It does not make sense to illustrate non-numerical concepts with numerical examples.
Badir's answer is a good one, but let's add the adversary to the explanation of watermarking. The adversary cannot remove the message and create a copy of the host without the embedded message.
Cryptography: is a method to hide information, (or: authenticate information, authenticate users, exchange keys, etc.). It is based on keys and applying keyed transformation to data and the data changes to encrypted data so that with the key one can reverse the transformation (or check authenticity, etc). There are many primitives in cryptography: symmetric encryption where key for encryption and decryption is the same, public-key cryptography where the keys above are different. There are digital signatures, MAC-ing information, and so on many other methods. RSA is a primary example of public key, Diffie-Hellman is a primary example of key exchange, AES is a primary example of symmetric key ciphers etc. Go read more and see various numerical examples of the above concepts.
Steganography: is a method to modify text or other form of information transforming the source to a target file (say text) which encode a message but without knowing the key the text simply looks like a normal text (while secretly embedding a message). Say the key tells you to take every 5th word and look at its third letter and this is the hidden message. Again, work out examples yourself for numerical methods (there are plenty).
Watermarking, is a method to embed an identifier in some hidden way in a file (typically in non text files) so that if you have the key you can authenticate the existence of the watermark. The method is typically hidden to the naked eye, but may be detectable. It's goal is authentication of the source of the data.
I agree to most of what you write Moti, but you exclude steganography by cover synthesis and steganography by cover selection in your definition of steganography. It is not a requirement for steganography in general that it modify a source text. The source text is irrelevant for the purpose, and the stego-system may very well synthesise a cover dependent on the message. The requirement is that the adversary cannot distinguish between the steganogramme and normal, non-secret communication.
Actually, the example you give makes much more sense with cover selection or synthesis. Modifying the third letter of every fith word will look weird, and may raise suspicion. It is, however, possible to select or synthesise a text which encodes the message you want in this way.
If you check out «steganography by cover synthesis» in google scholar you will quickly find more details.