I have been working on word segmentation and part-of-speech tagging for Hongkongese. Then GPT-4 came out, and I heard its Hongkongese/Cantonese skills has vastly improved, so I subscribed and tested out how well it does. Set Up GPT-4 understand what is Universal Dependencies. It was easy to get it to output in…

In my previous story, I evaluated the performance of several NLP systems against Hong Kong data. One of the top performers is the CKIP Transformers. It is an interesting system, because it was built from fine-tuning a large language model (LLM) for the word segmentation task. The CKIP models are…

In natural language processing(NLP), often the first step is to tokenize a string, then the second step is to annotate the tokens with part-of-speech(POS) tags. There are many Chinese word segmenters and POS taggers, but almost none for Cantonese. The strategy is usually to load a custom dictionary into a…

Continuing to provide archive links to articles pertaining to case DCCC265/2022. Article list collected from the court news site The Witness. 《立場》被指煽動案｜控方欲呈587篇文章分類表 稱再有6篇涉煽動 作者包括林彥邦、區家麟 - 法庭線 The Witness 控方代表為署理高級助理刑事檢控專員伍淑娟、署理高級檢控官徐倩姿；辯方代表為資深大律師余若薇、大律師馬維騉及黃卉儀，案件由法官郭偉健審理。thewitnesshk.com 2020.7.1 林彥邦 速讀《港區國安法》 速讀《港區國安法》 | 林彥邦 | 立場新聞 國安法來了， 還是讀讀條文吧 。 一、治外法權 先講最關心的，「駐港國安公署」，簡單講，公署人員有治外法權。 條例列明，駐港公署人員執行職務時，「不受香港政府管轄」；而且他們「不受」香港執法人員扣押、截查。…web.archive.org

At Toasty News, we obviously care deeply about news. Currently, there is a case against Stand News, the former editor-in-chief Chung Pui-kuen and the former acting chief edit …

At Toasty News, we collect and publish statistics about Hongkongese usage on independent news sites from Hong Kong. For previous reports, check 2018 and 2019. Using machine learning, we characterize each non-duplicate article we found into the following languages: Hongkongese — also know as Hong Kong Cantonese or Yue. This…

When I was younger, I loved reading the dictionary. It was fun to read random entries and imagine the situations when the words would be used. I didn’t have Internet then, the dictionary was my window to the world without ever stepping outside. Not quite sure why, my curiosity waned…

I got access to TensorFlow Research Cloud for a month, so I spent most of that time training transformer models with Hongkongese data. This post summarize the models and other NLP resources I created so far. ELECTRA and XLNet Models ELECTRA is the transformer model trained with a generator/discriminator pair. It is very efficient…

There are two character sets for writing Chinese characters in the world, Simplified Chinese (SC) and Traditional Chinese (TC). TC is mostly used in Taiwan and Hong Kong, and SC is is mostly used in China. As the name implies, SC has simplified forms, and multiple characters in TC could…