TWO-COMPONF~ TEACHING SYSTEM THAT UNDERSTANDS ~ CORRECTS MISTAKES 
Elena BORISSOVA 
Institut Russkogo yazyka imP~shkina 
Moscow 
Abstract 
This paper presents a computer-tool tea- 
ching system supplied by a language proce- 
ssor. Its aim is to correct mistakes in 
texts written by foreign students learning 
Russian as a second language. Since a text 
may include grammar mistakes, the system 
cannot use morphological analysis to fool 
extent. So one must compile a programm 
capable of finding and correcting mistakes 
without traditional means of analysis. 
To solve this problem we prepone a system 
that includes a vocabalary and rules of 
finding and re-writing words. So the pro- 
cess consists of finding word stems and 
than correcting word endings . Semantic 
and syntactic information("a model of ru- 
ling"/Mel'~uk 1974/~ necessary for that 
is written in the vocabulary of verbs 
as a frame. The slots of this frame con- 
tain semantic and morphological infor- 
mation about words that depend on this 
vord~. 
The vocabulary containing now approxi- 
mately 200 lexemes is enough for beginners 
I • Introduction 
As a rule, computer-tool teaching pro- 
grams ca~ do without language processors. 
That gives us an opportunity to use Perso- 
nal computers and other available and in~ 
expensive means. But such programs can be 
used only for several types of exercises: 
mostly those that include substitution or 
insertion of listed words and endings in a 
sentence. 
68 
USSR 
Such exercises train the student to use 
correct\]F ~rds and expressions° It is im~ 
possible to modell exercises of the so cal~ 
led communicative type / Kostemarov et ale 
1982/ that form the skills of spontaneous 
speech ( a pupil constructs some sentences 
expressing his thoughts on a definite toy 
pic). While doing such exercises one cannot 
deal with a limited list of variants as th6~ 
re exists an infinite number of them(they 
are innamerable if a pupil has mastered 
even an elementary course). That is why 
we need a language aualisis processor in 
a computer-assisted teaching program aimed 
to teach somebody t%s~ the language~ 
to ' c emm~uic ate '. 
There exists a lot of language analysis 
and synthesis programs of Russian applied 
in Automatic translation, Natural langu~ 
age dialogue systems etc. Nevertheless it 
is impossible to ase them in computer as~ 
sisted teaching programms.0n the one hand 
the majorit~ of them are oriented to the 
scientific or technological language. On 
the other - and more importantly - as a 
rule they cannot deal with texts contai- 
ning mistakes. Those systems that'under~ 
stand' a text with mistakes do net notice 
them or they correct them only printing 
them on a display/Carbonell,Hayes "198#/ 
So it is necessary to compile a pro- 
gram capable of finding and correcting 
mistakes° This problem is new for compu- 
tational linguistics. 
2~ Description of language processor 
To solve the problem we propose a two- 
compone~tt language processor that provl~des 
morphological, synt~.actic and (to some ex- 
tent) semantic analysis of a text with mis- 
takes, ~d then it synthesises correct sen- 
tences which express identical meaning to the 
analysec~ ones. This processor deals with se- 
parated ~:entences but some information 
must be used while analysing subsequent sen- 
tences of the given text (e.g. information 
about the sex of a speaker as in a Russian 
text tlds information is necessary for ag~ 
reement of a predicate in the Past and 
a subject expressed by the pronoun of 
the 1st person singular: ya chodil~I was 
going, 
The 1st ste~. The processor executes a mor- 
phological analysis of a sentence by means 
of a stem vocabulary which includes vari- 
ants of the stems of each verb (e.g.CHOD-, 
CHOJ~ of the verb CHODIT' ~to goOetc.), 
noun~ adverb, adjective, pronoun. This 
list lacludes the typical incorrect vari- 
ants oZ these stems (e.g. ZOV-, ZAV- of 
the verb ZVAT'~to cal~; a typical mistake 
is ZAVU instead of ZOVU). The first task 
of the processor is to find a verb and 
to identify it in the vocabulary. 
The 2rid ste~ . The system uses syntactic 
information of the vocabulary. Every verb 
stem is supplied with information of the 
morphological, syntactic and semantic 
features of words w~ch are ruled by the 
verb (e.g. JIT' ~to live~v chem 'place'). 
Since all the nouns in the vocabulary are 
supplied with semantic information as well 
(eog; DOM(a house~'place'), that enables 
the s~stem to find appropriate nouns for 
the verb. 
Then in accordance with the morphological 
information the system eynthesises a correct 
case form of the noun (e.g. V DOME) which 
is compared with the form written by the pu- 
pil. The difference is marked as a mistake 
that can be commented on by the list of ex- 
planations (e.g. a pupil: Jivu dom, a correct 
rect form, synthesised by the system is Ji- 
vu v dome, mistake: "a wrong case form") 
The3d step. Then the system accomplishes 
agreement of subject and predicate (e.g. 
Student sg,m~ sg.m rA student live~) ac- 
cording to the semantic information the verb 
is supplied with (e.g; JIT' 'person' STU- 
DENT 'person') and morphological informa- 
tion of the subject. The temporal and as- 
pectual characteristics of verbs depend 
on adverbs (e.g. vchera CyesterdayOpast - 
jilAtlived~past) and some other facts. Ac- 
cording to this information the system 
synthesises verbal forms and compares them 
with those written by the pupil. 
The 4th step. The agreement between adjec- 
tives and nouns is executed in the same way 
as the previously - by finding words accor~ 
ding to the semantic features (e;g; no~ 
Cnew~'thing','place'... dom~a hoase~,place ,) 
and then by changing of the forms accor- 
ding to the morphological information (e.g. 
dom m - novyi m) 
3.Some notes on system exploating 
The result of this system's work should 
be a correct text with the correction of 
mistakes. A system based on the same prin- 
ciples but more complex should correct 
some syntactic mistakes in word order, 
usage if conjunctions etc. 
If the result of the correction allows 
two possible variants of a text, the com- 
puter prints:"De you want to say"..." or 
69 
"..."? (Possible variants - in inverted 
commas). If a sentence is not admissible 
by the given system, a computer prints: 
"I do not understand you, say it another 
w~". 
The system can ask other questions as well. 
In particular, if a pupil prints a personal 
name unknown to the system it asks:"Is it 
a male or female?" and then this name is 
inserted into the vocabulary with morpho- 
logical characteristics fe~m. or mas__cc. 
Besides the grammatical information the 
vocabulary should include some encyclopaedi~ 
cal information important for a pupil. E.g. 
if a foreign pupul has come to Moscow al~ 
ready then the phrase Ja priedu v ~oskva 
~I'll come to Moscow~is wrong. In order to 
correct such mistakes one inserts into 
the morphological and syntactic infor- 
mation an inscription : PRIYEHAT' ~I ~- 
past- v Moskvurto M~scow which means that 
a prase about 1 p.sg. and Moscow is correct 
only in the past. 
Prelimenary input of proper names which 
a learner ma~ have occasion use in a 
text is desireable as well. Otherwise 
mistakes of the type Ya priehal iz kuba 
(instea~ of s Kuby) woul6 not be correc- 
ted. 
The system is intended both for testing 
compositions and dialogues. Since systems 
for advanced students would be too sophis- 
ticated and would have to include comp- 
lete information about the language, 
nowadays we restrict ourselves to a sys- 
tem for beginners (150 lexemes in the vo- 
cabulary). The system will be realised on 
the IBM-PC 
References

Mel'~uk I.A ~ 0pyt teorii lingulstlceskih• .v 
modeley 'Smysl~=~Text,. Moscow Nauka 1974 

Kostomarov V.G. ,Mitrofano~ 0 ~etodiceskoeV 
rukovodstvo dla prepodavateley russkogo 
yazyka inistrantsam. Moscow Russkiy yazyk 
1982 p. 7 

Carbonell J.G. Hayes Ph. J. Coping with 
extragrammaticality. In: COLING 8~ P.437~a43
