Teaching the Computer Swedish: Morphology and Phonology

Teaching the Computer Swedish: Morphology and Phonology Gerard M. Dalgish Stockholms Universitet ABSTRACT: This paper examines the role the computer can play in constructing, from rules, derived forms of Swedish morphology and a characterization of Swedish pronunciation. It describes computer rules that mimic generative morphophonemic rules that produce nominals, adjectives, and verbals, and phonological rules that represent their pronunciations. I will also discuss how such rules can be implemented in a context of Computer-Assisted Language Learning. A corollary of this discussion is the degree to which the computer and the student need to cooperate within such activities. 1. Background: This paper will discuss how the computer can mimic various aspects of linguistic knowledge, and how these aspects can be made applicable to the language learning context. In the absence of complete reliability for tools like mainframe-level semantic/syntactic parsing or computer-assisted translation techniques, and given the current constraints on availability of such tools in language learning contexts, this paper will focus instead on "low budget" methods of characterizing linguistic processes, and on examining how these can be made applicable to ordinary Computer-Assisted Language Learning (CALL) situations such as microcomputer classroom contexts. The concepts that are relevant for this discussion have been previously established in the literature: (1) the role of the computer as pedagogue or slave; and (2) open-endedness or creativity. The first of these needs no introduction and little discussion.1 The second concept is not so well known, and therefore a brief background of some of my own earlier attempts in this area for English as a Second Language (ESL) CALL will be given below, before a discussion of related points in Swedish. 2. Open-ended English Processes: A few examples of computer characterizations of English morphological-level phenomena will serve as a good background to further discussion. One case is plural formation in English. A native speaker of English has the linguistic ability to create the correct plural for any given noun in English, even for a new, previously unencountered word. It is this ability that I call "open-endedness." A computer program can mimic this ability, and use its created construct within a CALL lesson. The lesson proceeds as follows: the computer requests a singular noun from the user, the user inputs one, the

CALICO Journal, Volume 7 Number 4

55

computer analyzes it and derives a plural, but before displaying it asks the user for his/her guess as to the plural. The user's answer is compared to the computer's construct, and the user is informed of the match or lack thereof. The computer then provides an explanation of its answer by displaying to the user a description of the rule(s) it used to derive the plural form. A related activity is for the user to "query" the computer about the plural of a noun, without a test; the computer simply provides the answer and the explanation, if the user desires it. In both cases it is the computer that is the workhorse for the user, and in both cases the same rules underlie the derivation or creation of the plural form. It is simply the application of what the computer "knows" that differs in the two cases. Let us examine a little more closely the sorts of rules and processes used by native speakers and "knowledgeable" computers to produce the correct plural form. Several "checks" must be made: Is the input form an irregular noun (e.g., "child")? Is it a noun that cannot have a plural (e.g., "furniture")? Is it a noun that has two possible plurals (e.g., "fish")? Is it already a plural (e.g., "books")? To answer these questions correctly, the computer—like the native speaker—needs access to a list of irregular nouns, a list of so-called noncount nouns, a list of double-plural nouns, and a check to see if the noun ends in "s" (whereupon the computer asks, "Is your noun already a plural?"). If the answer to all these checks and the question is, "No," then the string is examined by a series of program statements that identify the final elements of the string and determine which of the regular plural formation rules apply. Note that the computer may have to "depend" on the user's knowledge for help in this procedure: in the above scenario, the question "Is your noun already a plural?" must be overtly asked of the user; it requires a truthful and knowledgeable response in order to distinguish between a genuine singular sfinal noun like "class" and a "bogus" noun the user might accidentally input like "books."2 The degree to which the knowledgeable computer depends on the cooperation of the user is an important factor in this discussion. Other morphological categories in English compatible with this concept of open-endedness are the formation of past tense and progressive ("ing") verb forms and the determination of the choice between a or an before vowel-sound initial words.3 3. Swedish CALL and Morphological Formations: Certain aspects of Swedish adjective, verbal and nominal morphology can be handled in much the same way as in English morphology, and the computer characterizations of these word formation processes can be incorporated into CALL for Swedish. The outline of the instructional modules for all these categories is this: the user starts from a menu requesting a choice among (a) working on vocabulary and/or formations from closed lists of adjectives, nouns, or verbs (regular and irregular);4 (b)


56

searching for a particular English or Swedish word; (c) typing ANY Swedish word and having the computer produce the appropriate forms; and (d) typing ANY Swedish word and having the computer produce a phonetic representation of its pronunciation. For choices (a)-(c) the user has the option of being tested on the form(s), or simply having the forms printed on the screen. The programs contain other features (testing and scorekeeping, random functions, explanatory modules, printouts, practice typing) that are not the focus of this paper. For choices (a)-(c), some of the same computer procedures are used. If the user chooses (a) in noun formations and then selects, say, a regular noun, the computer randomly selects a regular noun from a list containing Swedish base forms and the English glosses. The program then proceeds to derive the other morphologically related forms (plurals and others, cf. below); it does not store these other forms. If a word is searched for (choice (b)) and found, the same derivations take place. And in the case of a user typing in any word (choice (c)), the computer analyzes the string and proceeds in much the same way. The latter case then is simply the most open-ended example in the continuum, and is the focus of the following discussions. In what follows, I will briefly outline the morphological formation processes for adjectives, verbs, and nouns and discuss the computer characterizations. Adjective Formation: Adjectives in Swedish are inflected to agree with the noun they modify in gender, number, and definiteness. There are two genders, "common" gender and neuter; number can be singular or plural; and an adjective can be definite or indefinite. The dictionary or base form is the common, singular, indefinite form, from which the neuter, plural and definite forms are derived. It happens that the plural and definite forms are the same in adjective formation. (1) Common grön ny mogen svart söt vild röd tunn

Neuter grönt nytt moget svart Sött vilt rött tunt

Plural & Definite gröna nya rnogna svarta söta vilda röda tunna

English green new ripe black sweet wild red thin

The computer program derives the neuter and plural forms in much the same way that typical morphophonemic analyses might proceed. The first check is for whether the common form is an irregular adjective; irregulars are listed in a lexicon. If the form is not irregular, the program then determines the number of


57

syllables in the common form. This check serves to differentiate between monosyllabic n-final roots like gr6n and polysyllabic n-final roots like mogen for plural and definite formation. The neuter form derivations of svart-svart, and sötsött show that roots ending in a consonant plus t behave differently from those ending in a vowel plus t. The computer thus checks a form ending in t to see if the element preceding t is in its inventory of vowels, and separates the forms accordingly. The neuter derivations of words like vilt, rött and tunt are similar to svart and söt; they are derived the same way. Incorporating these procedures into a CALL lesson on adjective formation with the modules (a)-(c) described above provides the means of mimicking linguistics open-endedness. For module choice (a), the list of regular adjectives with English glosses need only contain the base form: other forms will be derived by the computer and shown to the user (or momentarily withheld if the user wishes to be tested). When the user wishes to search for a form from these lists (choice (b)), its derived forms are produced by rule. And if, as in choice (c), the user types a "new" adjective (one not on the list of irregular adjectives), the program "creates" the correct neuter and plural/definite forms from the user's base form. Verbal Formation: Swedish verbal morphology involves the derivation of infinitival, present, past (or imperfect), supine (used in compound perfect tenses) and past participle forms. The imperative form is the root in most cases, and the base from which the others are derived. There are four conjugations: the first three are regular; the fourth (the "strong" verbs class) is irregular in past and other formations, and so must be memorized. Representatives of the regular verbs are given in (2): (2)

Root: Inf kalla

kalla Pres. kallar

Eng.: "call" (class 1) Past Supine Partc. kallade kallat kallad

Root: Inf bôja

böj Pres. böjer

Eng.: "bend" (class 2) Past Supine Partc. böjde böjt böjd

Root: Inf köpa

köp Pres. köper

Eng.: "buy" (class 2) Past Supine Partc. köpte köpt köpt

Root: Inf bo

bo Pres. bor

Eng.: "dwell" (class 3) Past Supine Partc. bodde bott bodd

The infinitive form is simply the root plus a; the a is deleted if another


58

vowel precedes. The present tense form is simply the root plus -er; again a vowel is deleted if there is another vowel preceding. Given the imperative form, the computer can easily derive these two forms. The past, supine and past participle forms are more difficult and need some discussion. For the moment, let us assume the computer is able to determine the conjugation of the verb (we will discuss how it "knows" this shortly). The past tense form for conjugation 1 is the root plus -de; for conjugation 3, the root plus -dde. Conjugation 2 has two sub-groups of roots: one ends in voiced consonants (e.g., -j) and takes -de in the past; the other in voiceless consonants (e.g., -p) and takes -te in the past. Supine forms are the root plus t except in Conjugation 3, where they are root plus tt. Participle forms are the root plus d except for voiceless-final-roots of conjugation 2, which are root plus t, and for conjugation 3, which are root plus dd. In this analysis, the native speaker, the learner, and, in some sense, the computer program, need to know the particular conjugations of the verb in order to derive the other forms correctly. How can they do that? Consider first the case involving a list or lexicon of regular verbs and English translations, as in lesson module choice (a). If this list were to include the conjugation number, that would solve the problem. But this "solution" is clearly undesirable if we wish to capture any notion of open-endedness, and it will fail to derive "new" regular verbs that the user might type in, as in choice (c) of the module discussion above.5 The better solution is to force the computer to determine from the root which conjugation the verb belongs to. Conjugation 1 verbs end in the vowel a; conjugation 2 verbs end in a consonant; conjugation 3 verbs end in a vowel other than a. The computer can then "make a guess" as to the conjugation, assign the verb to that conjugation, and cycle the form through the regular rules that, as they are currently written, do refer to conjugation numbers which in turn are used as pedagogical aids and in explanations. The result is that any verb that the user inputs can be analyzed, assigned a conjugation, and used as a base to form the correct derivations. In the modular CALL lesson described above, in choice (a) the list of regular verbs need not include reference to conjugation. Nor will it be necessary for the user to know the conjugation should s/he select choice (c), to "create" the derived forms from the imperative form s/he types in. In short, the program mimics the ability of the native speaker to analyze base forms, assign a "novel" verb to a conjugation, and to produce derived forms. Nominal Formations: Nouns in Swedish are inflected like adjectives, for gender (common and neuter), number (singular and plural), and definiteness (definite and indefinite); nouns are further grouped among five classes according to how they form (indefinite) plurals. Unlike adjectives, there are separate forms for indefinite and definite plurals.6 The singular indefinite form is assumed to be the


59

base form. Class 1 nouns are common gender, most ending in a in the base form, with the indefinite plural formed by adding the suffix -or. The definite singular form takes n, a reduced form of the definite common suffix -en. (3)

Class 1 "girl" (common gender)

Indefinite flicka

Singular Definite flickan

Plural Indefinite Definite flickor flickorna

Class 2 nouns are common gender; many base forms are monosyllabic. Polysyllabic base forms often end in unstressed -e, -el, -en or -er; nouns ending in the nominal suffixes -dom, -ling, -ning and -ing are also found in this class. The plural indefinite is formed with the suffix -ar. Plurals of base forms that end in unstressed -e, -el, -en or -er show loss of the root -e before the plural endings. Vowel-final roots and -el, -en and -er finals roots take -n, a reduced form of the definite singular common suffix. (4)

Class 2 "horse" (common gender) Singular Indefinite häst

Plural Definite Indefinite hästen hästar "feather" (common gender)

Singular Indefinite fjäder

Plural Definite Indefinite fjädern fjädrar "lake" (common gender)

Singular Indefinite sjö

Definite hästarna Definite fjädrarna Plural

Definite sjön

Indefinite sjöar

Definite sjöarna

Class 3 nouns can be common gender or neuter. The base forms of common nouns are (1) monosyllabic consonant-final; (2) polysyllabic loan words (often from French) with "unusual" stress at the end; (3) polysyllabic words with certain suffixes (-or, -er, -arie) and others. The base forms of neuter nouns are typically polysyllabic and end with the suffixes -eri, -eum, and -ium. The plural indefinite form has the suffix -er; there are complications in the definite singular and indefinite plural forms for nouns ending in -or or -eum.


60

(5)

Class 3 "color" (common gender)

Indefinite färj

Singular Definite färjen

Indefinite armé

Singular Definite armén

Indefinite konstnär

Singular Definite konstnärn

Indefinite museum

Singular Definite museet

Indefinite färjer "army" (common gender) Indefinite arméer "artist" (common gender) Indefinite konstnärer "museum" (neuter gender) Indefinite museer

Plural Definite färjerna Plural Definite arméerna Plural Definite konstnärerna Plural Definite museerna

Class 4 nouns are neuter vowel-final; plural indefinite forms end in -n. (6)

Class 4 "apple" (neuter gender)

Indefinite äpple

Singular Definite äpplet

Indefinite knä

Singular Definite knäet

Indefinite äpplen "knee" (neuter gender) Indefinite knän

Plural Definite äpplena Plural Definite knäna

Class 5 nouns are neuter or common. Most neuter nouns are monosyllabic consonant-final. Common nouns end in the suffixes -are, -ande, -iker and -ier. The indefinite plural is the same as the singular. The definite plural takes the suffix en for neuter, -na for common. (7)

Class 5 "child" (neuter gender)

Indefinite barn

Singular Definite barnet

Indefinite lärare

Singular Definite läraren

Indefinite barn "teacher" (common gender) Indefinite lärare

Plural Definite barnen Plural Definite lärarna

As in the case of verbs, the native speaker, the user of the program, and the program itself, must "know" certain information in order to derive definite and plural forms for nouns. To produce the correct definite forms, the gender of


61

the noun must be known, since the definite endings are -(e)n for common nouns and -(e)t for neuter nouns. To form the plurals, the gender and the noun class must be known. With a closed list of items, as is the case for choice (a) of the instructional modules described above, one solution would be to list regular nouns with the gender and noun class number they take. But for modular choice (c) we again have the difficulty of a "novel" or previously unencountered noun for which the user, the program and perhaps even the native speaker, cannot "know" the noun class and the gender. Somehow the program must determine this information for any noun. The best point of attack is gender, because both definite formation and plural formation must have access to it. It turns out that for many nouns, if we know the gender we can successfully figure the noun class number, and therefore determine plural formation. Gender is arbitrary to some extent, but even beginner's books list gender relatively accessibly and far more readily than noun class. So for modular choice (c) the computer also requests the gender of any noun the user inputs. With "knowledge" of the gender, the program then figures the noun class. If the noun the user inputs is neuter, for example, it will be in classes 3, 4, or 5. If it is a neuter consonant-final root, it will be in class 5, unless it ends in -ium or eum, where it will be class 3. If it is a vowel-final neuter noun, it will be in class 4, unless it ends in -eri or -ori, when it will be in class 3. So, with information about a noun's gender, the computer can "guess" which class the noun goes in, and construct the definite and plural formations. Even with information about gender, the case is much more complicated for common gender nouns. Most polysyllabic common nouns can be correctly assigned a noun class because of their endings: a-final polysyllabic common nouns are in class 1, most others fall into classes depending on their suffixes. Monosyllabic vowel-final common nouns are most often in class 2. But a large number of monosyllabic and some polysyllabic roots present difficulties. For instance, a number of polysyllabic and monosyllabic loan words are in class 3: soldat ("soldier"), and sallad ("salad") are stressed on the final vowel, and the monosyllabics page ("page"), and chans ("chance") are pronounced to sound like French, marking them as loans. For such words, the program is forced to ask the user if there is unusual stress or "foreign" sound in the word; if the user answer is, "Yes," then the computer's "guess" is that the word is in class 3. If the answer is, "No," and if the word is "possibly" in class 2, since the majority of monosyllabic common nouns are class 2. After this point (for polysyllabic common nouns that have not been captured by any of these tests), the computer concedes that it simply cannot determine the class of the noun. Note that as in English plural formation, the program depends on a felicitous response from the user, in this case to the questions about gender and


62

loan words. Presumably the user is motivated to get the right form and will answer as best s/he can.7 Compounding: Another check these programs must make is for the presence of prefixes and suffixes in compound formations. The program's ability to analyze a compound has implications for both the morphological processes described above and the phonological ones to follow. Compounding influences the morphological formations described so far. Recall that for choice (c) of the modules (typing a "novel" item) the first step is always to determine if the user's item is irregular (just as with the English plural formation procedures). The user's item is checked against a list of irregulars (adjective, verb, or noun), and the user is informed if a match is found. An additional check must also be made, however: the program must determine if the user's item contains part of an irregular word. This is because it is impossible for a list of irregulars (nouns, verbs or adjectives) to also include every conceivable compound word that is based on an irregular form. Yet compound words behave like the noncompounded, or simpler forms, precisely in the areas of morphological formation we have been discussing. One case is verbal formations. Assume that the user inputs an irregular verb skriv ("write") for the program to analyze. The program checks its list of irregular verbs, finds skriv among them, and presents it to the user as irregular (displays its forms, tests the user, etc.) If the user's next item is "underskriv ("sign, give a signature"), the program checks the user's input to see if the leftmost part of the item matches any of a set of common verbal prefixes in Swedish, like under, -over, etc. If there is such a match, then that prefix (under in this case) is momentarily stripped off, and the remainder (skriv in this case) is sent to the list of irregular verbs, where it will be matched. The user is informed that his/her item underskriv "seems to consist of a prefix under and an irregular verb skriv, and again the parts are displayed.8 A similar test must next be performed for compounds in which the first element is not a productive prefix in the language. An example for nouns would be the case of ombudsman/ombudsman ("ombudsman/ombudsmen"), which behaves exactly like the irregular simple noun man/man ("Man/men"), but would not be listed in a lexicon of irregular nouns.9 Here, the search scans the rightmost part of the user's item and compares it to the irregular singular list for a match. When "man" is matched, the user is informed that his/her item "seems to be a compound noun consisting of the (undefined) element ombuds plus the irregular noun 'man'."10 This procedure of stripping the user's item in search of prefixes and suffixes is also important in producing phonetic representations of the pronunciation of the item, as we shall discuss next. To summarize, with help from the user the computer program can go a long way toward the ability of a native speaker to derive a variety of


63

morphological formations. Lists of irregulars and common compounding elements must also be employed in the correct analysis of basic forms, but of paramount importance is the felicitous input of the user whenever a formation depends on arbitrary or nonpredictable information. 4. Swedish Phonological Processes: Swedish CALL must also focus on pronunciation. One method is to provide phonetic representations of the pronunciation of certain items encountered in the lessons: singular and plural nominal forms; present, past and supine verbal forms; etc. In addition the user might want to know how an item—any item—is pronounced in Swedish; this corresponds to choice (d) of the modular lesson schema discusses above. The user can input any words and have the computer display its phonetic representation.11 For the most part, the orthographical representation of Swedish words can be related to a phonetic one. The program contains arrays for orthographical vowels, long vowels and short vowels. The word is checked for the number of syllables, and then for the presence of various common prefixes and suffixes, which are stripped off and assigned their own (invariable) pronunciations, leaving the root form. This root is then scanned for various orthographical combinations that have special pronunciation (Swedish -hj, -dj, and -lj are pronounced like English y for instance) and for other such processes. The program then determines the vowel length for each vowel by checking for a lengthening environment (a vowel is long in an open syllable) and substituting a long vowel; otherwise the vowel is assumed to be short. Final -e, -el, and en in polysyllabic words have the -e reduced to "schwa." Since palatalization of -k to sh and -g to -y occurs only before front vowels—one of which is e—palatalization must wait until reduction of -e has taken place; this guarantees that a word like sekel ("century") is pronounced with k, and not sh. Prefixes and suffixes must be analyzed separately from roots in phonology, just as in morphology. Consider the agentive suffix -nar that appears with the word galde ("debt") in the simple compound galdenar ("debtor"). The e in galde and in galdenar is actually pronounced as reduced "schwa," but the computer would find e in a lengthening environment and so would incorrectly lengthen it. Stripping off -nar first will result in correctly analyzing the remaining root galde- in exactly the same way as the word galde, with reduced final e. Stripping off prefixed and suffixes from words also guarantees that the syllable count of the remaining root is correct; syllable count can determine vowel length. The word dag ("day") has a long vowel a because the word is monosyllabic with a final consonant. In the compound middag ("midday"), the vowel a is pronounced just as in dag. If mid- is not stripped off, dag seems to be in a "polysyllabic" environment, and so not lengthened, hence incorrectly short. If mid is stripped off and the syllable count is reduced, the root dag—in middag will


64

be correctly analyzed as monosyllabic and a will be lengthened just as in the word dag. In summary, an accurate phonetic representation can be rendered that provides the user at the very least with a reminder of some of the differences between orthography and pronunciation. The role of prefixes and suffixes is important in determining vowel length, syllable count, and other phonological phenomena. Conclusion: This paper has outlined methods of incorporating the concept of linguistic open-endedness in certain applications for Swedish CALL, specifically by mimicking aspects of linguistic knowledge and by enlisting that knowledge to aid the student/user in the language learning environment. Similar applications to other languages should be obvious. We have seen how the program depend on felicitous user input and knowledge that is beyond the scope of the program itself. In a real sense, the computer and the user are cooperating in trying to solve the puzzles of morphological formations and phonological representation of pronunciation. Acknowledgment A nine-month Fulbright teaching/research grant made this paper possible. My thanks to colleagues and staff at Stockholm University for their assistance: Professor Magnus Ljung, Nils-Lennart Johannesson, Anna-Brita Stenstrbm, Ingegerd Folcker, Dave Minugh, Ingrid Almqvist and Gunnel Melchers. Notes 1

See Higgins, Johnson, Pennington and Dalgish (1985) for further discussion of this point. A similar proposal for dealing with the morphophonemics of an obscure Bantu language is found in Dalgish (1986). 2 It seems to me that eliciting user cooperation with the computer is pedagogically desirable even if competing methods without such cooperation are available, and even in spite of the risks of obtaining user responses that are either infelicitous or simply honestly wrong. Since a record of the computer/user transaction can be kept, the teacher and user can later review what went wrong and clarify for the user what s/he might have misunderstanding. In both cases, a learning experience well worth the trouble. 3 See Dalgish (1987a) for a discussion. 4 Irregular nouns, adjectives, and verbs must be listed in a file with the forms that cannot be predictably derived; the case is identical to English irregular nouns and verbs. For vocabulary testing, some regular items with English glosses must also appear in files. 5 It might also be unreasonable to expect the user typing in the verb to know which conjugation the verb is in. 6 The simplest cases are discussed first; exceptions and complications will be added later. The indefinite articles en (for common nouns) and ett (for neuter nouns) are not discussed here; in some analyses, it is these elements that are used as suffixes in the formation of definite nouns. 7 Presumably, only reviewers intent on crashing programs have any genuine interest in being infelicitous. A hard copy of the computer proceedings could be examined by the user and teacher/native speaker expert together to determine what, if anything, went wrong. 8 Note the tone of the computer's message: a modest claim about its analysis. This is in keeping with the role of the computer as slave and not as omniscient magister.


65

9 The element ombuds cannot be said to have the same degree of productivity as a prefix like under, and so ombuds is not listed among the set of prefixes to be searched. 10 In certain cases, the inability to recognize regular compounds might well confuse the computer. If the user's items is a compound that consists of a productive or common prefix like under plus a fictional but possible monosyllabic vowel -final common noun--which should thus be analyzed as a class 2 noun--the computer will first strip off the prefix under and then correctly analyze the remaining string as a monosyllabic, vowel-final--and therefore class 2--noun. But if the user's item is a compound that consists of some other "prefix", like ombuds, plus the same monosyllabic vowel-final common--and therefore class 2 noun, the program has no way of knowing that ombuds should be stripped off, and so will incorrectly count the words as a polysyllabic words and possibly analyze it as being ambiguous between class 2 and class 3. There seems to be no principled way of resolving this problem. 11 At this stage of development, a phrase or sentence will be analyzed as a group of individual words, not as a unit. This study will not include stress or tone in Swedish, an admittedly important exclusion. The pronunciation of orthographic o is non-predictable: e.g., Swedish "kort" can be pronounced (approximately) as [kôrt' "Short" or [kûrt] 'card".

Bibliography Dalgish, G. 1987(a). "Utilizing Randomization and Open-endedness in ESL Software." Journal of Computer-Based Instruction, Spring 1987, Vol. 14, No. 2, p. 73-77. Dalgish, C. 1987(b). Some Uses of Computers in Teaching English as a Second Language, The Issue of Control. In The Computer in Reading and Language Arts, Haworth Press, 81-93. Dalgish, G. 1986. "CALL (Computer-Assisted Language Learning) for Uncommonly Taught African Languages; Computer Characterizations." CALICO (Computer-assisted Language Instruction Consortium) Journal, Vol. 4, No. I September 1986, pp 27-38 Dalgish, G. 1985. Computer-assisted ESL Research and Courseware Development Computers and Composition, 2(4), 45-62 Göransson U., and H. Lindholm. 1981. Nybörjarsvenska. Kristianstad Boktryckeri, Kristianstad Higgins, J. and T. Johns. 1984. Computers in Language Learning, AddisonWesley Johnson, D. 1985. Using Computers to promote the development of English as a Second Language. A report to the Carnegie Foundation. The Carnegie Foundation: New York. McClean, R. 1969. Teach Yourself Swedish, Random House, New York. Pennington, M. (ed) 1989. Teaching Languages with Computers, Athelstan Svenska Akademiens Ordlista över Svenska Spraket, 1986, Nörstedts Förlag, Stockholm.


66