This page shows visualizations of some width-3 1-d convolutional filters from Google's lm_1b language model. Each column corresponds to one position in the filter, and shows the characters with the most positive weights. Use the checkbox in the bottom-right to also see the most negative weights (may be slow).

Below that are examples of words for which the filter emits the highest values. A filter's response is its maximum value over all substrings it sees in the word. So if a filter has high weights on 'c' in the first position, then 'a', then 't', it will assign equally high scores to 'cat', 'fatcat', 'concatenate', etc. The portion of the string in blue is the substring the filter is responding to.

'^' and '$' represent beginning and end of word markers, respectively. '_' is a padding character. Literal versions of those characters are escaped with a backslash.

Use the links at the top to see filters of other widths.

Check out my blog post here for a bit more context.

Filter 0 (bias = -0.62) #

/ k
O v
L t
G q
<BOS> b
o B
Q g
Y w
K i
\194 h
" e
X a
^ F
8 p
S H
<EOS> n
5 T
6 r
l u
I
. p
? J
( P
y z
H v
\$ B
F b
h K
! \195
Q 0
u l
s o
M %
" ;
t k
S x
/ i
: 9
A G
& R
t b
I x
\194 J
Q f
/ G
T 9
$ Z
A p
l K
D \195
, 3
S 8
O v
C w
B
P
g
-
y
E

Non-zero for 14.5% of words.

Filter 1 (bias = -0.39) #

f u
K D
w T
W d
N v
<BOS> l
x j
Q -
S R
, 0
O Y
y g
F h
e J
X H
B 1
3 C
5 z
2 \194
Z t
U f
u F
Y e
H x
R j
& p
1 S
z \$
T c
Z t
\195 .
A r
) N
a n
i w
/ g
L \163
: 5
\194 o
; E
9 m
x g
8 t
B y
W p
Q -
\194 j
N i
7 P
6 r
" T
q l
X O
v D
4 d
5 s
b G
3 M
a f
2 U

Non-zero for 17.1% of words.

Filter 2 (bias = -0.45) #

<BOS> k
e u
X U
l B
L h
j v
D C
O R
5 i
- b
6 T
d Y
2 a
E s
J c
7 '
/ H
Q V
. m
N A
Q v
X x
r c
M p
Y s
H t
: i
Z a
? z
; d
/ w
! 0
" k
( f
& n
L C
R g
N h
m 5
\195 B
Q m
" w
Y f
X -
\194 v
$ p
7 s
8 a
R z
9 d
V u
W g
S i
U
n
x
e
.
y
l

Non-zero for 13.5% of words.

Filter 3 (bias = -0.50) #

<BOS> c
X d
E v
O x
W u
K C
N .
S -
3 p
H '
M s
Q n
J k
4 g
2 y
5 z
Y m
I a
e U
^ h
W u
Q R
X j
" D
! H
x n
: h
p C
K J
w U
a r
b -
\194 l
q o
2 L
; g
\163 T
? M
6 )
V i
- x
D k
d f
J S
u B
1 h
Z b
2 t
I A
0 p
P F
7 W
6 a
j m
\195 ,
3 K
G '
9 y
R r
L c

Non-zero for 15.7% of words.

Filter 4 (bias = -0.45) #

N -
B g
F p
, G
A j
U J
S v
L d
K i
Q m
f 0
a P
/ z
W V
8 b
t c
" w
s \195
\194 T
y l
Y f
G .
i F
O x
g n
z c
V N
W v
X ?
E '
1 y
T d
H \$
A -
2 u
4 r
3 e
J q
I M
& m
W c
H z
Q v
X C
E o
4 D
" l
2 m
3 s
q n
I d
Y f
7 x
N .
$ p
6 '
e -
0
g
t

Non-zero for 13.6% of words.

Filter 5 (bias = -0.58) #

X -
Z f
Q x
K k
D p
/ s
8 v
L i
M j
W b
" g
^ h
6 '
2 l
N t
1 u
Y r
G n
7 d
3 m
R X
u v
r 6
s e
U 0
- 5
' B
? p
& q
k W
( T
Q D
" 2
\195 x
C K
! l
o 4
Y L
O j
: F
F h
Q y
P q
j n
X u
E w
J o
7 a
L i
D x
S v
6 k
5 -
G .
8 g
K p
9 t
V H
Z c
2 W

Non-zero for 11.6% of words.

Filter 6 (bias = -0.57) #

R m
9 y
Q p
<BOS> g
U l
B i
8 o
k h
" t
C -
7 f
I O
Z e
\194 M
V j
N w
b L
v S
u .
z G
f h
m g
P E
K u
; .
M A
p c
V a
X ?
' T
/ q
J v
% D
, \163
i d
n (
l t
6 0
Z U
: r
g m
j U
4 P
7 z
F f
S p
E o
5 T
h K
8 B
e t
Q u
V k
3 y
2 i
6 a
X l
Z ,
9 \195
0 v

Non-zero for 19.8% of words.

Filter 7 (bias = -0.60) #

B d
w p
W -
K g
<BOS> l
M P
N j
X C
Z D
q r
E s
b h
v G
V R
2 u
Q y
" '
9 1
m 7
<EOS> c
? p
. P
W F
( f
q D
H j
! d
: G
w c
Y %
u S
Q s
- C
N 0
" l
& K
Z z
X L
\194 )
5
g f
V y
X U
0 F
G r
7 N
Y u
1 s
Z o
6 B
5 t
j a
2 ,
i x
4 K
J .
I L
\194 m
C O
w S

Non-zero for 9.9% of words.

Filter 8 (bias = -0.66) #

<BOS> T
Q D
s y
- p
w h
W P
" d
S H
' c
^ B
<EOS> L
\194 F
V m
E t
q
r
l
f
k
1
U p
u g
Q j
Z -
B i
L S
& w
D x
" o
T e
R f
/ O
) 5
\194 l
Y E
H h
X n
8 t
! s
? G
- A
' h
V t
f T
P H
b y
R D
J a
x L
9 1
s q
v W
Q N
j O
k E
\194 c
Z o
" g
p B
7 i

Non-zero for 11.0% of words.

Filter 9 (bias = -0.56) #

y -
C w
F I
c E
P <BOS>
Y J
h j
" i
k t
8 2
K .
m e
p u
Q n
G s
V g
L o
B z
T v
b a
- F
J S
X t
\195 c
Y y
9 s
6 f
1 k
o A
w C
Z h
0 \$
: U
3 r
2 x
7 \163
n ,
u e
i .
l E
Y -
h w
Q z
S d
" f
V P
\194 m
W J
4 e
7 D
H E
8 U
$ \195
x p
C u
k o
X I
9 l
K
v

Non-zero for 9.5% of words.

Filter 10 (bias = -0.44) #

H x
U v
L f
Z c
P b
1 o
D w
/ S
y g
X .
3 -
6 k
Q t
7 '
2 e
I s
C p
8 <BOS>
, j
^ h
b t
Y d
V I
B n
G y
J -
X D
x p
9 C
K ,
Z \$
" c
) .
8 f
L u
M s
6 e
% r
m i
0 H
' y
\194 p
t h
j a
- K
. P
C x
$ L
s 1
Q 3
I H
V f
R B
S 8
z r
/ q
v i
b
e
2

Non-zero for 15.3% of words.

Filter 11 (bias = -0.48) #

A -
/ v
, d
B g
I u
C p
Q b
N x
H j
W r
5 G
X J
t e
K f
n m
Z E
^ o
4 P
2 \163
'
X s
Q o
H m
8 z
7 f
2 l
6 w
1 '
" g
q -
D x
9 t
3 S
4 n
W %
P k
e i
? c
Z .
E A
G F
z f
w h
g u
Z t
m x
K k
L d
O q
X ,
A v
c R
Y H
V B
. N
U j
E \194
M I
\195 9
/ r

Non-zero for 10.4% of words.

Filter 12 (bias = -0.51) #

j a
V y
<BOS> o
4 u
5 .
7 U
S T
Z r
6 h
M m
F q
3 b
2 d
X L
C A
I x
J v
9 z
Q l
^ \195
Q g
X d
N p
Z h
/ i
B s
! G
? E
( j
. S
& x
K y
: -
\194 u
q %
M 1
n e
W 0
w c
V o
d w
D B
1 b
7 f
C k
P K
H m
6 o
8 N
I M
0 x
2 .
4 E
L v
p W
5 r
\194 S
l z
j t
X s

Non-zero for 9.9% of words.

Filter 13 (bias = -0.48) #

<BOS> y
J p
j c
X f
l x
- h
\194 a
I o
V s
Q ,
Y k
7 C
6 r
0 F
^ n
Z U
R i
\195 O
9 A
M K
R p
Z x
U f
u v
Q l
? o
( i
& d
H t
r m
Y e
) ,
M y
" a
V q
A W
N h
! c
C K
E 0
k d
w -
B D
K u
z h
V l
W j
A L
U y
b .
G 1
Z x
Y F
E 7
S H
Q f
T e
O n
X 6
c P

Non-zero for 13.5% of words.

Filter 14 (bias = -0.51) #

B -
K g
C r
Z d
c u
V h
k j
U E
9 H
v O
F l
n e
z i
f o
5 .
8 p
M y
, \163
6 I
x Y
M x
Z p
m a
- h
? d
/ v
u S
K A
R t
J q
' c
Y E
& \163
! 0
X F
N s
( i
H k
r g
Q l
Q j
W o
X f
" J
q l
a -
Z t
. D
b i
$ s
\194 S
V e
8 p
Y O
P
T
E
M
g
F

Non-zero for 8.4% of words.

Filter 15 (bias = -0.53) #

. J
' B
s K
Q e
C E
g P
\194 f
/ T
0
N
9
o
3
\195
M
D
2
v
j
F
G n
Y f
X .
O a
M x
K t
E q
" -
T d
3 F
Z v
0 A
J u
V l
Q k
8 '
W s
) h
6 \$
4 p
x Z
W M
h F
v P
a J
\194 j
p D
o L
Y r
S K
i n
" U
t e
q N
' \195
d R
O H
l C
s f
$ 3

Non-zero for 11.5% of words.

Filter 16 (bias = -0.58) #

w y
z .
I h
i r
J F
2 m
0 u
5 L
E f
B '
<BOS> M
W d
v "
K H
3 c
6 D
9 Q
4 b
V N
1 x
Q f
" %
W L
\194 l
Y m
( J
T x
X -
& P
C n
b
a
s
\195
p
d
e
F
U
o
Z h
G x
X u
w H
z f
M a
K A
2 k
Q r
V t
E y
0 l
J i
c ,
5 o
$ F
6 q
- B
g S
D R

Non-zero for 10.1% of words.

Filter 17 (bias = -0.48) #

R .
C g
P h
Q m
9 l
8 w
U e
" A
7 b
3 t
, x
I q
1 L
6 -
\194 v
^ E
<BOS> j
Z a
' o
2 T
- C
E A
v c
X n
u p
" f
W y
e ,
: o
J K
6 k
Q r
\194 t
2 F
? B
d S
) x
j h
! l
M L
b O
V t
Y f
. P
Z p
h y
x d
q i
g E
Q U
X ,
9 D
7 s
\194 T
n e
R I
A K
B o
8 S
' 2

Non-zero for 12.4% of words.

Filter 18 (bias = -0.58) #

i .
4 c
3 z
j d
H D
J U
5 '
F r
6 C
S G
V v
f y
B a
h T
e u
2 Q
M m
7 \163
W -
9 g
M p
v I
) P
" l
b i
Z ,
Y d
V t
9 n
\194 a
? A
X O
B r
8 C
W %
E f
u y
4 1
6 s
. o
Q f
Y v
X B
$ p
Z P
" x
g k
V a
G t
7 F
\194 U
4 m
d
u
e
b
T
o
K
y

Non-zero for 13.3% of words.

Filter 19 (bias = -0.51) #

l v
Y u
O F
o k
S e
L f
A E
G q
/ d
z \163
<BOS> U
K x
5 y
X c
, b
\194 a
<EOS> w
j 9
J 8
^ "
N d
Q m
r i
? s
R y
X f
9 l
w u
q %
( U
" p
B L
E P
K h
W x
Z F
b ,
& D
Y H
O -
\194 -
Q m
x p
C y
9 i
F r
8 g
B T
5 P
7 O
S H
V u
6 w
N E
4 \195
" e
v o
$ G
' d
c J

Non-zero for 22.0% of words.

Filter 20 (bias = -0.46) #

U -
B .
k j
z d
R h
K l
V y
9 e
Q g
Y n
W D
b f
<BOS> t
E o
a m
" p
<EOS> F
G M
Z L
2 c
W j
K F
O u
X h
y s
p g
; v
Q R
! )
2 l
w k
a -
, x
" S
P .
/ b
1 C
I J
? V
: '
A -
z f
c j
Q u
W J
. P
a i
w h
X r
K x
t p
/ R
B m
T F
N M
I '
U d
Z k
\194 s
2 e

Non-zero for 13.8% of words.

Filter 21 (bias = -0.22) #

<BOS> y
Y f
J p
b n
X c
B F
E d
z C
Q t
R ,
V s
9 '
\195 h
W x
\194 m
" i
<EOS> D
T P
0 g
G e
G f
Y d
g x
V F
Z v
R t
? a
X e
) q
& B
M ,
Q \$
! D
C N
J .
3 l
p
y
u
n
P .
U c
V h
I o
i x
C N
6 g
X q
Q r
z w
/ y
$ e
7 v
l -
\194 f
, n
s \163
Y b
J E
R t

Non-zero for 11.3% of words.

Filter 22 (bias = -0.68) #

K u
5 -
, g
6 r
D k
2 b
0 R
c H
8 h
X Y
/ m
B j
W '
L \195
N V
O .
p s
\194 i
f E
9 U
- A
j B
: c
' a
O h
; k
s T
" q
J C
M t
! v
% F
Q D
n
U
N
0
L
x
b
h -
F v
A w
H z
S J
C T
L o
7 E
y M
4 \195
, u
8 m
x K
5 b
Q e
1 O
Y B
6 t
3 k
l W

Non-zero for 7.6% of words.

Filter 23 (bias = -0.43) #

U o
<BOS> h
z g
m N
Q r
s j
' e
P 0
\194 c
V y
k 1
b O
" 3
C q
/ w
^ E
v n
d J
a 5
X H
Y f
0 .
T y
1 m
V F
) e
G \$
C x
i !
7 ?
g -
J s
R N
k r
4 L
9 '
H a
\194 %
z :
6 w
x u
b A
V t
X r
W D
v H
K g
B I
m d
" -
\194 R
6 j
Q n
f s
p U
Y E
9 o
8 1
' h
k .

Non-zero for 16.6% of words.

Filter 24 (bias = -0.43) #

K -
U .
P g
B h
O x
, v
3 d
2 u
M q
Z j
z b
N l
5 '
/ \194
9 H
f r
G e
8 t
J n
C Y
S U
Q f
Y m
" P
( n
\194 a
: v
7 w
X u
z
p
d
B
\195
y
%
K
D
J
c
H v
/ c
Y f
7 k
l T
A e
i F
1 b
I r
X B
4 M
Q t
L E
$ \163
W D
x
p
j
0
m

Non-zero for 4.9% of words.

Filter 25 (bias = -0.39) #

Y .
S c
i -
3 d
h v
l w
J '
5 m
o g
H b
4 \163
O n
7 e
6 f
, z
1 a
B u
L s
K r
9 Z
V u
Z D
X t
g N
i d
m r
w E
! F
G T
Y R
p o
W U
n f
4 v
5 e
6 \$
b .
K c
s
\163
Q u
" -
X h
K .
P d
V t
R j
Y w
G n
k i
9 g
8 e
O H
C l
p s
$ m
r L
7 v
' o
b D

Non-zero for 21.4% of words.

Filter 26 (bias = -0.46) #

. p
Q i
N z
Z P
H k
<BOS> T
X o
/ J
F G
\194 v
8 O
q K
7 m
" f
^ l
4 U
n \195
A B
0
s
v y
\194 r
x H
J U
l A
0 g
9 ?
6 h
B E
- F
o u
X P
5 O
z 1
V k
b \163
W s
7 C
' t
Y i
P x
C h
I -
p v
Q u
z j
U S
D J
T o
O 4
r f
K b
c s
/ e
X E
y 9
A 5
t Y
G 6
d 3

Non-zero for 27.1% of words.

Filter 27 (bias = -0.46) #

p b
s .
P B
C q
d v
O w
S A
y l
, h
G T
8 u
3 J
1 N
Q m
2 \195
5 M
" L
F H
7 Y
I <BOS>
V f
Y t
X r
Z F
W d
H s
6 o
4 D
i c
m N
) .
G \$
b u
M y
! e
J S
7 ,
w O
g E
3 R
Y w
\194 g
" A
V c
x I
u t
6 z
b O
m r
h E
' n
8 .
9 p
M o
f a
v N
F e
$ 2
7 G
Q D

Non-zero for 11.9% of words.

Filter 28 (bias = -0.32) #

H f
4 K
Z z
V O
7 t
1 l
n U
Y o
W m
g P
X p
q s
3 D
2 L
6 y
h F
9 T
- c
^ S
Q r
X h
J x
Z y
6 c
5 p
V a
& f
Q r
M .
I d
/ k
K u
2 v
Y ?
\194 b
3 g
7 q
F
\163
s
g f
G B
Y K
V F
A N
. P
Z o
C x
Q ,
X v
z y
w U
$ u
j 9
H e
8
M
D
J
d

Non-zero for 13.8% of words.

Filter 29 (bias = -0.61) #

<BOS> y
V -
X u
Q t
7 f
Z p
Y o
L v
5 d
G w
8 i
6 a
9 m
S c
\194 h
J q
^ n
4 k
j T
l x
s P
S T
E B
w b
- m
g q
j ;
4 p
5 X
\$ l
3 a
e D
2 L
G v
O k
( \195
o Y
i d
M U
I y
p v
y u
, -
K j
X k
L R
P b
/ T
O J
Q E
1 .
A t
2 M
6 g
5 B
G c
a w
3 o
W e
7 h

Non-zero for 15.7% of words.

Filter 30 (bias = -0.45) #

C u
Z -
V v
Q t
7 T
F i
G m
5 o
8 y
X W
A w
9 f
L d
0 E
n e
c h
<EOS> H
<BOS> U
R O
S M
W j
? l
" J
y n
Q P
( 0
H p
! %
a D
U -
: C
h f
E \195
u z
q L
. g
O R
Y F
\$ 5
& o
X o
L -
V u
b i
Z t
G f
Q I
Y w
g ,
. n
m v
A s
l k
7 R
F O
8 p
M U
6 d
\194 a
" N

Non-zero for 19.3% of words.

Filter 31 (bias = -0.59) #

<BOS> A
W .
v L
V y
" n
p D
9 h
X u
k H
Q t
P l
6 m
b N
E /
2 U
J F
\194 r
G o
7 c
x C
o H
R m
9 t
K h
G T
8 q
N i
O A
5 k
& a
J y
0 g
3 u
S e
7 b
Q .
z d
C v
s W
c w
L v
y k
/ -
m j
X g
Z w
H R
Q i
D t
K p
U x
A o
. s
M E
6 c
8 I
l J
$ 0
r
'

Non-zero for 14.7% of words.

Filter 32 (bias = -0.55) #

Q j
O v
W k
G n
<BOS> F
z u
S T
s h
K f
" t
/ H
<EOS> D
^ q
Y P
E M
U e
d
-
J
r
E m
r v
N C
e l
O z
3 '
2 s
Q n
J t
9 k
8 x
\$ c
R i
I \194
? U
" .
X d
7 V
\163 u
( A
Y F
W c
X f
H C
J p
l k
w y
- t
\195 x
/ s
L d
$ j
\194 P
e
D
r
S
'
v
\163

Non-zero for 17.1% of words.

Filter 33 (bias = -0.58) #

Y w
H .
h -
R z
k v
V d
S a
F c
M o
4 p
" n
3 l
C x
7 q
Q e
8 t
i I
r \163
P g
U <BOS>
b t
g f
! y
V M
G ,
x O
. T
Y F
7 K
z o
Z D
X U
? N
Q i
0 P
: m
9 e
v H
- B
a u
j y
J k
5 h
6 m
9 U
7 H
- T
X u
0 i
e A
\194 r
2 p
l g
N a
8 Y
Q c
E t
I C
F b
3 \163

Non-zero for 21.6% of words.

Filter 34 (bias = -0.43) #

X C
2 c
W k
Z p
E '
6 t
<BOS> o
4 g
e -
K u
3 r
N s
8 R
5 l
Q h
w d
9 m
B T
7 i
L v
T x
O F
t b
o n
W h
M f
K a
E !
D V
w s
z L
& C
c 7
" P
v k
G A
\194 \$
Y r
0 p
X H
Z t
V T
b p
9 y
J O
G o
8 h
7 D
R ,
6 c
X e
Y f
L q
Q r
4 i
U d
s S
3 \163
z I
5 A

Non-zero for 15.7% of words.

Filter 35 (bias = -0.82) #

9 m
R .
J y
3 -
E d
B p
7 t
8 c
4 g
5 '
N n
Y f
S w
6 v
2 a
0 l
<BOS> q
K u
Q x
V \163
A v
H f
C b
I x
( B
l m
1 K
g M
h w
& e
/ J
7 )
t -
D k
Q E
Y 9
r W
: z
, ;
! o
F o
' J
k K
x \195
t O
s G
f L
v N
C r
\194 Y
S D
. 1
V 3
d R
Q 0
c E
" T
m z
$ Z
n M

Non-zero for 15.3% of words.

Filter 36 (bias = -0.51) #

M p
H .
u z
T c
4 a
J x
j g
B G
i d
Y l
6 s
3 A
V o
k r
9 b
W w
R \163
\194 y
F O
1 L
z f
& F
. y
! M
Q m
A k
l h
G P
: i
Y e
( H
I p
g )
w B
a t
R j
O T
\195 u
- n
X v
G u
p t
X -
Q j
y v
K l
" s
8 h
O n
Z B
c f
P J
W .
V i
$ A
Y H
\163 R
7 I
1 k
2 w

Non-zero for 20.7% of words.

Filter 37 (bias = -0.59) #

- k
d f
L B
7 t
l F
1 S
X c
Z p
. x
D K
H T
6 y
/ v
J m
I s
u b
2 M
Q ,
\194 r
^ i
U j
K g
! h
Q v
P S
a t
; -
Z e
z x
/ 0
X F
L i
\195 c
& 4
: 5
m o
y \$
% s
b E
r n
C p
Z f
V y
9 r
R e
0 l
U q
G h
5 a
z m
\194 t
8 -
4 x
7 o
c d
6 O
Y .
$ i
M P
s \163

Non-zero for 15.3% of words.

Filter 38 (bias = -0.44) #

K -
B .
9 u
G t
8 d
V m
P H
0 n
5 h
3 l
X g
2 y
6 '
S q
7 o
k w
z j
4 r
Q /
C e
o H
O F
- d
S )
K 1
G D
Y C
& I
% k
' A
" q
s P
z n
f T
x h
w a
: u
N 2
. t
W i
Q m
I f
E v
S -
A p
O x
2 n
W y
X u
$ c
7 M
4 o
5 h
k
b
d
P
'
J
q

Non-zero for 11.9% of words.

Filter 39 (bias = -0.53) #

x R
c U
p H
d <BOS>
l u
. Z
o k
D J
y E
h M
v V
L I
t w
g 3
f
0
S
a
C
Q -
& v
Y p
/ i
Z d
X f
G x
K e
C u
" w
\194 q
L n
M J
S a
V j
U I
( o
A h
E
g
- h
J c
I T
' B
s A
Q t
R D
P y
$ k
j F
O v
C
L
q
H
a
0
x
U
N

Non-zero for 2.9% of words.

Filter 40 (bias = -0.59) #

h f
Y E
7 k
\194 U
1 w
H P
6 m
4 e
5 K
8 z
l b
n r
/ B
x s
^ F
9 t
C T
3 p
X M
0 \163
Q k
X m
7 t
8 A
6 h
9 i
2 T
" g
5 o
3 c
4 B
: y
\194 p
Z a
( w
\$ U
E u
d r
J z
! l
d k
s B
F T
D b
5 Y
C m
j q
7 W
I o
2 h
6 r
8 M
L w
G H
4 y
S v
0 N
z K
1 i
E \195

Non-zero for 15.1% of words.

Filter 41 (bias = -0.48) #

p u
G .
P A
O L
i h
W F
X N
K t
V H
2 B
3 U
1 D
z M
6 T
x j
Q E
" l
' e
7 r
5 q
Z x
X p
! o
Q h
? k
M g
W S
/ l
H v
2 j
K s
L R
6 -
( i
U r
N t
w c
: f
8 '
y z
D p
F x
u g
j w
H G
M b
L o
t a
\194 W
/ y
U O
C c
R z
I K
6 r
7 \163
l i
Z q
J h
T k

Non-zero for 15.4% of words.

Filter 42 (bias = -0.46) #

Y F
v r
u e
i f
T j
W .
U Q
\194 N
z S
m p
o g
J c
1 \163
B E
H P
- b
6 x
0 A
w y
l 8
Z -
X p
Q f
Y x
& v
K t
/ d
8 g
9 h
L e
V j
U k
6 .
7 s
M r
" i
3 q
G a
( u
\194 n
G d
U f
z x
K -
A v
Z F
L p
Y q
R t
O e
\195 '
J j
o h
B \194
E n
w .
3 \163
0 k
M 6
T I

Non-zero for 4.7% of words.

Filter 43 (bias = -0.37) #

<BOS> m
I y
Q h
O u
S b
\194 x
X f
R .
5 g
l c
7 n
^ v
t M
, -
z Z
/ k
H
a
F
U
g f
Z x
! o
H S
m N
? v
V B
G F
X s
A E
w 9
Y R
i \$
1 t
n ,
) e
C O
y K
M "
/ 8
S u
V -
Q a
F d
j o
X U
M \195
\194 z
$ y
" v
4 n
5 L
7 p
1
w
D
J
m
q
i

Non-zero for 15.7% of words.

Filter 44 (bias = -0.55) #

n T
7 c
9 t
H v
a m
Q o
I O
Z g
4 M
8 D
6 G
3 y
V z
A -
N E
5 p
2 e
X \163
B j
F d
K d
J F
o t
; h
O C
b D
B s
w A
Y u
X g
\195 .
W H
M c
N \$
G j
% y
9 n
r 1
R I
" )
S -
Q m
F p
C i
\194 \195
s b
5 u
t P
$ y
c J
" r
8 o
4 H
7 q
, w
/ a
A l
N g
h
T

Non-zero for 18.3% of words.

Filter 45 (bias = -0.56) #

S J
y v
t -
s b
h Z
F \195
, 0
O n
A 9
f X
U V
" w
E <BOS>
. z
H P
c q
' B
Q 6
u <EOS>
d 7
Q h
K g
P v
X j
I -
/ x
Z u
N s
! l
, S
2 m
8 t
; i
U .
& b
O k
9 o
" T
3 e
W c
Q U
X k
W z
" G
\194 B
I c
q C
e R
- o
$ s
d K
. x
t b
H J
2 A
L
\195
p
f
n

Non-zero for 7.8% of words.

Filter 46 (bias = -0.66) #

<BOS> A
- g
' D
\194 L
" T
W c
Q E
^ G
f h
/ 0
F
a
z
r
e
p
l
B
U
1
v A
- r
\194 y
W F
X L
" U
6 C
J %
9 h
q H
w s
x g
0 S
; ,
V a
' P
T \$
Y N
: D
M c
Q -
/ p
L i
A g
F w
U v
N J
B j
\194 d
" x
K o
. e
X 1
C n
D E
$ u
, 0
Z h
S G
8 2

Non-zero for 6.3% of words.

Filter 47 (bias = -0.37) #

9 m
7 y
5 .
3 g
N U
2 c
4 b
8 u
I T
6 p
<BOS> k
S z
Q -
E '
R M
0 v
1 d
J t
W l
, L
Y n
T f
v r
\194 .
G ?
S F
W N
0 !
" -
O w
z Z
V A
; H
6 \$
c I
X L
h u
x a
i e
o y
Q i
. p
" T
/ w
L k
\194 J
' v
$ o
s t
Z h
X 1
U B
8 j
n
H
0
e
g
q
2

Non-zero for 11.1% of words.

Filter 48 (bias = -0.37) #

D f
R y
<BOS> w
\194 p
7 m
j K
0 s
J x
Y a
v O
T W
9 i
Q S
u o
d ,
X e
I E
l U
C A
6 c
Q i
. p
N h
? H
& 1
( g
" k
/ d
! l
K P
M 4
c J
' x
Z j
r a
\194 %
X u
O T
R v
: 0
Q g
" j
W D
B d
N -
X p
\194 c
K i
b 0
$ G
/ l
' 1
9 J
Y e
T
t
h
n
C
o

Non-zero for 9.5% of words.

Filter 49 (bias = -0.59) #

X u
2 h
<BOS> .
I o
6 c
W U
V T
3 l
Q v
5 s
Z A
4 L
7 m
P z
9 D
K g
n t
^ r
1 b
8 d
x u
o j
c -
K H
W P
a t
8 F
5 U
. M
0 E
9 I
B i
N e
z T
G k
X s
Y d
\194 r
! m
q ;
d h
v f
c r
D H
z b
T L
0 u
p m
I N
t M
\194 F
\163 Y
C R
2 A
X n
Q x
G J
1 S
W B
O s

Non-zero for 27.3% of words.

Filter 50 (bias = -0.62) #

w l
y \194
f Y
K D
m 7
c T
r j
. d
Z 0
n 6
N J
M v
s R
U h
a <BOS>
e 9
g 1
p u
k X
E L
Q f
W j
a -
A m
X p
I c
: M
& o
( v
! x
2 n
H '
q F
E g
U k
" s
N C
Y J
/ y
i
W C
X d
Q s
" p
M D
Y c
w P
$ R
\194 j
m n
Z g
K u
V F
q r
/ l
b z
0
U
k
I

Non-zero for 11.4% of words.

Filter 51 (bias = -0.61) #

2 k
1 g
D b
d m
I V
N h
6 S
3 x
8 Y
9 r
/ '
, p
W j
u A
5 l
o G
K F
- c
O s
U C
z F
K g
& h
U j
9 e
; \$
B H
W .
R y
Y f
o n
v t
J S
\194 A
O r
\195 ?
" m
X s
Q x
T M
o k
O F
K g
N m
/ v
Y t
Q b
" e
\194 j
9 i
$ h
X p
8 u
R f
W d
3 H
5 T
G P
s
V

Non-zero for 10.7% of words.

Filter 52 (bias = -0.78) #

d k
y b
D B
L v
1 R
H <BOS>
2 V
/ x
O '
P j
6 S
X f
p r
, s
W c
a J
3 F
I w
l 9
U g
u j
Y F
v e
B f
a p
U P
h g
W M
T \$
o -
R S
z 5
\194 r
& G
q E
" X
k I
A Z
x m
9 J
Q f
7 m
X o
6 w
4 .
8 c
" t
\194 n
2 y
Y B
9 x
$ A
1 r
V K
3 a
I p
5 k
H v
b
z

Non-zero for 12.0% of words.

Filter 53 (bias = -0.49) #

<BOS> p
j P
S a
. K
M o
s \195
F l
g B
' y
Q 1
t z
E x
V i
\194 L
" d
- J
4 q
e T
^ U
Z ,
z h
R F
- y
v L
k f
' S
I \$
Q H
; e
\195 x
b 4
& A
! 5
V %
r ,
P 3
w M
C i
T N
\194 6
Q f
Y x
X v
Z p
G k
/ t
L n
$ -
" e
U F
D i
V h
H o
s
w
B
a
q
j
d

Non-zero for 8.8% of words.

Filter 54 (bias = -0.42) #

9 m
4 t
2 .
3 -
8 '
1 f
7 l
6 r
5 p
0 <BOS>
E o
J k
Z y
G b
H T
U g
R v
L w
D c
X O
X h
V L
Q l
W x
M .
Z f
) d
" u
; a
k %
w y
T A
G o
K s
\194 F
2 r
n
\$
-
m
P .
D x
J h
G q
U w
L t
K v
Z a
8 k
3 -
6 W
1 b
7 '
0 g
9 n
2 f
R A
5 o
X m
M c

Non-zero for 5.5% of words.

Filter 55 (bias = -0.67) #

m E
' 2
f 3
. 1
k 9
l N
b 4
y 0
t J
p 7
s 5
U 8
<BOS> I
V R
M 6
/ e
z H
C q
v o
P h
N m
I y
9 g
Q i
7 p
( k
8 h
R %
2 s
5 b
q f
\$ G
X M
3 -
& U
0 u
6 '
\194 T
1 Y
D x
G A
Q h
P t
" B
O n
E q
- H
X a
8 .
V l
p u
Z N
$ v
z ,
R x
s L
3 i
' T
K k
S w

Non-zero for 10.3% of words.

Filter 56 (bias = -0.60) #

- F
. C
<BOS> k
w h
z B
E P
u p
' T
Q 1
W 0
m i
/ D
<EOS> S
s c
I ,
o y
a 8
4
7
H
( m
7 v
Q k
\$ z
4 T
3 U
5 B
j b
N c
8 p
: a
2 t
S f
6 y
I P
9 u
X %
1 K
H x
? )
K t
6 .
P j
2 r
a g
3 c
z M
U F
W n
p T
8 h
i k
9 '
1 e
x N
G u
5 D
7 -
, q
J A

Non-zero for 8.2% of words.

Filter 57 (bias = -0.39) #

S u
<BOS> n
Q m
O U
j -
G v
Y a
5 y
\194 w
F \195
E d
" H
s q
7 Z
t P
g B
^ 1
l J
V b
i
\194 p
6 y
D r
v m
X k
9 g
0 i
7 f
8 s
5 w
Q %
( P
& a
4 -
N O
2 ?
J '
B U
j n
" b
J y
z h
B p
K d
R f
9 m
N i
Z x
X t
\195 H
Q -
0 F
5 '
E g
I u
L k
A c
G e
w \163
Y q

Non-zero for 6.4% of words.

Filter 58 (bias = -0.49) #

V T
<BOS> t
Z y
x O
\194 o
F r
b i
Q u
9 p
7 E
6 w
X H
8 I
5 D
' -
C a
4 d
<EOS> h
^ \163
L q
R y
s p
V D
- T
' h
J c
9 q
k e
\194 ?
) m
S H
j L
z t
& d
Y O
Q A
b \163
u K
C o
7 1
O v
N k
/ b
K j
y x
o g
Q T
, F
Z h
3 u
W V
X s
r B
$ t
L E
n e
d
0
i
J

Non-zero for 16.5% of words.

Filter 59 (bias = -0.27) #

. J
' K
- z
<BOS> 2
f E
h U
x 0
t G
" B
\194 1
Q P
r 3
m L
n 5
F D
q T
y \195
S i
6
9
) t
Z o
V l
6 r
8 A
X .
4 a
7 %
9 O
2 u
3 N
5 T
M -
G h
C f
1 z
0 q
Q \195
F s
" I
W J
Q u
S -
" j
Y n
X D
\194 R
$ P
O d
\195
0
r
9
1
F
v
e
Z
U
f

Non-zero for 11.4% of words.

Filter 60 (bias = -0.54) #

p u
W F
X M
I s
Q L
2 j
q U
O h
a m
w D
<BOS> .
1 R
P f
i B
^ J
" l
7 A
\163 S
K c
v
2 f
E r
4 .
3 l
5 m
G h
X u
W t
Z y
V x
6 a
w d
) o
7 p
9 n
J k
8 q
Q b
1 v
0 '
C u
g W
G -
l a
j q
V y
A f
S E
c N
z e
Y d
0 w
D "
L x
Z H
5 v
F 2
m o
/ \163
\194 3

Non-zero for 13.1% of words.

Filter 61 (bias = -0.60) #

s p
h v
S T
L P
A c
H K
u z
F <BOS>
4 0
U k
. w
3 q
Y X
y J
7 B
E t
a D
8 o
x -
/ m
z H
K F
O u
G h
o m
& e
R d
Y j
9 f
0 .
Q y
S -
5 t
3 M
J q
8 ?
/ v
% \$
, n
N k
Q g
W v
I j
" o
X -
a D
, m
N c
$ J
/ T
G
u
0
M
l
h
z
d
i
L

Non-zero for 17.2% of words.

Filter 62 (bias = -0.61) #

2 m
1 f
0 .
3 '
9 t
5 k
7 M
I b
J F
6 y
8 h
4 r
G u
E S
z <BOS>
D s
X l
i x
W j
P g
i c
H D
W L
- z
h K
Y .
4 G
k U
: N
I y
V F
; o
3 C
1 Z
q f
u m
7 A
j 0
2 &
6 r
K g
O .
B d
f c
N h
o u
S n
9 -
, D
W H
Q v
3 j
" t
P C
J m
R A
8 q
X T
E y
Y 1

Non-zero for 15.0% of words.

Filter 63 (bias = -0.46) #

J t
Y y
L p
R d
V q
z e
9 w
B H
G \163
b I
<EOS> -
<BOS> W
l .
K i
Z c
U g
0 h
5 T
\195 r
8 a
K F
O .
W v
G d
3 u
Y f
1 j
w b
i -
2 x
o e
5 t
X l
Z h
& q
4 r
/ \$
, m
z D
'
Y f
H K
m o
g J
h x
Q P
$ z
V p
. E
M B
" 9
' 0
X 5
\194 2
W s
T a
N
e
,
F

Non-zero for 16.1% of words.