By Dominika Polak

1

FINITE-STATE TECHNIQUES

• By Dominika Polak
• based on
• TITLE Natural Language Processing in
• Prolog
• AUTHOR G.Gazdar C.Mellish
• EDITION Addison Wesley 1989

2
OUTLINE

• Finite-State Transitional Network / Finite-State
  Automaton /
• Diagram of FSTN
• Function of FSTN generation / recognition
• Non-deterministic FSTN
• FSTN with jumparc
• Symbols from SAMPA
• Abbreviations
• Application

3
FINITE-STATE TECHNIQUES

• Finite-State Automaton
• A device which has a finite
• number of states over which
• we traverse by means of arcs.

• Finite-State Transition
• Network
• An ilustration (with labels) of
• finite state automation to
• - recognize elements of lg.
• - generate elements of lg.

4

• DIAGRAM OF FSTN

Y
B
E
3
2
!
1
4
5
2
3
E
L
1 INITIAL STATE
I
2 FINAL STATE
B,Y,E,L,I,! - LABELS
5
FIRST FUNCTION OF FSTN GENERATION
( Creating all possible transitions )
f.e
QuestionHow to get from Kurpinskiego to UAM by
tram ?
Description of the net NameThe way by
tram Initial state1 Final state4 1-(From 1 to 4
by 12)-1-4 (12) 2- 1-2 (14) 2-4 (10) 3- 1-2
(14) 2-3 (14) 3-4 (6) 4- 1-2 ( D) 2-3 (14)
3-4 (6) 5- 1-2 (D) 2-4 (10)
10
14

6
14
D
2
4
1
3
Kurp.
PKS
M teatr
M dwor.
12
SECOND FUNCTION OF FSTN RECOGNITION
(Finding out particular finite transition)

Question
Is it possible to find out sequence of
transition 14,15,10 ?-NOT RECOGNIZED
D,14,6 ?-RECOGNIZED
6
FSTN - RECOGNIZING \ GENERATING
y
2
3
b
e
!
4
5
1
l
2
3
e
i
RECOGNIZING
GENERATING
From 1 to 2 by b From 1 to 2 by l From 2 to 3 by
y From 2 to 3 by i From 3 to 4 by e From 4 to 5
by !
B,y,e From 1 to 2 by b From 2 to 3
by y From 3 to 4 by e B,e,y Not
recognized
7
NON-DETERMINISTIC FSTN
A net with at least one state from which more
than one arc with the same label traverses.

y
e
!
b
1
2
3
4
5
y
Non-deterministic state 2 has two choices
Return to state 1 (in attempt to find the
following b
Move to state 3
FINITE STATE NETWORK WITH JUMP ARC
A net which has unlabelled jump arc, which means
that when the machine is generating or
recognizing, it can traverse such an arc and jump
without consuming any symbols from the input
string.
8
SAMPA (Speech Assessment Methods Phonetic
Alphabet) is a machine-readable phonetic
alphabet. In the present study IPA phonetic
transcription was not used for encoding phonemes
because some of its symbols are inconvenient for
computer processing, e.g. QNZD
SAMPA
f.e
POLISH
ENGLISH
SAMPA
SAMPA
X hymn xImn
V cut kVt Ts cyrk
tsIk D
this DIs o was vos
Q pot pQt
9
ABBREVIATIONS
A
From 1 to 2 by A A abbreviates a,b,c
1
2
a
From 1 to 2 by a From 1 to 2 by b From 1 to 2 by c
b
1
2
c
10
APPLICATION OF FSTNs
Used in linguistic analyses (esp. in
phonology, morphology and syntax) almost all
branches of industry (most machines are FSAs)
telecommunication (e.g. storage of telephone
numbers) etc.
11
SUMMARY
1) Finite-State Transitional Network- an case of
a finite state automation. 2) Diagram of
FSTN-has initial and final state. 3) Function of
FSTN a) to generate - creating all
strings that agree with the knowledge of
FSTN b) to recognize - finding out
strings that agree with the knowledge of
FSTN 4) Non-deterministic FSTN - A net with at
least one state from which more than one
arc marked by the same label.traverses 5) FSTN
with a jump - A net with unlabelled jump
arc. 6) SAMPA-Symbols for phonetic
representations.
12
THE END
OF
FINITE-STATE TRANSITION NETWORKS