You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
49 lines
2.6 KiB
49 lines
2.6 KiB
From text to speech: The MITalk system
|
|
|
|
3.Nouns are very often compounded into classifier strings (e.g.
|
|
|
|
cathode ray tube cleaning fluid).
|
|
The local level merely interprets the ATN grammar.
|
|
|
|
The global parsing strategy proceeds as follows: it looks for the longest noun
|
|
group (a noun phrase up to the head noun, possibly including an initial preposition)
|
|
that it can find beginning with the first word in the sentence. If it locates one, then
|
|
a “node” representing that constituent is constructed, the “current word” pointer is
|
|
advanced to the word after that constituent and the process begins again at that
|
|
point. If no noun group is found, the parsing logic attempts to find the longest
|
|
verb group starting at the word pointer. If it is successful, then a “node” is built,
|
|
the pointer is incremented, and the process begins again. If neither type of group
|
|
can be found at a certain point in the sentence, no node is created and the pointer is
|
|
simply moved to the next word in the sentence and the process begins again.
|
|
|
|
At the local level, the parser uses the ATN to find a constituent. There are
|
|
two pointers, one pointing to the word in the sentence currently being examined
|
|
and one pointing to the current state in the net, which begins at the initial state for
|
|
noun groups or verb groups. The parser tries each arc leading from the current
|
|
state in the order in which they appear in the net. This net is shown in Tables 4-1
|
|
(noun group) and 4-2 (verb group).
|
|
|
|
Testing an arc is done as follows:
|
|
|
|
1. If the arc label is JUMP or POP, then the exit routine associated with
|
|
|
|
that arc is tested. If it is successful, then for JUMP the state pointer
|
|
is advanced to the destination state (the word pointer is not
|
|
incremented), and the process begins again at that state. For POP, a
|
|
node is built if the popped constituent is longer than any found so far,
|
|
and the process continues with the next arc leaving the state. (That
|
|
is, parsing is exhaustive.) If the tests are unsuccessful, then the par-
|
|
ser simply checks the next arc leaving the state.
|
|
|
|
2. If the arc label is a part of speech and the current word does not have
|
|
this part of speech, the parser continues with the next arc. If the cur-
|
|
rent word does have this part of speech, the exit routine is tested. If
|
|
successful, the word pointer is incremented and the state pointer is
|
|
advanced to the destination state of the arc. If it fails, the next arc is
|
|
attempted. If the parser is to test the next arc for some state and no
|
|
arcs remain, then the state pointer is reset to the state from which the
|
|
arc led which brought the process to the current state, and the process
|
|
begins again with the next arc in the new state.
|
|
|
|
46
|