grammarians do not separate morphemes while build dependency tree, they make like this:

this is my example. another example from https://en.wikipedia.org/wiki/Dependency_grammar :

better way of grammarians is to connect not only words but blocks of words, like this:

this is my example. another example from https://en.wikipedia.org/wiki/Government_and_binding_theory :

and i make analysis / dependency tree separating morphemes and using them as if they are “separate words”, like this:

2013-november-30: this was not correct, order of have, he, s, read – i fix it:
in other words, the separateness of words is only in writing, it is not feature of language itself.

though there is past tense shown as “[PAST]” in the 2nd image in the 2nd article ( http://upload.wikimedia.org/wikipedia/en/7/7c/HeSmashedTheVase1.png ), it is not written like separate word (ie as “ed”) , even it is not written at all, only root of verb is written, so it is not clear. but in other images they do not separate morphemes: “likes”, “saw” have only 1 connection per every of them. and they show some “almost glued” parts separately in the first article i have linked/shown:

– ” ‘s “, ” ‘ll “, ” ‘ve ” are separated, but again, “would” is not separated as will+ed.

i have written (about) this in tatar language yesterday: http://qdinar.wp.kukmara-rayon.ru/2013/11/25/grammatika-no-nicik-yasa-w-doros/ , https://vk.com/wall17077748_2708 .

i had written suggestion to write morphemes separately in ural-altaic languages, last year in several forums.

2015-02-21 : i have made a paper about this and other things: //qdb.wp.kukmara-rayon.ru/?attachment_id=311 .