As already mentioned by @ta in the Wikipedia etymology link, the 'radix', is the the base of your trie. In this case we mean the numeric base, and we'll consider storing binary data. Radix R = 2 ^ x, where x >= 1.
Take the example of a binary (2-ary) trie. The radix is 2, so at each node you can compare one bit. The two children will handle all of the possible outcomes:
- the bit is 0
- the bit is 1
The next level of complexity would be a 4-ary trie. As @Garrett mentioned above, the radix must be a power of two so that it can always handle all possible sorting outcomes of the binary data we're using it for. A 4-ary trie can compare two binary bits with the four possible outcomes:
- 00
- 01
- 10
- 10
These four options each lead to a different child node.
Now, in answer to your question about the radix for the English alphabet. You want to encode letters from a to z (26 letters) so will need to have a radix of at least 2^5 = 32. This is the smallest radix that will let you switch between every letter and conform to the 'powers of two' rule. 2^4 = 16 wouldn't handle all of the letters.
As an example, let's imagine the following encoding:
- 00000 represents 'a',
- 00001 represents 'b',
- ... etc
- 11001 represents 'z',
- 11010 to 11111 are not in use (yet)
we can now do a comparison of five bits at every node in the tree, so every node can now switch between any Roman-alphabet letter. If you want a trie that will handle upper case letters then you will require a larger radix. A radix of 2^6 will give you enough to do this, but it comes at the cost of more wasted space (unused branches) in the trie.
Further reading: Sedgewick, Ch 15.4, on Multiway Tries. The 3rd edition of Algorithms by Cormen is generally excellent but doesn't do much for multiway tries.