Well, a good example is "ice cream" - I've seen it described as a "compound word without a hyphen" - which is fair because while ice cream has ice and cream as things that are used to make it, "ice cream" is definitely something distinct that is not trivially decomposable from the individual words.
A good definition of "a unit of semantic meaning" might be "fulfills a 'part of speech' in a sentence" - but even that is hard in English. For example: "I set the table up" - up modifies set to create the "phrasal verb" set up, and isn't related to its typical meaning of position. But remove that little word and the whole meaning of the sentence drastically changes. I've seen "up" in this instance described as an adverb, which tends to be the "throwaway stat" for words that don't fit any of the other traditional parts of speech.
Another I like is contractions. If say “can not” but it sounds like “can’t”, how many words is that?
I might say that “not” does not have any semantic meaning in isolation, as its only use is to negate the meaning of some other word, so perhaps “can not” and “can’t” are both just one word.
Perhaps such graphemes and phonemes aren’t actually words.
Another interesting question is whether words are truly a product of language (the natural skill) or writing (the synthetic technology). Did “words” exist before writing?
A good definition of "a unit of semantic meaning" might be "fulfills a 'part of speech' in a sentence" - but even that is hard in English. For example: "I set the table up" - up modifies set to create the "phrasal verb" set up, and isn't related to its typical meaning of position. But remove that little word and the whole meaning of the sentence drastically changes. I've seen "up" in this instance described as an adverb, which tends to be the "throwaway stat" for words that don't fit any of the other traditional parts of speech.