reading att-format problem ?

20 Sep 2005

      Hi,

I tried FSA with very small examples and find out
that:

1. FSA can not read the acceptor in AT&T format. Does
it just only work with transducer in AT&t format ? Or
am I wrong ?

- suppose I have the file name "test.fst" which
contains:

0       1       A
0       2       B
0       3       C
1       0       eps
1       2       B
1       3       C
1
2       0       eps
2       3       C
2
3
- RUN "fsa.linux-x86_64-standard att:testlm.def.DEF
infor". The following message appeared:

PROGRAM DEFECTIVE:
precondition type() == TypeTransducer violated
in virtual void
Fsa::StorageAutomaton::setOutputAlphabet(Fsa::ConstAlphabetRef)
file Storage.cc line 63

stack trace (innermost first):
#2  fsa.linux-x86_64-standard [0x483b53]
#3 
fsa.linux-x86_64-standard(_ZN3Fsa7readAttEPNS_16StorageAutomatonERSi+0x81d)
[0x45ea3d]
#4 
fsa.linux-x86_64-standard(_ZN3Fsa4readEPNS_16StorageAutomatonERKSs+0xf1)
[0x45d3e1]
#5 
fsa.linux-x86_64-standard(_ZN3Fsa4readERKSsN4Core3RefIKNS_8SemiringEEE+0x2c2)
[0x45dd02]
#6 
fsa.linux-x86_64-standard(_ZN7FsaTool4mainERKSt6vectorISsSaISsEE+0x262f)
[0x43eaff]
#7 
fsa.linux-x86_64-standard(_ZN4Core11Application3runERKSt6vectorISsSaISsEE+0x25)
[0x48ece5]
#8 
fsa.linux-x86_64-standard(_ZN4Core11Application4mainEiPPc+0x886)
[0x48ffb6]
#9  /lib64/tls/libc.so.6(__libc_start_main+0xdb)
[0x38e3e1c4bb]
#10 
fsa.linux-x86_64-standard(__gxx_personality_v0+0xd2)
[0x437dca]

Aborted

2. Problem with "epsilon edges" ?

I found that, with a WFST in AT&T format, FSA
correctly treats the symbol "eps" as an epsilon label.

However, with a FST (without weights on edges), FSA
correctly treats "eps" of the output label as an
epsilon label but "eps" of the input label is kept as
the distinct label.

Here is an example:

WITH WFST: 

Suppose we have a WFST file named DICT1.def

0       1       eps     eps     0
1       2       a       eps     0
2       3       a       eps     0
3       4       a       eps     0
4       5       eps     A       0
0       6       eps     eps     0
6       7       a       eps     0
7       8       a       eps     0
8       9       b       eps     0
9       5       eps     B       0
0       10      eps     eps     0
10      11      a       eps     0
11      12      a       eps     0
12      13      c       eps     0
13      5       eps     C       0
5

- RUN:  fsa.linux-x86_64-standard att:DICT1.def
write,file=DICT1.fsa

We have DICT1.fsa as we expected:

<?xml version="1.0" encoding="UTF8"?>
<fsa type="transducer" semiring="tropical"
initial="0">
<input-alphabet>
<symbol index="0">a</symbol>
<symbol index="1">b</symbol>
<symbol index="2">c</symbol>
</input-alphabet>
<output-alphabet>
<symbol index="0">A</symbol>
<symbol index="1">B</symbol>
<symbol index="2">C</symbol>
</output-alphabet>
<state id="0">
<arc target="1"></arc>
<arc target="6"></arc>
<arc target="10"></arc>
</state>
<state id="10">
<arc target="11"><in>0</in></arc>
</state>
<state id="11">
<arc target="12"><in>0</in></arc>
</state>
<state id="12">
<arc target="13"><in>2</in></arc>
</state>
<state id="13">
<arc target="5"><out>2</out></arc>
</state>
<state id="5"><final/>
</state>
<state id="6">
<arc target="7"><in>0</in></arc>
</state>
<state id="7">
<arc target="8"><in>0</in></arc>
</state>
<state id="8">
<arc target="9"><in>1</in></arc>
</state>
<state id="9">
<arc target="5"><out>1</out></arc>
</state>
<state id="1">
<arc target="2"><in>0</in></arc>
</state>
<state id="2">
<arc target="3"><in>0</in></arc>
</state>
<state id="3">
<arc target="4"><in>0</in></arc>
</state>
<state id="4">
<arc target="5"><out>0</out></arc>
</state>
</fsa>

- But with FST named DICT2.def:

0	1	eps	eps
1	2	a	eps
2	3	a	eps
3	4	a	eps
4	5	eps	A
0	6	eps	eps
6	7	a	eps
7	8	a	eps
8	9	b	eps
9	5	eps	B
0	10	eps	eps
10	11	a	eps
11	12	a	eps
12	13	c	eps
13	5	eps	C
5

and RUN:  fsa.linux-x86_64-standard att:DICT2.def
write,file=DICT2.fsa

then we've got the unexpected result: DICT2.fsa
contained:

<?xml version="1.0" encoding="UTF8"?>
<fsa type="transducer" semiring="tropical"
initial="0">
<input-alphabet>
<symbol index="0">eps</symbol>
<symbol index="1">a</symbol>
<symbol index="2">b</symbol>
<symbol index="3">c</symbol>
</input-alphabet>
<output-alphabet>
<symbol index="0">A</symbol>
<symbol index="1">B</symbol>
<symbol index="2">C</symbol>
</output-alphabet>
<state id="0">
<arc target="1"><in>0</in></arc>
<arc target="6"><in>0</in></arc>
<arc target="10"><in>0</in></arc>
</state>
<state id="10">
<arc target="11"><in>1</in></arc>
</state>
<state id="11">
<arc target="12"><in>1</in></arc>
</state>
<state id="12">
<arc target="13"><in>3</in></arc>
</state>
<state id="13">
<arc target="5"><in>0</in><out>2</out></arc>
</state>
<state id="5"><final/>
</state>
<state id="6">
<arc target="7"><in>1</in></arc>
</state>
<state id="7">
<arc target="8"><in>1</in></arc>
</state>
<state id="8">
<arc target="9"><in>2</in></arc>
</state>
<state id="9">
<arc target="5"><in>0</in><out>1</out></arc>
</state>
<state id="1">
<arc target="2"><in>1</in></arc>
</state>
<state id="2">
<arc target="3"><in>1</in></arc>
</state>
<state id="3">
<arc target="4"><in>1</in></arc>
</state>
<state id="4">
<arc target="5"><in>0</in><out>0</out></arc>
</state>
</fsa>

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com

Owen TN

Stephan Kanthak

tags

participants (2)