AN EXPERIMENT WITH HEURISTIC PARSING OF SWEDISH
Benny Brodda
Inst. of Linguistics
University of Stockholm
S-I06 91
Stockholm SWEDEN
carriers of semantic Information - were substituted for by a mere "-", whereas other formatives
were left in their original shape and place. These
transformed texts were presented to subjects who
w e r e a s k e d to fill in the gaps in such a w a y that
the texts thus obtained were to be both syntactically correct and reasonably coherent.
ABSTRACT
Heuristic parsing is the art of doing parsing
in a haphazard and seemingly careless manner but
in such a way that the outcome is still "good", at
least from a statistical point of view, or, hopefully, even f r o m a m o r e a b s o l u t e point of view.
The idea is to find s t r a t e g i c s h o r t c u t s d e r i v e d
f r o m g u e s s e s about the s t r u c t u r e of a s e n t e n c e
b a s e d on s c a n t y o b s e r v a t i o n s of linguistic units
In the sentence. If the guess comes out right much
p a r s i n g t i m e can be saved, and if it does not,
m a n y s u b o b s e r v a t i o n s m a y still be valid for rev i s e d guesses. In the (very p r e l i m i n a r y ) e x p e r i ment reported here the main idea is to make use of
( c o m b i n a t i o n s of) s u r f a c e p h e n o m e n a as m u c h as
p o s s i b l e as the base for the p r e d i c t i o n of the
s t r u c t u r e as a whole. In the p a r s e r to be developed along the lines sketched in this report main
stress
is p u t on a r r i v i n g
at i n d e p e n d e n t l y
working, parallel recognition procedures.
The result of the e x p e r i m e n t
was r a t h e r
astonishing. It turned out that not only were the
syntactic structures mainly restored, in some few
cases also the original content was reestablished,
a l m o s t w o r d by word. (It was b e y o n d any p o s s i bility that the subjects could have had access to
the original text.)
Even in those cases when the
text i t s e l f w a s not r e s t o r e d to this r e m a r k a b l e
extent, the stylistic v a l u e of the v a r i o u s texts
was almost invariably reestablished; an originally
lively, n a r r a t i v e story c a m e out as a lively,
n a r r a t i v e story , and a p i e c e of r a t h e r dull,
f a c t u a l text (from a s c h o o l text book on s o c i o logy) invariably came out as dull, factual prose.
The work reported here Is both aimed at simul a t l n g c e r t a i n a s p e c t s of h u m a n l a n g u a g e perc e p t i o n and at a r r i v i n g at e f f e c t i v e a l g o r i t h m s
for a c t u a l p a r s i n g of r u n n i n g text. T h e r e is,
indeed, a great need for fast such a l g o r i t h m s ,
e.g. for the analysis of the literally millions of
words of running text that already today comprise
the data b a s e s in v a r i o u s large i n f o r m a t i o n ret r i e v a l s y s t e m s , and w h i c h can be e x p e c t e d to
e x p a n d s e v e r a l o r d e r s of m a g n i t u d e b o t h in importance and In size In the foreseeable future.
I
This experiment showed quite clearly that at
least for Swedish the information contained in the
c o m b i n a t i o n s of s u r f a c e m a r k e r s to a r e m a r k a b l y
h i g h d e g r e e r e f l e c t s the s y n t a c t i c s t r u c t u r e of
the o r i g i n a l text; in a l m o s t all cases also the
s t y l i s t i c v a l u e and in s o m e f e w cases even the
s e m a n t i c c o n t e n t was kept. (The extent to w h i c h
this is true is probably language dependent; Swed i s h is r a t h e r rich in m o r p h o l o g y ,
and this
property is certainly a contributing factor for an
experiment of this type to come out successful to
the extent it actually did.)
BACKGROUND
T h i s type of e x p e r i m e n t has since then b e e n
repeated many times by many scholars; in fact, it
ls one of the s t a n d a r d w a y s to d e m o n s t r a t e the
c o n c e p t of r e d u n d a n c y in texts. But there are
several other important conclusions one could draw
f r o m this type of e x p e r i m e n t s . F i r s t of all, of
course,
the o b v i o u s c o n c l u s i o n
that s u r f a c e
s i g n a l s do carry a lot of i n f o r m a t i o n about the
s t r u c t u r e of s e n t e n c e s , p r o b a b l y m u c h m o r e than
one has been inclined to think, and, consequently,
It could be w o r t h w h i l e to try to c a p t u r e that
I n f o r m a t i o n in s o m e kind of a u t o m a t i c a n a l y s i s
system.
This is the p r a c t i c a l side of it. But
there is more to it. One must ask the question why
a language llke Swedish is llke this. What are the
theoretical implications?
The g e n e r a ! idea b e h i n d the s y s t e m for heuristic parsing now being developed at our group in
S t o c k h o l m dates m o r e than 15 y e a r s back, w h e n I
was m a k i n g an i n v e s t i g a t i o n ( t o g e t h e r w i t h Hans
K a r l g r e n , S t o c k h o l m ) of
the p o s s i b i l i t i e s of
using computers for information retrieval purposes
for the Swedish Governmental Board for Rationaliz a t i o n (Statskontoret).
In the c o u r s e of this
i n v e s t i g a t i o n we performed some psycholingulstic
e x p e r i m e n t s a i m e d at f i n d i n g out to w h a t extent
s u r f a c e m a r k e r s , such as endings, p r e p o s i t i o n s ,
conjunctions
and other (bound) e l e m e n t s
from
t y p i c a l l y c l o s e d c a t e g o r i e s of linguistic units,
could serve as a base for a syntactic analysis of
sentences. We s a m p l e d a c o u p l e of texts
m o r e or
less at r a n d o m and p r e p a r e d t h e m in such a way
that stems of nouns, adjectives and (main) verbs these c a t e g o r i e s b e i n g thought of as the m a i n
Much Interest has been devoted in later years
to t h e o r i e s (and s p e c u l a t i o n s ) about h u m a n per-
66
ception o f linguistic stimuli, and I do not think
that one s p e c u l a t e s too m u c h if one a s s u m e s that
s u r f a c e m a r k e r s of the type that a p p e a r e d in the
described experiment
t o g e t h e r c o n s t i t u t e imp o r t a n t clues c o n c e r n i n g
the gross s y n t a c t i c
structure of sentences (or utterances), clues that
are probably much less consiously perceived than,
e.g., the a c t u a l w o r d s in the s e n t e n c e s / u t t e r a n ces. To the extent that such clues are a c t u a l l y
p e r c e i v e d they are o b v i o u s l y p e r c e i v e d s i m u l t a n e o u s l y with, i.e. in p a r a l l e l with, other units
(words, for instance).
linguistic unit that can be identified at whatever
stage of the analysis, according to whatever means
there are, i_~s identified, and the significance of
the fact that the unit in q u e s t i o n
has been
identified is made use of in all subsequent stages
of the analysis. At any time one must.be prepared
to reconsider an already established analysis of a
unit on the g r o u n d that e v i d e n c e a ~ a l n s t the
analysis may successively accumulate due to what
analyses other units arrive at.
In next s e c t i o n we give a brief d e s c r i p t i o n
of t h e a n a l y s i s s y s t e m for S w e d i s h that is n o w
under d e v e l o p m e n t at our g r o u p in S t o c k h o l m . As
has been said, m u c h effort is spent on t r y i n g to
m a k e use of s u r f a c e s i g n a l s as m u c h as possible.
Not that we b e l i e v e that s u r f a c e s i g n a l s play a
more important
r o l e t h a n a n y o t h e r t y p e of
linguistic signals, but rather that we think it is
i m p o r t a n t to try to o p t i m i z e each single subprocess
(in a p a r a l l e l
system)
as m u c h as
~osslble, and, as said, it m i g h t be w o r t h w h i l e
to look careful into this level, b e c a u s e the importance of surface signals might have been underestimated in previous research. Our exneriments so
far s e e m to i n d i c a t e that they c o n s t i t u t e exc e l l e n t units to base h e u r i s t i c g u e s s e s on. Another reason for concentrating our efforts on this
level is that it takes time and requires much hard
computational
w o r k to get such an a n a r c h i s t i c
system to really work, and this surface level is
reasonably simple to handle.
The above way of looking upon perception as a
set of i n d e p e n d e n t l y o p e r a t i n g p r o c e s s e s is, of
course, more or less generally accepted nowadays
(cf., e.g., L i n d s a y - N o r m a n
1977), and it is also
g e n e r a l l y a c c e p t e d in c o m p u t a t i o n a l linguistics
that any p r o g r a m that a i m s at s i m u l a t i n g perc e p t i o n in one w a y or other m u s t have features
that s i m u l a t e s (or, even better, a c t u a l l y performs) p a r a l l e l processing,
and the a n a l y s i s
system to be described below has much emphasis on
exactly this feature.
A n o t h e r c o m m o n saying n o w a d a y s w h e n discussing
parsing techniques is that one should try
to i n c o r p o r a t e
" h e u r i s t i c devices" (cf., e.g.,
the m a n y s u b r e p o r t s related to
the big ARPAproject c o n c e r n i n g S p e e c h Recognition and Understanding 1970-76), a l t h o u g h there does not
seem
to exist a very precise consensus of what exactly
that
w o u l d mean. (In m a t h e m a t i c s
the term has
been t r a d i t i o n a l l y
used to refer to i n f o r m a l
reasoning,
e s p e c i a l l y w h e n used in c l a s s r o o m
situations.
In a f a m o u s study the h u n g a r i a n
m a t h e m a t i c i a n Polya, 1945 put forth the
thesis
that h e u r i s t i c s
is one of the most i m p o r t a n t
p s y c h o l o g i c a l d r i v i n g m e c h a n i s m s behind m a t h e matical
- or s c i e n t i f i c
- progress.
In AIl i t e r a t u r e it is often used to refer to shortcut
search m e t h o d s in s e m a n t i c networks/spaces; c.f.
Lenat, 1982).
II
AN OUTLINE OF AN ANALYZER BASED ON
THE HEURISTIC PRINCIPLE
F i g u r e 1 b e l o w s h o w s the g e n e r a l o u t l i n e of
the system. E a c h of the v a r i o u s boxes (or subboxes) represents one specific process, usually a
complete computer program in itself, or, in some
cases, independent processes within a program. The
big "container",
l a b e l l e d "The Pool", c o n t a i n s
both the l i n g u i s t i c
m a t e r i a l as w e l l as the
c u r r e n t a n a l y s i s of it. E a c h p r o g r a m or process
looks into the Pool for things "it" can recognize,
and when the process finds anything it is trained
to recognize, it adds its o b s e r v a t i o n to the m a terial in the Pool. This added material may (hopefully) help o t h e r p r o c e s s e s in r e c o g n i z i n g w h a t
they are trained to recognize, w h i c h in its turn
may again help the first process to recognize more
of "its" units. And so on.
One reason for trying to adopt s o m e kind of
h e u r i s t i c device in the a n a l y s i s p r o c e d u r e s is
that one for m a t h e m a t i c a l
reasons k n o w s that
ordinary, "careful", parsing algorithms inherently
s e e m to refuse to w o r k in real t i m e (i.e. in
linear time), whereas human beings, on the whole,
s e e m to be able to do e x a c t l y that (i.e. p e r c e i v e
sentences or utterances simultaneously with their
production). Parallel processing may partly be an
a n s w e r to that d i l e m m a , but still, any process
that c l a i m s to a c t u a l l y s i m u l a t e s o m e part of
human perception
must in s o m e w a y or other
s i m u l a t e the r e m a r k a b l e a b i l i t i e s h u m a n beings
have in g r a s p i n g c o m p l e x p a t t e r n s ("gestalts")
seemingly in one single operation.
The s y s t e m is n o w under d e v e l o p m e n t
and
during this build-up phase each process is, as was
said above, e s s e n t i a l l y a c o m p l e t e , s t a n d - a l o n e
module, and the Pool exists simply as successively
u p d a t e d text files on a disc storage.
At the
moment some programs presuppose that other programs have a l r e a d y been run, but this state of
a f f a i r s w i l l be valid Just d u r i n g this b u i l d ~ u p
phase. At the end of the b u i l d - u p phase each
p r o g r a m shall be able to run c o m p l e t e l y independent of any other program in the system and in
a r b i t r a r y o r d e r r e l a t i v e to the others (but, of
course, usually perform better if more information
is available in the Pool).
Ordinary, careful, p a r s i n g a l g o r i t h m s are
often organized
according
to s o m e
general
p r i n c i p l e such as "top-down",
"bottom-to-top",
"breadth
f i r s t " , " d e p t h f i r s t " , etc., t h e s e
headings
r e f e r r i n g to s o m e s p e c i f i e d type of
"strategy". The h e u r i s t i c m o d e l we are trying to
work out has no such preconceived strategy built
i n t o it. O u r p h i l o s o p h y
is i n s t e a d
rather
a n a r c h i s t i c (The H e u r i s t i c Principle): W h a t e v e r
67
In the ~ e c o n d phase s u p e r o r d i n a t e d c o n t r o l
p r o g r a m s are to be i m p l e m e n t e d . T h e s e p r o g r a m s
w i l l f u n c t i o n as "traffic rules" and via these
systems one shall be able to test various strategies, i.e. to test w h i c h r e l a t i v e order b e t w e e n
the different subsystems that yields optimal resuit in s o m e k i n d of " p e r f o r m a n c e metric", s o m e
e v a l u a t i o n p r o c e d u r e that takes both speed and
quality into account.
tires; the word minus this formative must still be
pronounceable, otherwise it cannot be a formative.
S M U R F w o r k s e n t i r e l y w i t h o u t s t e m lexicon; it
a d h e r e s c o m p l e t e l y to the "philosophy" of u s i n g
surface signals as far as possible.
NOMFRAS, VERBAL, IFIGEN, CLAUS and PREPPS are
other "demons" that recognize different phrases or
w o r d g r o u p s w i t h i n s e n t e n c e s , viz. noun phrases,
verbal
complexes,
infinitival
constructions,
c l a u s e s and p r e p o s i t i o n a l phrases, respectively.
N-lex, V - l e x and A - l e x r e p r e s e n t v a r i o u s (sub)lexicons; so far we have tried
to do without them
as far as possible. One s h o u l d o b s e r v e that s t e m
l e x i c o n s are no p r e r e q u i s i t e s for the s y s t e m to
work, adding them only enhances its performance.
The programs/processes shown in Figure i all
represent rather straightforward
F i n i t e State
Pattern Matching
(FS/PM) procedures. It is rather
t r i v i a l to s h o w m a t h e m a t i c a l l y
that a set of
i n t e r a c t i n g FS/PM procedures of the type used in
our s y s t e m t o g e t h e r w i l l
y i e l d a s y s t e m that
formally has the power of a CF-parser; in practice
it w i l l yield a s y s t e m that in s o m e sense is
stronger,
at least f r o m the point of v i e w of
convenience. Congruence and similar phenomena will
be r e d u c e d to s i m p l e local o b s e r v a t i o n s . T r a n s formational
v a r i a n t s of s e n t e n c e s w i l l be rec o g n i z e d d i r e c t l y - there w i l l be no need for
performing some kind of backward transformational
operations. (In this respect a s y s t e m llke this
w i l l r e s e m b l e Gazdar's g r a m m a r concept; G a z d a r
1980. )
The format of the material inside the Pool is
the o r i g i n a l text, plus a p p r o p r i a t e
"labelled
b r a c k e t s " e n c l o s i n g words, w o r d groups, p h r a s e s
and so on. In this way, the form of representation
is c o n s i s t e n t
throughout,
no m a t t e r h o w m a n y
d i f f e r e n t types of a n a l y s e s have b e e n a p p l i e d to
it. Thus, v a r i o u s p e o p l e can join our g r o u p and
write their own "demons" in whatever language they
prefer, as long as they can take sentences in text
format, be r e a s o n a b l y t o l e r a n t to w h a t types of
'~rackets" they find in there, do their analysis,
add their own brackets (in the specified format),
and put the result back into the Pool.
The c o n t r o l s t r u c t u r e s later to be s u p e r i m posed on the interacting FS/PM systems will also
be of a F i n i t e S t a t e
type. A s y s t e m of the type
then o b t a i n e d
- a s y s t e m of i n d e p e n d e n t F i n i t e
State A u t o m a t o n s
c o n t r o l l e d by a n o t h e r F i n i t e
State Automaton - will in principle
have rather
complex mathematical
properties.
It is, e.g.,
rather easy to see that such a system has stronger
c a p a c i t y than a Type 2 device, but it w i l l not
have the power of a full Type I system.
Now a few comments
to Figure
i
The "balloons" in the figure represent independent programs (later to be developed into indep e n d e n t p r o c e s s e s inside one "big" program). The
figure displays
those programs
t h a t so far
(January 1983) h a v e b e e n i m p l e m e n t e d and t e s t e d
(to some extent). Other programs will successively
be entered into the system.
T h e big b a l l o o n l a b e l l e d "The C l o s e d Cat"
represents
a program that recognizes closed word
c l a s s e s suc h as p r e p o s i t i o n s ,
conjunctions, pronouns, auxiliaries,
and so on. The C l o s e d C a t
r e c o g n i z e s full w o r d f o r m s directly. The S M U R F
b a l l o o n r e p r e s e n t s the m o r p h o l o g i c a l c o m p o n e n t
(SMURF = " S w e d i s h Murphology"). S M U R F itself is
organized internally as a complex system of independently
operating
"demons" - S M U R F s - e a c h
knowing "its' little corner of Swedish word formation. (The n a m e of the p r o g r a m is an a l l u s i o n to
the p o p u l a r
comic
strip
leprechauns
"les
Schtroumpfs",
which
in S w e d i s h
are called
"smurfar".) Thus there is one little smurf recogn i z i n g d e r i v a t [ o n a l m o r p h e m e s , one r e c o g n i z i n g
flectional endings, and so on. One special smurf,
Phonotax, has an important controlling function every other s m u r f m u s t a l w a y s c o n s u l t P h o n o t a x
before identifying one of "its" (potential) forma-
68
Of the v a r i o u s p r o g r a m s SMURF, N O M F R A S and
IFIGEN are extensively tested (and, of course, The
Closed Cat, w h i c h is a s i m p l e lexical lookup
system), and various examples of analyses of these
programs will be demonstrated in the next section.
We hope to a r r i v e at a crucial s t a t i o n in this
p r o j e c t d u r i n g 1983, w h e n CLAUS has been m o r e
t h o r o u g h l y tested. If CLAUS p e r f o r m s the w a y we
hope (and p r e l i m i n a r y
tests indicate that it
will), we will have means to identify very quickly
the c l a u s a l s t r u c t u r e s
of the s e n t e n c e s in an
a r b i t r a r y r u n n i n g text, thus h a v i n g a f i r m base
for entering higher h i e r a r c h i e s in the s y n t a c t i c
domains.
the p r o b l e m of h o w to m a k e it p o s s i b l e for each
process to work independently of all other in the
system.
The C l o s e d Cat (CC) has the i m p o r t a n t role to
mark words belonging to some well defined closed
c a t e g o r i e s of words. This p r o g r a m m a k e s no internal analysis of the words, and only takes full
words into account. CC makes use of simple rewrite
rules of the type ' ~ => eP~e / (blank)__(blank)",
w h e r e the i n s e r t e d e's r e p r e s e n t the "analysis"
("e" s t a n d s for " p r e p o s i t i o n " ;
P~ = "on"). A
s a m p l e o u t p u t f r o m The C l o s e d Cat is s h o w n in
i l l u s t r a t i o n 2, w h e r e the v a r i o u s m e t a - s y m b o l s
also are explained.
The programs are written in the Beta language
d e v e l o p e d by the p r e s e n t author;
c.f. B r o d d a Karlsson, 1980, and Brodda, 1983, forthcoming. Of
the a c t u a l p r o g r a m s
in the system, S M U R F was
d e v e l o p e d and e x t e n s i v e l y t e s t e d by B.B. d u r i n g
1977-79 (Brodda, 1979), w h e r e a s the others are
(being)
developed by B.B. and/or Gunnel KEllgren,
Stockholm (mostly "and").
III
EXPLODING
The s i m p l e e x a m p l e
above also s h o w s the
format of inserted meta-lnformatlon. Each Identified c o n s t i t u e n t
is "tagged" w i t h s u r r o u n d i n g
lower case letters, which then can be conceived of
as l a b e l l e d
brackets.
This format
is u s e d
throughout the system, also for complex constituents. Thus the nominal phrase 'DEN LILLA FLICKAN"
("the
little
girl")
will
be
tagged
as
"'nDEN+LILLA+FLICKANn" by N O M F R A S (cf. below; the
p l u s e s are i n s e r t e d to make the c o n s t i t u e n t one
c o n t i n u o u s string). We have r e s e r v e d the letters
n, v and s for the m a j o r c a t e g o r i e s nouns or noun
phrases, verbs or v e r b a l groups, and sentences,
respectively, whereas other more or less transparent letters are used for other categories. (A list
of used c a t e g o r y s y m b o l s is p r e s e n t e d in
the
Appendix:
Printout Illustrations.)
SOME OF THE BALLOONS
W h e n a "fresh" text is e n t e r e d into The Pool
it first passes through a p r e l i m i n a r y o n e - p a s s program, INIT, (not shown in Fig. i) that "normalizes" the text. The o r i g i n a l text m a y be of any
type as long as it Is r e g u l a r l y typed Swedish.
INIT t r a n s f o r m s the text so that each g r a p h i c
sentence will make up exactly one physical record.
(Except in poetry, p h y s i c a l records, i.e. lines,
u s u a l l y are of m a r g i n a l l i n g u i s t i c
interest.)
P a r a g r a p h ends w i l l be r e p r e s e n t e d by e m p t y records. Periods used to indicate abbreviations are
Just taken a w a y and the a b b r e v i a t i o n itself is
contracted to one graphic word, if necessary; thus
"t.ex." ("e.g.") is t r a n s f o r m e d into "rex", and so
on. Otherwise, periods, commas, question marks and
other t y p o g r a p h i c c h a r a c t e r s are p r o v i d e d w i t h
preceding
blanks.
Through
this each w o r d is
g u a r a n t e e d to be s u r r o u n d e d by blanks, and delimiters
llke c o m m a s ,
p e r i o d s and so on are
guaranteed to signal their "normal" textual functions. E a c h record is also ended by a s e n t e n c e
delimiter (preceded by a blank). Some manual poste d i t i n g is s o m e t i m e s n e e d e d in order to get the
text n o r m a l i z e d a c c o r d i n g to the above. In the
I N I T - p h a s e no l i n g u i s t i c analysis w h a t s o e v e r is
i n t r o d u c e d (other than into what a p p e a r s to be
orthographic sentences).
The program S W E M R F (or sMuRF, as it is c a l l e d
here) has been e x t e n s i v e l y d e s c r i b e d e l s e w h e r e
(Brodda,
1979).
It m a k e s a rather i n t r i c a t e
m o r p h o l o g i c a l a n a l y s i s w o r d - b y - w o r d In r u n n i n g
text (i.e. S M U R F a n a l y z e s each w o r d in itself,
disregarding the context it appears in). SMURF can
be run in t w o modes, in " s e g m e n t a t i o n " m o d e and
"analysis" mode. In its s e g m e n t a t i o n m o d e S M U R F
simply strips off the possible affixes from each
word; it m a k e s n o
use of any s t e m lexicon. (The
a f f i x e s it r e c o g n i z e s are c o m m o n prefixes, suffixes - i.e. d e r l v a t l o n a l morphemes - and flexlonal endings.) In analysis mode it also tries to
m a k e an o p t i m a l guess of the w o r d class of.the
word under inspection, based on what (combinations
of) word formation elements it finds in the word.
SMURF in itself is organized entirely according to
the h e u r i s t i c p r i n c i p l e s as they are c o n c e i v e d
here, i.e. as a set of i n d e p e n d e n t l y o p e r a t i n g
processes that interactively work on each others
output. The S M U R F s y s t e m has been the test bench
for t e s t i n g out the m e t h o d s n o w being
used
throughout the entire Heuristic Parsing Project.
INIT also changes all letters in the original
text to their c o r r e s p o n d i n g upper case variants.
( O r i g i n a l l y c a p i t a l letters are o p t i o n a l l y provided w i t h a p r e f i x e d "=".) All s u b s e q u e n t analysis p r o g r a m s add their a n a l y s e s In the form of
l o w e r case letters or letter c o m b i n a t i o n s . Thus
u p p e r case letters or words will belong to the
object language, and lower case letters or letter
c o m b i n a t i o n s w i l l signal meta-language information. In this way, s t r i c t l y text (ASCII) format
can be kept for the text as well as for the various stages of its analysis; the "philosophy" to
use text Input and text output for all p r o g r a m s
involved represents the computational solution to
In its s e g m e n t a t i o n
mode SMURF functions
formally as a set of interactive transformations,
w h e r e the s t r u c t u r a l c h a n g e s h a p p e n to be extremely simple, viz. simple
segmentation rules of
the type 'T=>P-",
"Sffi> -S" and "Effi>-E'' for an
arbitrary Prefix, Suffix and Ending, respectively,
but w h e r e
the "Job" e s s e n t i a l l y
consists
of
e s t a b l i s h i n g the c o r r e s p o n d i n g
s t r u c t u r a l descriptions. T h e s e are s h o w n in III. I, below,
together with sample analyses. It should be noted
that p h o n o t a c t l c c o n s t r a i n t s play a central role
69
in the S M U R F system; i n fact, one of the m a i n
o b j e c t i v e s in d e s i g n i n g the S M U R F s y s t e m was to
find out how much information actually was carried
by the p h o n n t a c t l c
component
in Swedish.
(It
turned out to be quite much; cf. Brodda 1979. This
p r o b a b l y holds
for other G e r m a n i c l a n g u a g e s as
well, which all have
a rather elaborated phonotaxis.)
IFIGEN parsing diagram
Aux
(simplified):
n>Adv)o
ATT - -
-A
# (C)CV
-(A/I)T
#
I
where '~ux" and "Adv" are categories recognized by
The Closed Cat (tagged "g" and "a", respectively),
and "nXn" are s t r u c t u r e s
r e c o g n i z e d by e i t h e r
N O M F R A S or, in the case of p e r s o n a l pronouns, by
C C (It should he worth mentioning that the class
of a u x i l i a r i e s in S w e d i s h is m o r e open than the
corresponding word class in English; besides the
" o r d i n a r y " V A R A ("to be"), HA ("to have") and the
modalsy, there is a fuzzy class of seml-auxillarles
llke BORJA ("begin") and others; IFIGEN makes use
of about 20 of these in the p r e s e n t version.) T h e
supine cadence -(A/I)'T is supposed to appear only
once in an i n f i n i t i v a l group. A s a m p l e output of
IFIGEN is given in Iii. 3. Also for IFIGEN we have
r e a c h e d a r e c o g n i t i o n level a r o u n d 95%, which,
again, is r a t h e r a s t o n i s h i n g ,
considering
how
little information actually is made use of in the
system.
NOMFRAS is the next program to be commented
on. The p r e s e n t v e r s i o n r e c o g n i z e s s t r u c t u r e s of
the type
det/quant + (adJ)~ + noun;
where the "det/quant" categories (i.e. determiners
or q u a n t l f l e r s ) are d e f i n e d e x p l i c i t l y t h r o u g h
enumeration - they are supposed to belong to the
class of "surface markers" and are as such identified by The C l o s e d Cat. A d j e c t i v e s and nouns on
the other hand are identified solely on the ground
of their "cadences", i.e. w h a t kind of ( f o r m a l l y )
e n d l n g - l l k e s t r i n g s they h a p p e n to end with. The
n u m b e r of a d j e c t i v e s that are a c c e p t e d (n in the
formula above) varies depending on what (probable)
type of construction is under inspection. In indefinite noun phrases the substantial content of the
expected endings is, to say the least, meager, as
both nouns and adjectives
in many situations only
have O-endings. In definite noun phrases the noun
mostly - but not always - has a more substantial
and r e c o g n i z a b l e e n d i n g and all i n t e r v e n i n g adJectives
have either the cadence -A or a cadence
f r o m a s m a l l but c h a r a c t e r i s t i c set. In a (supposed) d e f i n i t e n o u n p h r a s e all w o r d s e n d i n g in
a n y of the m e n t i o n e d c a d e n c e s are a s s u m e d to be
adjectives,
but in (supposed) i n d e f i n i t e noun
p h r a s e s not m o r e than one a d j e c t i v e is a s s u m e d
u n l e s s o t h e r types of m o r p h o l o g i c a l s u p p o r t are
present.
The IFIGEN case
illustrates very clearly one
of the c e n t r a l points in our h e u r i s t i c a p p r o a c h ,
namely the following: The information that a word
has a specific cadence, in this case
the cadence
-A, is u s u a l l y of very llttle s i g n i f i c a n c e
in
itself in Swedish. Certainly it is a typical infin l t l v a l c a d e n c e (at least 90% of all i n f i n i t i v e s
in S w e d i s h have it), but on the other hand, it is
c e r t a i n l y a very t y p i c a l c a d e n c e for other types
of words as well: FLICKA (noun), HELA (adjective),
DENNA/DETTA/DESSA (determiners or pronouns) and so
on, and these other types are by no comparison the
d o m i n a n t g r o u p h a v i n g this s p e c i f i c c a d e n c e in
running text. But, in connection with an "infinitive warner" - an auxiliary, or the word ATT - the
situation changes dramatically. This can be demonstrated by the following figures: In running text
words having the cadance -A represents infinitives
in a b o u t 30% of the cases. A T T is an i n f i n i t i v e
m a r k e r ( e q u i v a l e n t to "to") in q u i t e e x a c t l y 50%
of its o c c u r e n c e s (the o t h e r 50% it is a s u b o r d i n a t i n g conjunction). The conditional probability
that the c o n f i g u r a t i o n
ATT ..-A r e p r e s e n t s
an
i n f l n l t v e is, h o w e v e r , g r e a t e r
than 99%, prov i d e d that c h a r a c t e r i s t i c c a d e n c e s like - A R N A / O R N A and q u a n t i f l e r s / d e t e r m i n e r s
llke A L L A and
D E S S A are d i s r e g a r d e d
(In our s y s t e m they are
marked by SMURF and The Closed Cat, respectively,
and thereby "saved" from being classified as infinitives.)
G i v e n this, there is a l m o s t no overgeneration in IFIGEN, but Swedish allows for split
i n f i n i t i v e s to s o m e extent. Quite m u c h m a t e r i a l
can be put in b e t w e e n the i n f i n i t i v e w a r n e r and
the infinitive, and this gives rise to some undergeneration (presengly). (Similar o b s e r v a t i o n s reg a r d i n g c o n d i t i o n a l p r o b a b i l i t i e s in c o n f i g u r a tions of l i n g u i s t i c units has been m a d e by M a t s
Eeg-Olofson, Lund, 1982).
T h e F i n i t e State S c h e m e b e h i n d N O M F R A S is
presented in Ill. 2, together with sample outputs;
in this case the text has been preprocessed by The
Closed Cat, and it appears that these two programs
in cooperation are able to recognize noun phrases
of the discussed type
correctly to well over 95%
in r u n n i n g text (at a speed of about 5 s e n t e n c e s
per second, CPU-tlme); the e r r o r s w e r e s h a r e d
about 50% each between over- and undergenerations.
Preliminary experiments aiming at including also
SMURF and FREPPS (Prepositional Phrases) s e e m to
indicate that about the same recall and precision
rate could be kept for a r b i t r a r y types of (nons e n t e n t l a l ) noun p h r a s e s (cf. Iii. 6). (The syst e m s are not yet t r i m m e d to the extent that they
can be operatively run together.)
I F I G E N (Infinitive
Generator)
is a n o t h e r
rather
straightforward
F i n i t e State P a t t e r n
Matcher (developed by Gunnel K~llgren). It recognizes
(groups
of) n n n f l n l t e
verbs. S o m e w h a t
simplified it can be represented by the following
d i a g r a m ( r e m e m b e r the c o n v e n t i o n s for upper and
lower case):
70
IV
REFERENCES
Brodda, B. "N~got om de svenska ordens fonotax och
morfotax",
P a p e r s from the I n s t i t u t e Of
Linguistics (PILUS) No. 38, University of Stockholm, 1979.
versity of Helsinki,
Gazdar, G. "Phrase Structure" i_~nJacobson, P. and
Pullam G. (eds.), Nature of Syntactic Representation, Reidel, 1982
Brodda, B. '~ttre kriterler f~r igenkEnnlng
av
sammans~ttningar" in Saari, M. and Tandefelt, M.
(eds.) F6rhandllngar r~rande svenskans beskrivning - Hanaholmen 1981, Meddelanden fr~n Institutionen f~r Nordiska Spr~k, Helsingfors Universitet, 1981
Brodda, B. "The BETA System,
tions", Data Linguistics,
(forthcoming).
Lenat, D.P. "The Nature of Heuristics", Artificial Intelligence, Vol 19(2), 1982.
Eeg-Olofsson, M. '~n spr~kstatlstlsk modell f~r
ordklassm~rknlng i l~pande text" in K~llgren, G.
(ed.) TAGGNING, Fgredrag fr~n 3:e svenska kollokviet i spr~kllg databehandling i maJ 1982,
FILUS 47, Stockholm 1982.
and some ApplicaGothenburg,
1983
Polya, G. "How to Solve it", Princeton University
Press, 1945. Also Doubleday Anchor Press, New
York, N.Y. (several editions)•
Brodda, B. and Karlsson, F. "An experiment with
Automatic Morphological Analysis of Finnish",
Publications No. 7, Dept. of Linguistics, Unl-
APPENDIX:
Some
1981.
computer
illustrations
The following three pages illustrate some of the parsing diagrams used in
the system: Iii. I, SMURF, and Iii. 2, NOMFRAS, together with sample analyses.
IFIGEN is represented by sample analyses (III. 3; the diagram is given in the
text)• The samples are all taken from running text analysis (from a novel by
Ivar Lo-Johansson), and "pruned" only in the way that trivial, recurrent examples
are omitted. Some typical erroneous analyses are also shown (prefixed by **).
In III. I SMURF is run in segmentation mode only, and the existing tags are
inserted by the Closed Cat. "A and "E in word final position indicates the
corresponding cadences (fullfilling the pattern ?..V~M'A/E '', where M denotes a
set of admissible medial clusters)•
The tags inserted by CC are: aft(sentence) adverbials, b=particles, dfdeterminers,
efprepositions, g=auxiliaries, h=(forms of) HA(VA), iffiinfinitives, j=adjectives,
n=nouns, Kfconjunctions, q=quantifiers, r=pronouns, ufsupine verb form, v=verbal
(group)•
(For space reasons, III. 3 is given first, then I and II.)
Iii. 3:
PATTERN:
aux/ATT^(pron)'(adv)A(adv)'inf^inf A. .. :
..FLOCKNINGEN eEFTER..IkATTk+iHAi+uG~TTui
..
rDETr v V A R v O R I M L I G T i k A T T k + i F I N N A I
rJAGr gSKAg a B A R A a I H J A L P A i
- rDETr gKANg I L I G G A I
g S K A g rVlr iV~GAi
- rVlr gKANg a l N T E a iG~i
. ..ORNA v H O L L v SIG F A R D I G A i k A T T k + i K A S T A i
rDEr g V ~ G A D E g a A N T L I G E N a i L Y F T A i
gSKAg rNlr a N O D V A N D I G T V I S a
iGORAi
..rVlr h H A D E h a A N N U a a l N T E a u H U N N I T u iF~i
. . B E C K M O R K R E T eMEDe i k A T T k + I F O R S O K A i + i F ~ I
eMEDe V A T G A S eFORe i k A T T k + i K U N N A i + I H ~ L L A i
SKOGEN, L A N D E N g T Y C K T E S g iST~i
rDENr h H A D E h M I S S L Y C K A T S ele i k A T T k + i N A i
*** qENq kS g V ~ G A D E g I K V l N N O R N A + S T A N N A i
71
F R A M A T B O J D H E L A DAGEN..
qETTq K A D S T R E C K ele ..
eTILLe ikATTk+iSEi
..
qENq KARL INUTI?
VIPPEN?
HEM eMEDe S K A M M E N ...
eOMe N A R S O M H E L S T .
ePAe rDETr.
N~T eMEDe rDENr, k S ~ k
eUPPe P O T A T I S E N .
B A L L O N G E N FYLLD.
SEJ OPPE.
S T I L L A e U N D E R e OSS.
SITT M~L.
IIi. i:
SMURF - PARSING DIAGRAM FOR SWEDISH MORPHOLOGY
Structural
changes
PATTERNS " S t r u c t u r a l D e s c r i p t i o n s " ) :
I ) E_NOINGS (E):
2) PREFIXES (P):
X " 1/VS.Me "E#;
I'
#p>
X " V " F (s)
3) SUFFIXES (S):
I
l
E :> =E
- p - V " X ;
--
X " v " F "_S -
(s)
E#
#
I " V " x1
P => (-)P>
S :> /S(-)
where I : ( a d m i s s i b l e ) i n i t i a l c l u s t e r , F = f i n a l c l u s t e r , M = mor-he-m-eTnternal c l u s t e r , V = v o w e l , (s) t h e
" g l u o n " S ( c f . TID~INGSMA~),
# = word boundary, ( = , > , / , - ) = e a r l i e r accepted a f f i x segmentations, and
, finallay,
d e n o t e s o r d i n a r y c o n c a t e n a t i o n . ( I t i s t h e enhanced e l e ment in each p a t t e r n t h a t is t e s t e d f o r i t s s e g m e n t a b i l i t y ) .
B A G G ' E = v D R O G v . R E P = E T S L I N G R = A D E MELLAN STEN=AR , F O R > B I
TALLSTAMM AR , MELLAN ROD*A LINGONTUV=OR e l e
GRON I N > F A T T / N I N G .
qETTq STORT FORE>M~L hHADEh uRORTu eP~e SIG BORT'A eIe
SLANT=EN
• FORE>M~L=ET
NARM=ADE
SIG HOTFULL'T
dDETd KNASTR=
= A D E eIe S K O G = E N
.
- SPRING
BAGG'E SLAPP=TE kOCHk vSPRANGv
. rDEr L~NG'A KJOL=ARNA
VIRVI=ADE
eOVERe 0<PLOCK=ADE
LINGONTUV=OR
, BAGG'E KVINNO=RNA
hHADEh STRUMPEBAND
FOR>FARDIG=ADE
eAVe SOCKERTOPPSSNOR=EN
,
KNUT=NA
NEDAN>FOR
KNAN'A
a F O R S T a b U P P E b e P ~ e q E N q kS V ~ G = A D E K V I N N O = R N A
STANN'A
.
rDEr vSTODv kOCHk STRACK=TE
eP~e HALS=ARNA
. qENq FRAN
UT>DUNST/NING
eAVe SKRACK SIPPR=ADE
bFRAMb . rDEr vHOLLv
BE>SVARJ/ANDE
HAND=ERNA
FRAM>FOR
SIN'A SKOT=EN
- dDETd vSERv STORT kOCHk eRUNTe bUTb , vSAv dDENd KORT~A
eOMe FORE>MAL=ET
dDETd vARy aVALa alNTEa qN~GOTq IN>UT>I ?
- dDETd gKANg LIGG'A qENq KARL IN>UT>I ? dDETd vVETv rMANr
aVALa kVADk rHANr vGORv eMEDe OSS
- r J A G r T Y C K = T E d D E T d R O R = D E e P ~ e SEJ
gSKAg rVlr iV~GAI
VIPP=EN
?
- JA ? E S K A g rVlr i V ~ G A I V I P P ~ E N
?
BAGGE vSMOGv SIG eP~e GLAPP'A KNAN UT>F~R BRANT=EN
• kNARk
rDEr NARM=ADE
SIG rDEr FLAT=ADE
POTATISKORG=ARNA
eMEDe LINGON
kSOMk vSTODv eP~e LUT eVIDe VARSIN TUVA , vVARv rDEr aREDANa
UT>OM SIG eAVe SKRACK . oDERASo SANS vVARv BORT'A .
PASS eP~e . rVlr KANHAND'A
alNTEa vTORSv NARM=ARE
? vSAv
dDENd MAGR'A RUSTRUN
rVlr E K A N g a l N T E a G~ H E M e M E D e S K A M M = E N
aHELLERa
• rVlr
g M ~ S T E E a J U a iHAi B A R K O R G = A R N A
eMEDe .
- JAVISST
, BARKORG=ARNA
kMENk kNARk rDEr uKOMMITu
bNERb eTILLe STALL=ET
I<GEN
uVARTu rDEr NYFIK=NA
rDEr vDROGSv eTILLe FORE>M~L=ET
ele
-
-
72
Iii. 2: NOMFRAS - FS-DIAGRAM FOR SWEDISH NOUN PHRASE PARSING
quant +
dec
+
"OWN"
+
adJ
+
noun
IDETTA~
OENNAL__
-ERI-NI-~I
/j MI-T
A
LL"A~ ~
B~DA DEN
ER) "NAI-EN]
- PYTT , vSAv nDEN+L~NGAn
k V A D k v V A R v NU n D E T + D A R n
nDET+OMF~NGSRIKA+,+SIDENLATTA+TYGETn
nDEn GJORDE nEN+STOR+PACKEn
eMEDe SIG SJALVA eOMe kATTk nDET+HELAn
.. n D E T + N E L A n
alNTEa uVARITu
nETT+DUGGn
nDET+FORMENTA+KLADSTRECKETn
.. G R O N e M E D e H A N G B J O R K A R
kSOMk nALLAn
.. M O D E R N
, nDEN+L~NGA+EGNAHEMSHUSTRUNn
STORA BOKSTAVER
nETT+SVENSKT+FIRMANAMNn
eP~e nDEN+ANDRA+,+FR~NVANDAn
nDETn vVARv nEN+LUFTENS+SPILLFRUKTn
kOCHk nDEN+ANDRA+EGNAHEMSHUSTRUNS+OGONn
nETT+STORT+MOSSIGT+BERGn
. • SIG eMOTe SKYN eMEDe nEN+DISIG+M~NEn
eVIDe nDET+STALLEn
SAGA HONOM kATTk nALLA+DESSA+FOREMALn
..ARNA kSOMk nEN+AVIGT+SKRUBBANDE+HANDn
kSOMk nEN+OFORMLIG+MASSAn
- nEN÷RIKTIG+BALLONGn
• *nDETn alNTEa
vL~Gv nN~GON+KROPP+GOMDn
• ** T V ~ k S O M k B A R G A D E ~ D E N + T I L L S A M M A N S n
73
kATTk
VARA
RADD
eFORe
?
eAVe dDETd
.
alNTEa uVARITu
qETTq
..
FARLIGT
.
vVARv kD~k SNOTT FLE..
FYLLDE FUNKTIONER
.
kSOMk uVARITu
ele SKO..
, vSTODv ORDEN
..
kSOMk hHADEh
uRAMLAT..
VATTNADES
eAVe OMSOM
HOJDE SIG eMOTe SKYN..
kSOMk qENq RUND LYKTA
..
kDARk LANDNINGSLINAN
..
aAND~a alNTEa
FORMED..
.
VALTRADE
SIG BALLONG..
gSKAg VARA FYLLD eMEDe..
INUNDER
.