付執為炤。
即日仝中收過字內佛面銀柒拾大員正。再炤。
代筆人：胡典金。
為中人：胡世千。
在場人：胡文煨、胡德寶。
知見人：胡桂瀾、胡桂路。
同治六年正月　日立撥給出契字人金協福，即胡協成，即胡桂森、胡森雨。

Deed B

立給開墾田契字遷善北社通事遠山眉有承祖父遺下埔地壹所貫在頂沙轆寮庄東至消水溝為界西至路為界南至童林田為界北至港仔溝為
界四至界址明自今因乏銀費用無力自耕外託中引就向與寓鰲頭街蔡太和堂前去自備工本開墾成田時仝中面議備出給墾契銀貳拾肆大員正銀即日仝中交收足
訖其埔地隨即踏明界址交付太和堂前去掌管開築成業逐年配納番業主大租谷陸斗正滿每年至六早季付業主收取給單存照其埔地自此一給千休刈籐永斷
永為己業眉子孫日後不敢言及找貼情弊保此埔地係是眉承祖父遺下物業與社內族番親等無干亦無重張典掛他人併無來歷交加不明
為碍如有不明等倩眉出首一力抵當不干太和堂之事此係二比甘愿各無反悔恐口無憑今欲有憑仝立給開墾荒埔田契字壹紙付執為炤
即日仝中親收過開墾荒埔田契字內佛銀貳拾肆大員正完足再炤
再批明內添口字壹字再炤
知見併為中（印記一）
光緒拾壹年貳月　　　　日立給開墾荒埔田契字遷善北社通事遠山眉自筆（印記二）

What happened?

  • Few-shot prompting
  • GenAI LLMs make the NER more user-friendly

Retreival Augmented Generation (RAG)

flowchart TB
    subgraph Direct["Direct LLM Query"]
        direction LR
        U1[User Query] --> L1[LLM] --> R1[Response]
        style L1 fill:#f9f,stroke:#333
    end


flowchart TB
    subgraph RAG["RAG-Enhanced Query"]
        direction LR
        U2[User Query] --> RC[Retrieval Component]
        RC --> |Search| DB[(Knowledge Base)]
        DB --> |Relevant Documents| RC
        RC --> CP[Context Processor]
        CP --> |Enhanced Prompt| L2[LLM]
        L2 --> R2[Response]
        style L2 fill:#f9f,stroke:#333
        style DB fill:#bef,stroke:#333
        style RC fill:#fbf,stroke:#333
        style CP fill:#fbf,stroke:#333
    end

NotebookLM

  • NotebookLM is a GenAI tool developed by Google.
  • For free tier:
    • 30 notebooks
    • each notebook with 30-50 sources
  • You can access NotebookLM at https://notebooklm.google.com