Dept. for Speech, Music and Hearing

Quarterly Progress andStatus Report

Bliss symbol-to-speechconversion: ”Bliss-talk

Hunnicutt, S.

journal: STL-QPSRvolume: 25number: 1year: 1984pages: 058-077

http://www.speech.kth.se/qpsr

11. SPEECH SYNTHESIS

A. BLJSS SYMKlL-TD-SPEECH CCEWESI(=N: "BLISSTW' Sheri Hunnicutt

Al=s tract "Blisstalk" is an electronic communication board on which Bliss

symbols are selected by a magnet or by scanning, and their corresponding linguistic expressions are spoken by a hilt-in speech synthesizer (or written as text) in the chosen language.

Linguistic knowledge has been applied in a variety of ways in the realization of this device. A special phrase structure grammar has been written which marks clauses and phrases, referring to parts-ofspeech information in a lexicon containing words corresponding to Bliss sym- bols. Phrase order is then inspected to determine sentence type. The speechsynthesizer incorporates rules for pronunciation and prosody. Bliss-to-speech and Bliss-t-text programs have been developed for Swed- ish, English and French.

This paper discusses the development of Blisstalk, its structure and its modes of operation. The choice of natural language grammar is motivated and differences between this grammar and "Bliss grammar" explained and exemplified. An Appendix is included which lists possi- ble verb phrase types in English and indicates their availability to Blisstalk users.

Introduction

Bliss symbols were developed by the Austrian, Karl Blitz, in the 1940's. He was deeply impressed by difficulties in communication among people who spoke different languages, or even the same language with different intentions. While in China, Blitz -- now calling himself Charles Bliss - was inspired by the Chinese ideographs to develop his own set of characters. He hoped they could be used as the basis of a system of world-wide commonality of expression and understanding (McW nald, 1980). This system was set forth in his nearly 1,000-page work, Semantography (Bliss, 1965).

In 1971, a special education teacher at the Ontario Crippled Child- ren's Centre found a description of Semantography, and obtained a copy for a symbol mmunication project which had been instituted for non- vocal pre-reading children. The project staff, with consultation from Charles Bliss, developed vocabularies and procedures for use of the symbols which are now called Blissymbols. An institute for the purpose of dwelaping the Bliss system grew out of this work, being established in 1975. bated in Toronto, Ontario, Canada, it is called the Blissym- bolics Communication Institute.

The use of Blissymbols in Sweden began in 1976 at two regional habilitation centers, one in Gothen- and one in Linkijping. In 1977, the Swedish Blissymbolics Resource Center was formed. Interest in Blissymbolics grew rapidly in all of Scandinavia, and the formation of the Nordic Bliss Communication Committee came about a year later, in 1978. According to the Swedish Institute for the Handicapped, there are about 800 children in Sweden who use Blissymbolics in some form; with many of them, it is their primary means of communication. ; ,

The groups concerned with speech synthesis and vocal aids for handicapped at the Roydl Institute of Technology in Stockblm have, for some years, been interested in implementing a "talking Bliss system." This interest has been encouraged by the Blissymbolics Communication Institute and others concerned withvoice output communication aids, V O W (McNaughton, 1980). This system was realized for Swedish in early 1981, and has since been developed for English and French (Carlson, Granstrom and l3unnicutt, 1982a). Bliss users interact with a 500-symbol Bliss board which includes a (multi-language) text-to-speech system de- veloped by our speech synthesis group (Carlson, Granstr6m and Hdcutt, 1982b). This system presently contains a formant speech synthesizer implemented on a programmed signal processing chip and a powerful mi- computer. The Bliss-to-speech program transforms the symbol string indicated by the Bliss user to the corresponding well-formed sentence. Bliss-to-text programs have been developed, as well, which perform a similar transformation to well-formed written sentences. The user may intermix Bliss symbols and spelled words to produce the spoken or writ- ten message. s ?; 1

Linguistic knowledge has been applied in a variety of ways in the realization of this device. An algorithm for producing well-formed sentences employs a lexicon for pronunciation, part-of-speech and spe- cial feature information, a grammar to mark clauses and phrases, and morphological rules to produce correct inflectional endings. Words which are spelled using the board's alphabet squares are pronounced by the speech synthesizer according to grapheme-to-phoneme rules or an accompanying "exceptions" lexicon in the chosen language. A set of phonetic rules controls the language-dependent sound inventory, adjust- ing the realization of phonemes in quality, duration and pitch according to the linguistic context. (See Fig. 1.) '* w ii ! WY

b B * $L; 1

SPEECH Fig. 1. The carrplete text-to-speecfr system w i t h Bliss rules

Structure of Blisstalk - Chart and Uxicon

Cutpsi t ion of the B l i s s Board

The B l i s s board, c a l l ed "Blisstalk" (see Fig. 2), contains 504 squares, most of which a r e l e x i c a l i t e m s , arranged according t o t h e i r part of speech. This arrangement corresponds to the standard Swedish B l i s s chart. It is possible to reprogram the board to include any of the 1400 standard symbols (only 500 f o r French) by making subst i tu t ions . One can a l s o reprogram a square with a word, such a s a name, which is not included i n the set of 1400 standard symbols. The standard Swedish cha r t contains approximately 200 nouns, 80 verbs, 70 adject ives and adverts, 50 function words, 30 commonly used referents of persons (e.g., pronouns, "boy, " "fr iend, " "v i s i to r" ) , and 10 common expressions (e.g., l1hello, "yes1') .

Another group of symbols has a syn tac t i c function. A user may indicate which tense a verb is to be realized in, and may indicate that

., - -, --., a noun is t o be realized i n the plural. , h ' . f f^ + -. " .-* . '

Fig. 2. "Blisstalk". Nouns and pronouns may also be marked pssessive. I f it is desired t o mark the part of speech of a mrd which is spelled using the board's alphabet squares, t h i s can be acccanp- lished by choosing a part-of-speech symbol a f t e r the mrd. These symbols are included in the 1400 (international) stand- ard symbols, but do not appear on the standard W i s h chart.

It is possible to reprogram the board so that any square corres- ponding to a lexical entry (one of the 1400 standard symbols) can be placed i n any space reserved for lexical entries. One may also desig- nate either 4 or 16 squares to have the same value so that larger (and fewer) symbols may be used for beginners. This design feature allows a great deal of flexibility i n specifying the format of the board. Com-

munication between the Bliss board and other parts of the text-tu-speech system is accomplished via BLISSCII codes. It is possible to connect to any type of input or output device utilizing this code.

Loudness may be controlled by knob or command sequence, as may the speed of the speech. The speed of contact of the magnet may also be regulated, using a knob.

I * . < 'Y*. . + + ' There are three different "grammars" which are associated w i t h the

use of Blissymbols. me is the so-called telegraphic style. The other two are more proper grammars, one being B l i s s syntax and the m r being natural spoken language syntax.

J i 2 ;

The telegraphic style . . , , r

The telegraphic style i s employed by many Bliss users. It i s the easiest, shortest style of Bliss communication, uinittirrg function words and paying less attention to word order. It is often sufficient i n mersation between the Bliss user and well-known persons, and allow a conversatim to progress much more rapidly for users who experience a motor handicap. This style can be duplicated i n B l i s s t a l k by set- the board i n word mode. In this mode, syntactic analysis, and there- fore, word inflections are omitted.

b C. . C - - "1 Bliss and natural spoken language syntax

L I . . . . - There are other Bliss users, krowever, who strive for grammatical

perfection i n their linguistic express ion, either via Bliss syntax or natural languae syntax. Bliss syntax, as described by its inventor, Charles Bliss, was meant to be simple. His symbol system was intended for use (mainly by adults) i n situations i n which there was no common natural language. His advice was to choose short effective symbol strings which gave the content of the message.

The main features of B l i s s syntax are the following:

1) The order of a message is Subject-Verbobject.

This ordering usually follows common natural language ordering for declarative sentences i n Swedish, English and French, but not a1 ways. For example, very common Swedish declaratives beginning w i t h an adverb or a definite pronoun cannot be expressed this way.

1g&r fick jag e t t brw. I C .'3 *

D e t kan jag inte g6ra. ~'*--i e-l r., :

And common French declaratives w i t h a pronoun as an object cannot be expressed as Subject-Verbobject.

Je le suppose. > . ... , w r , . c " .

There are, of course, exceptions to this ordering i n Ehglish also, such as topicalized sentences (The m i l k I remembered, but the bread I for- got.), ht these are not especially common declaratives.

' " * - I x .

- 2) The negative element is placed before the verb. . .It: -,- .,

In this case, a l l three natural languages differ to some extent. In Swedish declaratives , the negative element "intell is placed after the first verb form i n a main clause, and follows the Bliss syntax only i n dependent clauses . . ', . ++ -:L ;% -".

Main clause: i

: . . , . ". I . . " , i ' H a n k o m m e r inte. ! . . . ..I ;~, ,* i ,$% ,<) , .? ,:, 2 - ,

, . ,, -. J3an ska in&!--. . : .. . .. . .. -- , I .. . 4 J f > !,;?'!->*: &,qiv7... : 1 . :.." .. ??\,,;. .?L ,' ,;. ..J\ ; t . . j w . , ;r .

Deperdent clause: , - .. .. . ,... - * , .- !.. : , - - - . $ i ; , .., $-l,,,-:- I*.. .Y?.:.;.?b,,:fc.. , >

P ,. 7 - . Han sager, a t t han inte kommer. . % .,ta~b: fw. . - - . Han sager, a t t han inte ska komma. *;IL 6 .OW . A k - -- '-, ,

In English, an auxiliary verb precedes the negative element "mt*' -- a form of the verb DO is inserted if no other auxiliary is used.

The most common Rench negative declarative places a negative element m each side of the f i r s t verb form, and allows optional pronoun objects between the first and second negative elements.

-+,, . - . . ,&& :,"' t , . ' , e,.*-* !,IS ". ' t .l.:fi:Tj .:S?.l- ;,--. Je ne sais pas. t .

,.i . ..> - .,. . . .: Jene l'aipas vu. . . i 9 . . ,. !?;:.y , > f. ; CC ,.. ,: - - 3) Modifying elements are placed before the words they modify except i n the case of combinations which have an alternate grammar partially depenlent upon focus and importance.

This ordering is followed for adjectives i n English and Swedish, ht not i n Rench.

Ad jective preceding:

He saw the red house. - Han s&g det rijda huset.

Adjective following: 1 I

I1 a vu la maison rouge.

The ordering for non-sentential adverb in all three languages is depen- dent upon whether t h e modified word is verb, ad jec t ive o r another ad- verb. In English for example: ,- +. : ,a . K- ,.' ,,L

. . . . .%., 9 !. ;. Preceding adjectives and other adverb:

I ' I , #... .'

They w e r e especially hungry. . i ., ;, - fi F. : . . , ... * . They ate awfully fast. t i i - . - a . \

* 1:- ' 8 9

Following verbs: . . " a -

They a t e fast. - 6 - ' 3' '

And the same order is f a u d i n French: .. %.? d c r ,

3 - 1. -,,.. f

11s etaient iculierement aff ames . 4) The question symbol, o r anather question word, is placed f i r s t i n a question. The declarative order follows.

Swedish has no sentence type corresponding t o t h e i n i t i a l quest lon symbol sentence, since all questions not beginning with a question word begin, instead, w i t h a verb. French, lmwwer, has such a common form:

- A

Est-ce que vous avez recu une l e t t re? - I

It could be s a i d t h a t English a l s o has such a form, i n which a form of the verb DO precedes a declarative.

9 , , . . 1 ... E ti ( . Did you receive a l e t t e r? -

The main verb, of course, is no longer a past form. It is also possible i n a l l three languages to say a declarative sentence w i t h an intonation which makes it understood as a question. This is perhaps the form which best corresponds t o t h e B l i s s syntax form, even though t he r e is no i n i t i a l marking. It may also take the form of a declarative followed by a tag question.

* --. . You know what I mean? I P

- ' " < , ' 3 '

Du vet vad jag menar? , I b. +.& , . .- I I

Tu sais ce que je veux dire? ;, ,(, + t t

C $ . 2 you're caning with me, aren't you? IXI ska gs med, vad? ., .. ,-

Vous venez avec mi, n'est-ce pas?

listener may change. Frcan the listener's (or reader's) standpoint, his &erstanding may be inhibited by the presentation of stimuli mm- priate to the medium of perception. F'rom the symbol user's standpoint, he may have to listen to inferior expressions of his thoughts, a rather frustrating experience.

These considerations point to a clear need for a programmed gram- mar. Wl t what sort of grammar? Should it be a grammar which minimizes input? Such a grammar would minimize physical effort, and that is certainly desirable in many instances, Mt it would be quite restrictive in output capability.

One might, for example, have a l i s t of sentences i n which one or two words are l e f t blank, and could be fi l led i n by the user after choosing the desired sentence. Such a grammar would give access to quick correct speech and would cover certain cases. I t may be seen, however, that it is quite restricted. Consider the possibility of making a l l unspecified nouns definite so that one less symbol needs to be chosen. To get "a car," one would indicate "a" and "car;" to get "the car," one would indicate only "car." This solution appears, a t f i r s t glance, to be useful, but would, i n fact, make the unspecified noun impossible. The sentence "Bread i s the staff of life.", for example, would read "The bread is the staff of the life." That is, some sentences would sound very strange, or produce an unintended meaning.

Assume, then, that we want to provide for a range of types of simple declarative sentences. A simple declarative sentence w i t h only a determiner ard noun or prorloun subject a d object and three verb tenses would require 48 sentence types. Allowing an optional adjective i n either the subject or the object noun phrase raises the required number of sentences to 192. With so many possible sentence types to choose from, the task of finding the desired sentence type becomes much more time-consuming than select- more squares on the board, even i f special selection algorithms are employed. We see, then, that th is type of grammar would allow l i t t le flexibility i n sentence type, altlmugh the vocalxlary need not be especially limited. y 3 , , % r . i: A

Another possible avenue to flexibility i s the use of transforma- tions from one sentence type to another. I t is possible, using these transformations, to construct a variety of syntactic forms w i t h ap- proxiately the same meaning. Each syntactic form is derived from a base form using a t least one of the stipulated transformations. Beginning w i t h a base form "He sold the farmer a horse.", we could derive over twenty syntactic f o m such as "He sold the horse to a farmer.", "for a horse to be sold to a farmer by him" and "there being a horse sold to a farmer". We would, perhaps be content w i t h only one third of these possibilities, Mt, even so, seven numbers would have to be referenced for the transformational possibilities i n addition to whatever was necessary to generate the original sentence "deep structure," that is, the original base form.

I t seems better, therefore, to choose a grammar i n which most of the words and the word order are specified by the user. This method

provides maximum flexibility for a sophisticated user without requiring reference to, or a knowledge of, anything except the vocabulary and grammatical symbols on the board i tself . I n particular, the user can rely on his or her (possibly passive) knowledge of order i n the lan- guage, and need not refer to the rules of Bliss syntax.

The grammar of Blisstalk - ~

r .P -. * t i , , $1 I * r : v f J +. The option that has been m e n for Blisstalk is to follow natural

spoken language ordering. This ordering requirement promotes flexibili- ty i n use, not forcing input into prescribed sentence structures and

, permitting the m o s t natural sounding speech output. A s for less sophis-

. ticated users who may not require such flexibility, Eugene McDonald

. .* states i n his book Teaching and Using Blissymblics, . .. -ypl) - , * - 3 . 2 ' * f t - . < - . '

"When teaching sentence construction to young . z b

or mentally retarded children, the symbol instruc- .r d p - < .

tor w i l l probably find that the word order of , , , English w i l l be easier for the children to learn ,.- + I . - : and to use than the word order of Bliss syntax." ,. . I

I F. . &.'.$ *

The grammar of Blisstalk can be described as a determinate finite state phrase structure grammar. It allows unrestricted input from the Blissymbol lexicon (1400 symbols) and from the text-mpeech system for spelled words. The grammar proceeds by f irst introducing phrase mar- kers, forming the input words into a single set of noun phrases and verb phrases (i.e., no choice is made among alternate phrase structures).

I 1 - 5 , r . e , k

. , . The phrase structure gramnar - ,.;- < t I 6 t. A A .:

Noun phrases and verb phrases are initially delimited by recogni-

. tion of which words can or must not appear i n them. Noun phrases can then be further divided by recognition of ordering conventions within them into double objects, subject-object pairs, or both. These con- structions may occur in declaratives, i n questions and, i n Swedish, in adverb-initial sentences. Verb phrases are split before a marked infi- nitive (one introduced by "to" i n English, by "att" i n Swedish). A prepositional phrase is considered to be a special case of a noun phrase. -7 , ;

The grammar's success i n delimiting phrases is a direct consequence of the fact that lexical part of speech is, for the most part, predeter- m i n e d by the Blissymbol input. Ambiguity results only i f a Bliss user attempts to use a symbol i n a function other than that determined by its

, place on the board. I t would be possible to allow a user to change the part of speech of a word to one of those specifiable. This facility would be easy to specify i n English noun-verb conversion since the singular form of a noun and the infinitive form of a corresponding verb

may be the same (e.g., simple roots such as "walk" and "sleep"). Many

Order

We f ind addi t ional support f o r adopting natural spoken language order for Blisstalk i n an a r t i c le on children's acquisition of syntax, Roger Brown and Ursula Bellugi (1964) note t h a t when young children imitate an adult's utterance, they preserve the word order of the model sentences. They also report that a t 18 months, children are likely to begin constructing in t eg ra l 2-word utterances with the prosodies of normdl speech. A l l the major varieties of English simple sentences up t o a length of 10 o r 11 words a r e produced by the age of 36 months. We may assume that the sentences may nat be morphologically standard, e.g., s trong pas t tense verb forms may not a l l be present. But, given the correct word order and a grammar , many inflectional endings and function words can be supplied by the l i s tener . And i n t he case of symbol-to- speech conversion, grammatically redundant forms can be supplied by rule. This conversion takes advantage of the child's abi l i ty to supply a rather advanced concept of order i n his/her internal grammar to p m duce speech output.

There are, of course, many expressions which cannot be exparded by r u l e because of possible ambiguity. The expression "boy book," f o r example, might mean "The boy wants his book," or "Give the b y a book," and there a r e a number of o ther p o s s i b i l i t i e s as w e l l . The Brown and Bellugi study showed that the m o t h e r s of the children they reported an responded to t h e i r children's speech with expansions about 30% of the t i m e . That is, the ch i ld might say "boy book," and the mother would expand with something l i k e "we'll give the boy h i s book back." It w a s noted that the mothers' expansions, l ike the ones abwe, preserved the word order of the children's speech. W e may therefore assume t h a t i n cases of reduced o r incomplete expressions, the preservation of word order w i l l aid communication i n a natural way.

One might say that this approach is not fai thful to the principles of B l i s s y m b l i c s - that the goal of international communicatim is mt being kept i n mind. Perhaps t h i s is true. But spoken and wri t ten, as opposed t o pointed language, has other requirements. Is it not more important that the individual user of a Bliss-to-speech or a Bliss-to- text device be able to communicate naturally with those who speak his own language? This seems t o present a problem: should a user then learn two grammars, one for non-spoken Blissymbols and one for spoken Blissymbols? It is certainly possible to write "translation" programs for a limited grammar and with the limited vocahlary of ~ ~ S C 1 I - c o d e d words. I would l ike to argue, however, that even though this might be an in te res t ing pro jec t f o r multi-lingual use, it is unnecessary f o r a s ing le language. Some B l i s s users a r e taught only natural spoken lan- guage order. And those who learn B l i s s syntax a r e ce r t a in ly learning bath grammars - one for Blissymbols and one that is internalized from l i s ten ing t o speaking users of h i s own native language. I t has been noted by teachers w b e students are using synthetic speech that these students' language capabilities are greatly enhanced by using synthetic

speech. We can expect that many B l i s s users w i l l go on to learn to read and wri te t h e i r own native language. Supporting t h i s development is certainly important.

To add a further assurance that learning two grammars is not prohi- bi t ive, it can be noticed t h a t a s imi lar s i tua t ion exis t s i n the deaf community where two grammars are used, one for deaf sign 1-e among sign language users themselves, and one fo r signed language which is normally spoken, i.e., signed Swedish, signed English, etc. There a re sane similarities i n deaf sign language syntax and B l i s s syntax, eq., marking time of occurrence f i r s t and then dispensing with verb tense

, markers, leaving out articles, placement of modifiers (includirrg nega- , tion). The building of compound symbols/signs a lso has some s i m i l a r

principles. In addition, the basic symbols, arrows and baselines of Blissymbols could be likened t o the "dez", "sig" and "tab" of sign

language There are, of course, many d i s s imi la r i t i e s also, and d i f ferent

. levels of development i n various areas of the two communication systems. Since B l i s s symbols have only been used fo r a l i t t l e over ten years by non-speaking users, it is s t i l l i n an ear ly developmental stage for

, ,-; actual use. It is quite clear, I-mwever, that to intellectually rnrmal . non-speaking persons, learning two grammars fo r the same symbols o r

signs is quite feasible. .-. . . .. lrlA- I , . . -= A I . -a . - A > . > I + --, 4: t .3~; :~~. *.r,:h*.,

%

+.- t . 3 -fc: 1 I + .

f "\:2C-: .

' 2 . s . . O o n ~ l u d b =ks ,i

be3 - s. ' I ' - , xi >3C&KZ . 7 +? f t l e i i * <

', This paper has been -ed w i t h B l i s s t a l k , a speaking electronic Ellissymbol board. Blisstalk expects symbols to be chosen in the order of natural spoken language, and allows a user t o compose many well- formed sentences w i t h flexibility of expression. As a symbol b a r d with

,. f , a "voice," it makes its users heard -- communication is not s o highly . dependent upon the willingness of a "listener" to watch, and to inter-

pret the symbols for himself. A system similar to and contemporary with Blisstalk is the Sahara

11, developed in France for the French language (Emerard, Graillot, Cyne and Lucas, 1979/1980). I t a lso uses a 500-symbol B l i s s chart, and allows output i n either speech or print. Sahara 11's lexicon contains mot morphs in phcnetic or orthographic form, their lexical category a d

, reference to rules applying to conjugations. Permitted syntactic struc- tures a re defined by a "precedence grammar" t h a t controls which con- s t ruct ions can precede, follow, or be i n the same construction with other constructions. I t a lso allows co-ordination a t the level of nouns, verb, phrases (which Blisstalk does not yet accanplish) and rel- atives. Synthesis is based on ( a b u t one thousand) diphones (Emerard, 1977). : 6

References Bliss, C.K. ( 1965) : Semantography , Semantography Blissymbolics P u b lications. 2rrd Edition. Sydney, Australia.

mlson, R, Granstrom, a, and Hmnicutt, S. (1982a): "~liss Communi- catim with Speech or Text Output," Conference Wrd, 1982 IEEE-I-, Paris, France. .+ > y -

Carlson, R., Granstrom, B., and Hunnicutt, S. (1982b): "A Multi- Language Text--Speech Module, " Conference Record, 1982 IEEE-ICASSP, Paris, France.

Emerard, F., Graillot, P., Cyne. G., and Lucas, J.J. (1979/1980): "Pro- theses de Parole Destinees a la Communication des Handicaps Moteurs Deficients de la Parole, " Recherches Acous tique, VI, CNET Lannion, France.

Emerard, F. (1977) : "Synthese par diphones et traitement de la proso- die," These, Grenoble, France. 5, - i ; 1J

McDonald, E. ( 1980 ) : Teaching and Using Blissymbolics , Blissymbolics Canmunicatim Institute, Tbmnto.

McNaughton, S. ( 1980) : "Bliss ymbolics a d Voice Output Cormnunicatim Aids," Presentation at VOCA Conference, May 22-23, Berkeley, California.

Quirk, R and Greenbaum, S. (1973): A Concise Grammar of Cmtempara- ry Bqlish, Harcourt Brace Jwanovich, Inc., New York.

(4) present participle (-a fom)

(a) progressive aspect

EX: ~eisswimningat thegymtoday. . (bl in Dartici~le clauses

- * Ex: S w h n h g early, I have the gym to myself.

(5) past participle (-ed form) - ' 3 ' 4

(a) perfective aspect (HAVE + verb + -ed) - ,, .. EJC: Theboyhasdrunkthewater.

(b) passive voice (BE + verb + 4) - 1

: We were surprised.

(c) in participle clauses

Surprised 2 the n- , he dropped his tiread --- - B. Auxiliary Verb Fom

Auxiliary verb forms are used together with another verb in a verb phrase. (The verb phrase may be split by a noun phrase in a question. These forms are available for the user of Blisstalk except for the contracted negative.

(1) Primary auxiliaries: DO, HAVE, BE ., ,

(a) mn-negative -V ' - PA

Ex: Did yw guess? .\,,. - , - (b) uncontracted negative

Ex: You did mt guess. - - > ., * (c) contracted negative

* Ex: ~idn't you guess?

( 2 ) Modal auxiliaries: CAN. MUST. W I I L , WOULD, COULD, *SHOULD, etc.

- , > , "*

Ex: Must yw go nod' - ( 3 ) Marginal modal auxiliaries: NEED, *USEI *MI etc.

Ex: Theyneedtogorrow. -