Title: Chinese Word Segmentation Applied to Written Cantonese

Abstract: Chinese text is written without spaces to separate words,
and individual characters can used be used both as a whole word and as
a part of a compound word: a word of two or more characters, e.g.  
(learn),  Éú (student),    (mathematics) all contain a common
character.  This leads to ambiguous sentences which could be read
differently depending on the word boundaries, or word segmentation,
inferred by the reader.  This causes no problem for human readers who
understand the meaning of text they read: most incorrect segmentations
would not make sense.  However, as a computer cannot understand the
meaning of a sentence, it just executes an algorithm, which makes this
a challenging problem to automate.

Chinese word segmentation (CWS) is incorporated into other text
processing tasks, such as information retrieval; text-to-speech and
machine translation, and has been the subject of several competitions.
The corpora used for these competitions are all standard written
Chinese (SWC) and mostly simplified characters.  However, SWC uses
different words and word order to spoken Cantonese and there is
another written form of Cantonese, which is used widely in informal
writing and when the exact wording of the speaker is important.  There
are very few applications of CWS to written Cantonese in the
literature.

In this talk I will introduce the problem of CWS, including some of
the complications of Chinese text in general and problems specific to
written Cantonese.  I will then describe some CWS algorithms and the
application of them to a written Cantonese corpus.