早安 is made up of 2 characters but is a single word. If you fall into the trap of thinking 1 character = 1 word, you won't understand a thing. In this case you'd have thought it meant "early safe" instead of "good morning".
> Same with Chinese language, thus lexing and parsing requires knowing many more words than in languages with spaces between words.
In English, can you get away with knowing the meaning of "good" and "morning" and not "good morning", and know that I'm greeting you instead of commenting on the quality of this morning?
Where I've had pain specifically with Thai is that I can't even know where a syllable begins and ends until I read a few "syllables" together and decide whether some vowels go with the consonant in front or behind it, and whether some an -ar should be pronounced as an -aan.
As mentioned in another comment, single syllable words are much more common in Cantonese, and word combinations are much more "free" in the sense that there are a lot more ambiguity as to what counts as a "word" and what is merely two single-character-words idiomatically used together. There are also cases where grammatical constructs (and also foul words) are inserted in between a two-character word/idiomatic combo, and sometimes the characters are reversed, to the extent that it used to be a meme: https://evchk.fandom.com/zh/wiki/Y%E5%B7%B2x
It's gotten to a point where, after thinking about it for a couple years, I've come to believe that segmentation on Cantonese is a fool's errand...
Of course, there's also classical Chinese where most of the time a character is a word.
The problem with proper nouns is that they don't end up in dictionaries, same with slang and other terms that for reasons don't end up in dictionaries.
The additional problem with Cantonese is that there's a larger class of words where the constituent characters can move around as if they were words themselves. Even for a native speaker with some experience in lexicography, it can be difficult to determine word boundaries as there are many cases where a word with characters X+Y can be interpreted as just word X and word Y with some idiomatic meaning. This issue is more pronounced in Cantonese because there are more single character words in active use.
I've actually done this before. My experience is that naive segmentation on Mandarin text with wordlist is probably 80+% accurate, while using the same algorithm in Cantonese text (with cantonese wordlist) will definitely end up "wtf".