In ye olde days it was common for processors to access memory in units other bytes. It was extremely common for computers to use memory in units of 36 bits (DEC, Sperry). There were deskop computers that used 14-bits.
A word is then the smallest unit of memory that a specific computer addresses. When the word is 8-bits, it is called a byte.
In your example the word becomes 16-bits.
You question is totally screwed up. If you have a word addressable machine, then it does not have bytes.
They appear to be saying
The cache size is 1 Megaword, and the line size is 4 words long
I am continually amazed at how academics come up with ways to make the simple complex.
There are actually two meanings of the term "word" in common use. (1) is the addressable unit size. (2) Processors generally support multiple integer sizes (e.g., 8, 16, 32, 64, 128 bits). Is it common for processor documentation to call one of this sizes greater than 8 a word.
The processor might have a MOVE BYTE instruction that moves 8-bits and a MOVE WORD instruction that moves 16, 32 or 64 bytes; whatever the processor documentation want to call a word.