J.

Linguistics (1978) r-zo Printed in Great Britain

Grammar on mathematical principles

ZELLIG HARRIS
Unioersity of Pennsylaani a

(Received zo May ry77)

This papert presents a grammatical theory which has certain mathematical

properties, and which produces the sentences of a language by means of two
simple processes, word-entry and entry-reduction, with the meaning of each
sentence being indicated directly by its construction. The reason for seeking a
rnathematical control on grammatical analysis lies partly in the inherent pos-
sibilities of such a connexion, and partly in the complex and somewhat haphazard
nature of grammar, which nrakes one wonder if there is not something simpler
and more principled involved. In the event, the system of grammatical analysis
which is reached here is not an alternative to descriptions of grarnmatical
patterns, but rather a complement to them; for in producing the sentences it
locates and explicates those patterns as resultants of the two processes, but
without cataloguing or arranging them.
The essential problem was to find one or more relations (or operations) on
the set of sentences in respect to which that set - or, as it turned out, an extension
of that set - is closed. To meet this problem we will establish certain overt
relations among English sentences, which can then be shown to hold also in
less noticeable situations. The result will be the set of grammatically possible
sentences;and it is this set which will be seen to have the desired mathematical
properties. Data for the analysis presented here are to be found in Harris (r976)
and to appear; further discussion is in Harris Q976a,b).

r. ReouctroN
First, we note certain relations among subsets of sentences, where one can say
that a word has been reduced in its phonemic shape. A simple case is that of
pronouns: we can save pronouns from being something unique in grammar, if we
describe pronominalization not as replacement of a noun by a pronoun, but as
reduction of a repeated noun to a short phonemic sequence - that which is
called 'pronoun'. The reason pronouns are thought of as separate words is that
there is a systemic degeneracy: different words, when repeated, are reduced to
the same phonemic sequence, Similarly, repeated words can be reduced to
zero ('zeroed'), as in Bach utrote cantatas and Mozart operas from Bach wrote
cantatas and Mozart u)rote operas, and in 1 ashed to be last from I ashed that I be
last, Here too different words are reduced to the same zero. 'pronoun' and
'zero' may therefore be used as verbs, like 'reduce',

[r] Given as a lecture in Somerville College, Oxford, t6 March ry77,

 ZELLIG HARRI

In certain sentence-positions, indefinite nouns such as things, people, someone

are zeroable. We can take tVe opposed huilding the monument, from trI/e opposed
people's buitding the monument, and He read for his examinatiozs from He read
things (or: tlrc releaant things) for his examinatioz.r. Assuming here the existence
and then zeroing of things is preferable to saying that read is intransitive as well
as transitive; for read is intransitive only in contexts in which an object of it
can be said to have been zeroed.
'I'here are also cases where it is clear that a particular word (or one of its
synonyms) is the indicated word in the given environment, and has been zeroed.
Thus the object of erpect is normally a sentence: I expect that John uill be late,
I expect John to leaz;e, or to come, etc. \Vhen we see I etpect ]oirl, then rather
than say that expect can also have noun as object, we say that to be here, or
to come, has been zeroed in the sentential object of expect. This, not only becausc
the meaning of I etpect John is thereby obtained, but also because nouns can be
crbjects of expect precisel,v to the extent that they can be subjects of De here, come'.
We no more sa\' *I expect time, *I expect the uniaerse, than we szy
*Time is here,
*The unicerse comes.
Also, when we compare my friend, the ambassador, with my friend, zaho is the
ambassador, and this book here with tftls booh which is here, and a correct copy
with a copy which is correct, we can say that which is, who is is zeroed, u'ith
required shifting of adjectival words to before the noun.
A more complicated case of zeroing of particular words is seen in the compound
stress. All compound worcls, At-X, aside from a few petrified cases, can be
obtained from XP'N, where the X is a noun or an adjective (including derived
ones) and P' represents certain prepositions (mostly/or, of) and certain verbs or
adjectives clepending on thc pair X, A (".g. lihe, special lo). Thus snow-sltoese
shoes for snow, repeated book-burnings+repeated burnings of books, slate-grey
wall*wall grey lihe slate*wall which is grey lihe slate. Many X, ly' pairs permit
the zeroing of a particular verb or preposition which is special to them: The
milhman is late<-The man who brings the milh is late; The milh-punch is strong<-
The punch contaning nilk is strong and presumably not from The punch coloured
like milh is strong.
Another kind of zeroing is that of 1 ask, I request To obtain a common deriva-
tion for all interrogatives, both yes-no and zaft-questions, it is best to take
Are you coming? not from You are coming but from I ask operating on You are
coming, yielcling I ask whether you are coming or not. Then Who is coming? wotid
come from I ask on the pair One is coming, Another is coming, yielding 1 asA
whether one is coming or another is coming, which by regular and-or zeroing
reduces to I askwhether one or another is coming, which reduces by zofr-pronouning
to I ash who is coming. In both forms of the questioning statements there is then
an optional change which places a question intonation on the object of asA,
dropping the whether and permuting the free subject with the following verb-
 GRAMMAR ON MATHEMATICAL PRINCIPLES

word. This yields 1 ash: Are you coming, or not?, I ash: who is coming?, and
further A re you coming? , I ash and who is coming?,1 asA. This whole development
occurs with all verbs whose object is zohether...or... (but not if there is
a
zeroed choice or the like as in t hnow whether he is coming or not, which there
is
reason to derive from 1 know the choice as to ushether he is coming or not), 8.g.,
He could not decide whether she was there or not--+He could not iecide: was she
there, or not? The connexion between question-intonation and this disiunction
is
supported by the fact that verbs lacking one lack the other: e.g. I doubt whether
he is there, but *1 doubt whether he is there or not and, *I doubt: Is he there?
The
relevance of this derivation to our present purposes is in showing that Are you
coming?, who is coming?, is obtained by zeroing of the main verb (with
its
subject) whose object the question had been, namely I ask.
A similar analysis yields Please come! from I request of you: please come! from
I request of you that you please come.
There are also examples of reduction of particular words not to zero but to
affixes. Thus, the relation of quietly to in a quiet manner, and the relation of
considerably to to a considerable degree, would be considered merely one of
paraphrase were it not that they are historically connected. The -ly is a reduction
of lic 'body, form' plus indirect case-ending when that was in a compound with
quiet, considerable, etc. Bt lic plus case-ending must also have occurred
as a
free word, not in a compound; and there it was replacedby in a manner, to
a
degree, etc. Hence quietly has today the status of a reduced variant of in quiet
a
manner. Given this understanding, we must go further and say that adverbs of
fact, such as truly, also had some preposition-noun form, such as as a true
matler, which is not merely their paraphrase but their unreduced variant; for
the free-word form of -/y with true must have been able to exist and must have
been in effect replaced by some word which can stand freely today in this
position.
A similar situation of affix-reduction and suppletion is seen in -hood. This is
historically reduced from the compound position of -had ,state, condition',
which in free-word position has been replaced by such words as srale. whereas
booh-burning is clearly from burning of books, childhood is less obviously from
the state of being a child; but it is that, not merely by paraphrase, but by historical
reduction in the suffix (and compound) position and suppletion is the free-word
position. This derivation explains how the suffix can carry its own modifiers.
Thus l1rs early childhood zras unhappy is from His earry state of being a child was
unhapp!, with early modifying state, i.e. the eventual suffix _hood.

z. AnaouNr oF INFoRMATToN
The reductions have a common property, that alr of them take place on words
which is one way or another contributed little or no information totheir sentence.
 ZELLIG HARRIS

This is the case for repeated rvords. Thus, since repetitional zeroing under
and takes place only for words which repeat a word in corresponding position,
having zero between Mozart d operas gives no less information than having
^
wrote. Similarly for pronouns, excePt for degeneracy in respect to which it is the
noun that is the antecedent. The indefinite nouns contribute no information.
All that things tells us after readis that there is something which is being read,
which we obtain anyhow if we take He read as a transitive with zeroed indefinite
object, rather than as an intransitive. The which is, who is are by far the most
likely words in their position, hence the least information-bearing: the onlv
other words there, which are not zeroed, would be which becomes or the like.
The he here tnder expect, and the various verbs zeroed for the compound, are the
most expectable, giyen their neighbours, and are inferrable fr6m the zeroing,
and even recoverable. And the prepositions zeroed for the compound are the
general ones for that relation, and very common, and hence contribute little
information.
A diflerent and interesting kind of informationalessness is seen in the case of
I ask, I the information of the word asA is contained in the
request. While
question-intonation which is imposed on its object, the subject and tense of
ash do contribute extra information. However, there is one subject and tense
which can be considered to contribute no information, and this happens in the
performative case. For a person to say 1 ash: Are you coming? is the same as for
hi..r ro say Are you coming?, because for the speaker to make the statement is for
the question to be asked. Hence I ash is zeroable. That indeed it is 1 asA and not
I asketl, He ashs, etc., which is zeroed here is clear not onl1'from the meaning
of the question form but also from its textual environment.
There are many other examples of reduction taking place where amount of
information is low. For example, a sentence which is the object ('complement')
of a verb (I promised John that I would get the booh) may zero its subject or object
if they are the same as the subject or object of the main vetb (I promised John to
get the booA). Which of these positions is the one that admits zeroing depends on
ihe main verb, in a seemingly arbitrary way. But there is reason to the choice of
position for zeroing. Thus, there is no referential zeroing under obserxe or
report, etc.: 1 obserged my slurring of the towels, I reported his seeing me. IJnder
piefer, admit,promise, the subject is zeroed if it is the same as the main subject:
I prefer to phone John, I admit phoning John, I promised John to phone him' Under
order or beg, the subject is zeroed if it is the same as the first object of the main
verb: 1 ordered John to phone them, I begged John to phone me. Under deserae,
suffer, undergo, the object is zeroed if it is the same as the main subject: John
deseraes our support, John suffered their attachs. Under defend, the object is
zeroed if it is the same as the first object of the main verb : I defend him from their
attack.
Each one of these verb sets has a property common to its members' That

+
 GRAMMAR ON MATHIMATICAL PRINCIPLES
propertv is not their having a common meaning: it is hard to say what me aning is
common to prefer and promise but not to beg or deseroe. Rather, it is the likelihoocl
that the subject, or object, under that verb be the same as the subject, or object,
of the verb itself. It is reasonable to expect that the object under they undergo,
they deseroe should be the same they (i.e. same as the main subject), or that the
subject under they admit, they promise should be the same they, and. so on. But
such higher likelihood means that when the main subject, or object, incleed
reappears in the stated other position it contributes less information there than a
different word would;ancl it is this low-information case that is zeroed.
A different example is seen in the reciprocal verbs. These verbs present
difficulties for grammar, not least in that they require and in their subject a
-
rather irregular situation. They can be obtained simply if we derive, say,
John
and Mary met from John and Mary met each other pronouned from
John met
Mary and Mary met John (which is why we do not normally have *John and his
doom met, *John and the brooh rnet from John met his doom,
John kept descending
until he met the brooh). Now, the pronouning to each other can take place with any
verb: John and Mary saw each other from John saw ilIary and Mary sazu
John.
But the zeroing of each other takes place only in verbs where given,\yzAr2 it is
most likely that also NztrlAr, so that little or no information is given there by
an occurrence of and AI2ll/l-r, which is pronouned by the each other, which can
then be zeroed.
whereas high likelihood favours reduction, low likelihood can block a reduc-
tion which would otherwise take place. Thus He farms extensiz;ely has two mean-
ings, one when reduced from to an extensiae degree, and the other when reduced
from in an extensiae manner (as against intensive farming). But He writes extensiaely
on this subject is only to an extensixe degree. It is not that one cannot say He
writes in an extensitte manner, but rather that manner is not common for the pair
a,rite, extensite and therefore is not reduced there to -1y, ercept contrastively.

3. INsEnrroN oF rNDEFrNrrI NouNS

Aside from the reductions, we must note another overt relation among sentences.
This involves the extra occurrence of indefinitc nouns such as tltat, sornething.
If we consider such sentences as
(r) What fell is a tlictionary.
(z) The dictionary is what fell.
(3) A dictionary is what I need.

we seem to have in (z) a fixed word-sequence is what inserted into a sentence;

from (z) one kind of esceptional permutation l,ields (r) and another
l,ields (3).
But English has also
 ZELLIG HARRIS

(4) -fhat which fell is a dictionary.

(5) The dictionary is that which fell.
(6) A dictionary is that which I necd.

The relativc clauses here show that (4) (6) are derived from
(7) That is a dictionary; that fell.
(8) The dictionary is that; that fell.
(9) A dictionary is that; I necd that.
justlike Adictionaryisthethingwhichfell<-Adictionaryisathing;thethingfell,
and, The book which fell is a dictionary+The booh is a dictionary;the boohJell.
Although (Z) (S) are stylistically uncomfortable, the sentences (+)-(0) imply the
existence of (ZF{S) in the grammar, for there is no other efficient wav of describ-
ing (4)-(6), and all other that which sentences. In turn, all sentences in the set of
(t) (:) are obtained from (4)-(6) sentences by a reduction of that zahich+zahat.
The permutations which would have been needed for (r), (3) are thus not needed;
they are inherent in the different orders availablc i" (Z)-(S).

4. OrHEn sHAPE cHANGES

There are very few other changes of shape in Ilnglish sentences. Almost all
permutations arise from a leftward shifting of a secondary sentence into its
primary. This happens especially if a noun or noun-like sequence in the second
is prtrnouned by wh as being a repetition of one in the first, thus creating a
relative clause: My friend left; I had told you about him goes both to My friend - I
had told you about him - left andto My friend left, whom I had told you about+My
friend, ztthom I had told you about, left. In the relative clause, uthich is, znho is (and
in some situations which, whom) are zeroable; certain residues then shift further
to before the antecedent: I'he sun, which is hright, shone--+The bright sun shone.

5. ExraNstoN BEyoND THE ATTESTED SENTENcES

We have seen certain kinds of reductions, mostly to zero, taking place on material
that contributes little information to its sentence; and we have seen, in the ls
what sentences, the insertion of indefinite nouns in a way that affects the nuance
of the sentence. We will now see that these reductions and insertions. in the same
conditions, can be used to derive the remaining sentences of English from
sentence-like formations (to be marked with t) that go beyond what is normally
said in English, and which will be characterized below as grammatically possible
rather than actual. These grammatically possible sentences are marginal to
English grammar, rather than external to it, as noted in 6 below. 'lheir im-
porta lies in this, that if we include them as descriptive (not always historical)
 GRAMMAR ON I\TATHENIATICAL PRINCIPLES
sources of the remaining English sentences, then the reductions and insertions
described above turn out to suffice for deriving all English sentences from a
subset of sentences (including daggered ones) having very simple structure.
A few major cases are the following:
Adverbs and subordinate conjunctions are obtained via zeroed which is ot a
sentence. They can be derived from predicates on the sentence, but not directly:
they are predicates on a secondary repetition of the sentence, which has formed a
relative clause. Ire addressed them quietly and He quietly addressed them areboth
from JHe addressed them, which was in a quiet manner (or: quietly)+He addressed
thern; his addressing then was in a quiet manner. Similarly, He left because they
phoned and Because thev phoned, he left are both from He left, which was because
they phoned+-He left; his leaoing was because they phoned. 'fhe various permuta-
tions of adverbs and subordinate clauses are preciselv those permitted by zeroing
of which rs. The derivation via two sentences with semicolon also explains why
both the verb and the adverb are each asserted in He spoke quietry. And it
explains why there is no *He falsely left, *He left improbably, although one can
say That he left is fake, That he left is improbable, and of course, He truly left,
He left probably. The reason is that the source would be *He left;that he left is
.false and
*I{e left; that he left is improbable, and it is the coniunction of
the two
which is rejected in each case.
English, like other languages, has various words of peculiar and apparently
rnultiple syntactic statuses. It turns out that for each such word, its various uses
can be obtained from a single source, by means of 'expectable-word' zeroing
such as has been seen above. Consider, for example, only, except, bzl. Because
of their concessive meaning, a favoured (expectable) first sentence under them is
one that contains eaeryone else (anyone e/se) and is otherwise the negation of the
second sentence under them. Thus, for but: Mary coohed, but js
John was late
possible, while Mary coohed, but John didn't (cooh) is more comfortable, and the
distinguished case is seen in Ez.teryone else coohed, but Mary did not (cooA). Now
we first consider the zeroings under only. We seem to have two syntactic statuses
for only: Q) I spoke, only John wouldn't listen (with conjunctional only) and
(z) Only John wouldn't listen. If we start, in the conjunctional type (r), with
Eoeryone listened, only John wouldn't listen, the first sentence is the favoured one
and zeroable, yielding Only John wouldn't listen, which is of tvpe (z) above.
T'he meaning of only in type (z) is precisely the sum of that in type (r) plus the
expectable sentence Eaeryone (else) listened. In (z) it seems as though only carries
this as a covert or implicit meaning, but actually this is merely the meaning of
the zeroed sentence. As to how it is that the conjunction and can occur before
the conjunction only: if we start with we argued and ezteryone listened only
John
did not listen we obtain We argued and only John did not listen,
As to except; it is an ordinary conjunction in John inztited her, except that she
 ZELLIG HARRIS

couldn't come, ItJohn agreed, except Mary did not agree, regtilar zeroing yields
as above, with
John agreed, except not Mary.It the favoured eoeryone+nol case,
notbeing said in the second sentence, the eoeryone-Sentence is not zeroable, but
the not is : from Ezteryone (else) agreed, except Mary did not aglee we have, by
repeated-word zeroing, Eoeryone agreed, except not Mary, and by expectable-
word.zeroing o^ not, Ez;eryone agreed except Mary. It is in this way that the
sentence with not comes out as a paraphrase of the one without zof. Similarly
f.or but : First, John inoited her but she couldn't come. Ftom Everyone (else) agreed
but Mary did not agree we obtain Eaeryone, but not Mary, agteed and Ezteryone
but Mary agreed with the same expectable-word zeroing of not. However, Dzl
permits also the expectable-sentence zeroing seen under only. Hence Eaeryone
(else) agreed, but Mary did not agree reduces to the rather r^re But Mary didnot
agree with stressed but in the sense of only.lt is in this way that we reach the
paraphrase between There isn't but one glass left and There is but one glass left,
both from There isn't anything (eke) left, but there is one glass left, with and
without zeroing of not. In this way too the apparently different uses of bzl as
conjunction and as synonym of only are obtained from the same source.
Another case of zeroing which goes beyond the obvious is seen when we assume
that any asseftional sentence, including such as become components of larger
ones, have azeroed I say,in a way comparable to the zeroing of I ask, I request.
E.g. I say that he is here--+I say: He is here.--+He is here, I say.--+He is here. There
are many indications that any sentence can carry a zeroable.I say: e.g. inter-
pellations as in He is uncertain, not to say confused, where the subject of say is
presumably I, zeroed from the first I say in tI say that he is uncertain, if it is not
because his car is
for me to say that he is confused. A simpler situation is F1e's home,
in the garage where because makes sense only if I say was present : I say he's home
because his car is in the galage. Various problems, such as those in direct and
indirect discourse, are simplified by this eminently possible, if uncustomary'
analysis.
A related analysis can apply to the tenses. Tenses are peculiar in that their
normal definition is metalinguistic, referring to the time of the speaking of the
sentence. However, there is also evidence that tense indicates the time-order
between verbs in the sentence - more clearly in some other languages than in
English. It is possible to obtain these and various other properties of English
tenses by starting with, for example,

tJohn expects Mary's coming; Mary's coming is after John's expecting

it'
--+tJohn expects Mary's coming' which is after John's expecting it.

Here John's expecting it is zeroabla as being a repetition:

--+John expects Mary's coming, after.

Here we can posit that after imposes will on Mary' s coming, much as 4sft imposes
 GRAMMAR ON MATHEMATICAL PRINCIPLES
question intonation on its object-sentence (in I ash: Are you coming?). Then we
obtain

John expects that Mary will come.

The tense on the main verb comes similarry from its time-orde r to the I say
on it:
tI say that John expects that N{ary will come; John's expecting that N,Iary will
come is before rny saying it.
--+tI say that John expects that N{ary will come, before my
saying it.
--+I say that John expected that Mary would come.
-+I say: John expected that Mary would come.
--+John expected that that Mary would come.

Note that the -ed on expect also attaches to verbs which had been previously
tensed
in respect to erpect: will come to wourd come, This is a roundabout way of
ob-
taining tense; but, as with the pronouns, it saves tense from being sui
generis in
grammar' by deriving it from ordinary constructions by
reductions.
-eir. of known
Like reductions, the insertion of indefinite nouns, too, exprains difficult
grammatical constructions. l-or example, there is the difference
between non_
restrictive (or 'descriptive') relative clauses, as in

A man, whom l know, phoned

and the comma-less restrictive

A man whom I know phoned;

in the restrictive form, phoning is asserted about someone identified

as a man I
know. The form and special meaning of the restrictive can be readily
obtained if
we start with

tSomeone phoned;the samc someone is a man; I know the man.

--+Someone phoned;the same someone is
a man, whom I know.
->Someone, who is a man whom I know, phoned.
where the earlier relative clause loses its comma when compressed
into a further
relative clause. Then, zeroing the indefinite someone and the
following who is,
we reach

A man whom I know phoned.

The special restrictive meaning comes from the fact that phoned does
not enter
into a grammatical construction with man wrtil after man had received
the
modifier uthom I knoz,, rn contrast, the non-restrictive case comes fronr

9
 ZELLIG HARRI

A man phoned; I know the man

-A man phoned, whom I know
+A man, whom I know, phoned,
where phoning is asserted of the man independently of my knowing him'
If we accept set, team, or the like as indefinite nouns, then the same insertion
enables us to derive andbetween nouns from andbetween sentences, and to make
the distinction betu'een, e.g., Gilbert and Sullioan u)rote operett4s and Mozart
and Beethoxen ztrote operas. We take the former from
A team wrote operettas; the team contains Gilbert and thc team cotltains
Sullivan
--+A team lvhich contains Gilbert and Sullivan wrote operettas,

where a team zahich contains (or is o.f) is indefinite-noun plus expectable-word

zeroing, hke something which is. This analysis also resolves the diliculties with
collective verbs, such gather.
^s
As a final example of a transformation, consider the passive, which was the
first to have been established. It is clear that the passive is not simply an in-
dependent rearrangement and form-change of the active, since all its ingredients
are otherwise knov'n in the grammar, in closelv related uses. The Dy plus subject
occurs in nominaiizations: The chopping of trees by settlers denuded the forest.
The -enl-etl occurs as an aflix having approximately the same semantic and
clescriptive (though not historical) relation to some such completive word as
state, as does -hood in the childhood example. This is seen in the very early
-erl on nouns (monelted, hard-headed), which descriptively can be considered to
be the same morpheme as -enl-ed, and which means'state'as in ill the state of
hati.ng money, etc. It is also seen in the development of the perfect, whose early
stage was, e.g., I hazte the boolt closed, i.e. t1 hazse the booh in the state of my
closing of it; note that I haae closed the book means that it is I who closed it, and
lacks passive connotation, so that the source is not t1 haae the booh in closed slate.
In a parallel way we have the passive: The book was found by John from t The book
was in the state of the finding of the booh by John, with o/ the book zeroed as
repetition. This derivation is equally hospitable to agent-less passives such as
A hitten is born uith its eyes closed (although nobody closes them) from fl
hitten is born in the state of its eyes' closing (or of the closing oJ its eyes). Similarly
for intransitives, where the resultants are only statives and not passive at all:
as in We are agreed from lWe are in the state of our agreeing; The sun is risen
frorn lThe sun is in the state of its rising. The completive aspect of the passive
thus does not have to be separately explained. It arises from the character of the
-enl-etl (and of its free-word variant state); and the extension of the passive to
the non-completive, as in 1t is being built is quite recent. The subject of the
passive is thus not a permuted object of the verb; rather, it is the subject of
is in a state, in reference to which the object of the verb is zeroed. This analysis
IO
 GRAMMAR ON MATHEMATICAL PRINCIPLES
does not require that state be a free-word source for -enf -ed, but only that
-enf -ed
be an operator like in a state (9 below). of course, non-periphrastic passives,
in
some other languages, would be analysed clifferentlv.

6. THn STATUS oF TIIE DAGGERED FoRMS

'Ihe daggered sentences have the property that they dilTer from attested sentences
of the language only by the same listable changes (made in the same conditions)
as hold among attested sentences. Furthermore they consist of English words
appearing in positions which those words have in the attested sentences of
English. No clear line can be drawn between them and the normal sentences;
indeed, already the sources of the that which zl sentences could have been
daggered, even though they are inescapable as being the relevant sources there.
Note also that the set of attested sentences is not well defined. Many sentences
are marginal, or used by some speakers or writers and not others. some of the
daggered sentences are grating to the ear without being excludable from
the
grammar in any principled way.
The relation of the daggered sentences to their normal resultants is not alwavs
historical. True, -hood is the reduced compouncl form of a word ,epl"ce,l by
state, and thus suppletive to it. But we cannot say this of -enl-ed. Nevertheless,
rlescriptively, the detailed conditions and meanings in which -enl-ed occurs
are
as though it were reduced from some such word as state. or rather,
-e,f -ed
retains, in its suffix position, properties that belong to a word such as state.
Just
as we saw in early childhood the word early as a modifier on -hood, so in c.s
thefreshly isolated heart (from a medical paper),
freshty, which is not an adverb
ontoisolate, appearsasadverb onthe-enf-ed:theheartfreshlyinthestateof the
isolating of it.
we therefore take these daggered sentences as constituting, together with the
normal sentences, an extended set of grammatically possible r"rl^."r.

7. Tun ENTR\. AND REDUCTIoN sysrEi\r

The extended set of grammatically possible sentences has mathematical and
linguistic properties. If we form in that set a subset
{1} consisting of only the
unreduced sentences, whether daggered or not, we find that they hurr" u ,.urr*-
parent structure, describable as follows:

In the sentences of {1} we define a relation { yi > {x} ({ y} is rater than

{X}; {X} is prior to { Y}), where {X}, { y} are sets of simple ig".,..utty aflixless)
words entering the sentence. The relation holds for
.j"ri tho*e .u.e, *h".e u
lvor! r of set { I/} is present in {1}- sentences only provided that some worcl
'Y of set {x} is present there: we say that )z requires x. Thus every {1}- sentence
containing probable must also contain a word of the set (verbs,' etc.) which

II
 ZELLIG HARRIS
*He is probable); everv
includes come,fall, large (His coming is probable, but not
of rock, etc.,
{1}- sentence containing/a// must contain a word of the set John,
but not necessarily a word of the set of come, probable, etc. (e.g. John fell, which
contains no other verb). (In the second case, we could say that roch requires
fall, rather than fall requires rock; btt/a// is morphologically and positionally
similar to probable and to the other words which have non-null requirements,
whereas rocA is not.)

The entry relation is a partial order: If Y>X, Z, tt may be that X> Z, or

z>x, or neither. This last case we see in John wears hats, where wears>John,
hats, but Johnlhats, and hatslJohn. If, for word-occurrences l', X, '4 in an
{1}- sentence, Y}X, and there is no,4 such that Y>A>X,
we say that Y in
that sentence is the operator, or next later entry, on X, and X is an argument'
or immediately prior entry, of Y; also that X was free for Y. In the linear form
of the sentence, the operator appears immediately after the first of its arguments,
as in John usears hats; if we accept some suffixes as oPerators, these appear after
in
their seconcl argument: John zDas seen by Mary the arguments of -enl-ed are
John, see, and those of see are Mary, John.

It is found that with particular ordered word sets,

each word is associated
members of which must be its inrmediately prior entries; we call these sets the
argument requirement (or entry requirement) for that word: e.g. thc requirement
foi wea, is two words of the John, ials class. To avoid infinite regress or cir-
cularity, there must be some words whose requirement is null (called primitive
arguments, ly'), some whose requirement consists only of primitive arguments
(th"re u.e called elementary operators), and some whose requirements include
operators (these are the non-elementary operators). operators are written with
subscripts to indicate their ordered arguments: elementary operators include
O, (".g. sleep in Johnslept), O,, (r.g. wear); non-elementary operators include
o, t. g. continue, probable in John's playing continued, That John is late is probable),
O,, t..g know ii John knozos that she is late), O ",(,.g'surprise in John's being late
t"iipixra me), O oo (e.g. cause in John's phoning caused my return), o,no(e'g' tell in
I iold John that she left), C)noo(e.g. attribute in I attribute her returntoJohn's
phoniig).It is found that the only word-classes needed for arguments are -l!r and
o: i.e. an argument of a word is either null, or it is any primitive argument,
or it is any operator of whatever kind. For example, there is no verb whose
object must be an elementary operator such as sleep and not a non-elementary
o.r" rrrch as know:If we can say I thinh that he slept, wecan say I think that he
krutws that she came.
Every operator forms with its arguments an {1}- sentence, and appears there
after the first one of its arguments, as in John wears hats. Every {1}- sentence is
fbrmed out of Ar and operators belonging to some ten subsets - the entry-classes
O n, O nn, O o, etc. with no other factors entering into its construction' There are
-
t2
 GRAMMAR ON MATHEMATICAL PRINCIPLES
few or no grammatical restrictions on the members of an argument set, aside
from graduated unlikelihood due to nonsensicality. Thus if wear requires two
ly', it is not necessary for the first to be animate and the second clothing. If we
consider metaphor, fairy-tales, and arrant nonsense, there is no primitive
argument i/ which can be absolutely excluded grammatically from either
argument-position.
The reductions, if they are made, are made on (or in respect to) words entering
a sentence at the moment at which they enter. They cannot normally be made
after further entries have come in. This fact is the source of the nesting of
grammatical constructions, e.g. the fact that later adverbs are farther from the
verb than earlier ones (except for intrusion with commas). Also. the determina_
tion of whether a word is contributing low information, and is thus reducible,
can be made on the basis of the relation of the word to its prior (in most cases
immediately prior) or next later entries. rf come is zeroable wder expect, and
manner rejects reduction to -/y when between ztrite and extensipe, it is all on the
basis of their likelihoods to their next entries to ilcpect, and, to write, extensioe
- -
without regard to the rest of the sentence.
These conditions on entry and reduction suffice as grammatical processes.
The grammatical patterns that emerge for each language are due to the specific
types of entry-classes and reductions, and to the similarities of physical shaoe
-*ord,
and of domain among the various reductions. Given a list of the ir, .u"h
entry class, |,tr , o n, etc., and a list of the specific reductions with the subdomains to
which they apply, these processes produce the actual sentences of the language.

8. NIernauATrcAL pRopERTTES

The set of grammatically possible sentences is closed under the entry-order

relation, making it amenable to mathematical investigation. Furthermore, the
objects which participate in the entry order are defined solely in terms of that
order: a word is defined as having entry requirements of N or o where ,Ay', o
are defined by their having either null or non-null entry requirement; this
makes the set formed by the entry-order relation a mathematical object. The
further grammatical events, mainly the reductions and the insertions of in-
definites, are defined on likelihoocl and other relations among the participants in
the entry-order relation.
The entities in the set which is closed with respect to this entry-order relation
are sentential word-sequences. But they could be anything else, and if these
other things had the same estimated likelihoods of entry that the words do, then
the new entries would constitute under the entry-order relation a set isomorphic
to the set of sentences, as indeed we have in writing vis-)-vis speech. In contrast,
concatenation as an operation in the set of all word-sequences, or manipulations
of word-position or the like without regard to entry status, does not distinguish

r3
 ZELLIG HARRI

sentences from non-sentences without a large ad hoc apparatus of grammatical

conditions.
The grammatical analysis of a sentence consists in finding what ordered entries
and reductions, the occurrence-conditions for each being satisfied by its pre-
cursor, yield the sentence; it constitutes a decomposition of the sentence into a
canonical form.
Abstractly, therefore, we are dealing here not with words defined as having
certain meanings and certain morphological structures, and not with a body of
rules statecl on particular words or combinations, but with the set of grammatically
possible sentences as a mathematical object. As a result of all this' it is possible
to state various relations among the entities, and the subsets, in the set of
sentences. And since the interpretation of the mathematical object is a real thing,
language, the subsets and relations have real interpretations too. Only a brief
sketch of some subsidiary relations can be given here.
We take the set {S} of grammatically possible sentences with homonymous
(ambiguous) sentences, which are reduced from different sources, being taken as
difierent sentences. In {S}, each operator which has precisely one operator in its
argument-requirements (called unary) maps the set {S} into itself, i.e. onto the
subset whose latest operator is that unary. Each binary operator, which has
two operators in its argument-requirement, maps {SxS} into {S}' Each
reduction maps a subset of {S} (having a particular latest or penultimate entry)
onto a subset (having that reduction). All of these are thus transformations and
partial transformations of the set {S}, and it is for this reason that these inter-
sentence relations were called transformations'
In {S}, the subset {1} of sentences containing no reduction is closed with
respect to the entry-order relation, and can be called a sublanguage, In n succes-
sive reductions in a sentence, the domain of the mth (*4") cannot be larger
than that of the z-rth (since the mth is defined on the resultant of the m-Ith);
hence while it may be that there are some unreduced sentences which have no
reduced one as image, there are no reduced sentences that lack an inverse
image in the sublanguage. This, plus the fact that the reductions do not alter the
information in the sentence, means that all the information in the reduced
sentences is present in their inverse images, i.e. in the sublanguage. Hence the
set {S} contains a sublanguage, {1}, whose sentences have the transparent
entry-order construction, unrestricted by reductions, and contain all the in-
formation in the sentences of the language.
Various algebraic structures can be defined in {S}. Equivalence relations can
be defined in {s}, yielding various partitions of the set. The set {s} is a semi-
group under and asbinary composition. If we take two sentences in it to be in the
same equivalence class E, when they contain the same ordered unary operators
and reductions, then and provides a binary composition in the set {Z} of equi-
valence classes, with (8, and. Eu) : (E,c. ona r). Theelementarysentences of the
r+
 GRAMI\{AIT ON MATHEMATICAL PRINCIPLES
language are reachable via the kernel of the naturar mapping
of {s} onto its
quotient set {n}; it is for this reason that they *"r. .utLa"t ernel-sentences.
(clf' fn' 6r of Harris (1957) and more generally on transformations
(Harris,
t97o' t968)') Another partition puts into one equivalence class
all sentences
having the same partially ordered entries: all sentences in
an equivalence class
are then paraphrases, and one is unreduced, i.e. a member
of the informational
sublanguage above.
Defining the reductions on entries rather than on sentences
makes possibre a
finitary grammar: a finitc set of elements and operations, and
finite domains
(prior entries, not whole sentences) for these.

9. Pnorrnrrls oF LANGUAGE
T'his system of grammar can be presented as a set of formation
and transforma-
tion rules on an alphabet consisting of ry' and o. Given which
words are in each
entry class and which are in the domain of each restricted reduction,
this system
provides not merely the grammar of sentence structures
but the grammar of
actual sentences - the indication of which word-sequen"a,
ur"" sentences,
although one would still neecl information about likelihoods
and about which
reductions are common. In general, the domains of the
restricted reductions
require listing, at least for borderline cases, even though they
are not arbitrary
and must be related to extra-high likelihood (yierding zeroing), extra-low
likelihood (semantic rejection of a combination), ani the
liie. A.,ulolgi" and other
processes on the resultants of entry-and-reduction
may lead to tie spread of
certain constructions and grammatical patterns.
The theory tells what can be found in a sentence and where
it can be found:
red,ctions apart, every sentence must contain at least one primitive
argument.A/,
and it must contain as many A--argument positions unier its
op.ritor. u. it
contains ly' (before reduction). And for every operator or operator-pair
in it, it
may contain an operator with one o or two o respectively
u. urg,r-..ts. since
thc position of each oper_ator is immediately after the first of its-ordered
argu-
ments, and since each reduction leaves a trace (as follows
from the recoverability
of zeroed words), with the main permutation being just the
secondary-sentence
shift, the possible locations of the entries and their possible physical
,hrp., u."
known. An effective decision procedure can be formulated
for analysing every
sentence into ordered reductions and entries. There
are, of course, considerable
complexities due to degeneracies of reduction, many of which
are only rocar and
can be resolved by referencc to other entities in the sentence.
The sentence analysis presented here reveals the nature of grammaticar
transformations. (r) Some transformations are simply reductions,
as"in He reads
by zeroing of the indefinite appropriate object in He reads
things; or a succession
of reductions as in fls chitdhoott was happy from For one to ie
happy ts a state;
r5
 ZELLIG HARRIS

his being a child was that state. (z) Other gramrnatical transformations are
reduced - or as though reduced from added entries which had very general
meanings, as in the passive -en, where The trees are in a state of the chopping of
the trees by settlers first zeroes of the trees, then reduces in a state of the chopping
to chopped, so that it may seem that the object of chop has moved to subject
position while chop has somehow changed to is chopped. (3) Yet other trans-
formations are the results of distributing the words of a sentence through two
sentences, filling in the remaining positions with non-specifying words (e.g.
demonstrative pronouns). The non-specifying words are then reduced or
zeroed,but their presence leaves a nuance on the sentence; as in Whatfellis the
booh, or in the restrictive relative clause. What is common to all of these is that
the words of the 'source' sentence appear in the same or equivalent grammatical
relations in the transform, and that the change can be made in all or many
sentences of the source form. The reason for this is that the added words in the
expanded source forms are of such general meaning (state, that, srtmething) that
they do not afiect the word likelihoods (selections), and contribute only a nuance
of meaning, and that they therefore invite reduction to affixes or the like which
can then be considered 'constants' of the transformation. And indeed the
transformations which are pure reduction (I) are pure paraPhrases. The others
(2,3)have a difference in nuance, and this difierence is visibly produced in the
source form, by state, that, etc. in their given positions, and is thereafter para-
phrastically preserved in the reduction. It is thus seen that transformations are
not a set of word-manipulations coming full-blown, and not a grammatical
process at all, but an effect. They are the effect of reductions and of broad-
selection entries. The only grammatical transformations which are found in a
language are those that can be reached by such entries and by the known
reductions.
The method of analysing a sentence ,4 is to look for a source which (I) is made
out of words that have entered the source solely on entry-order grounds' and
(z) is reducibleto A by attested reductions in attested conditions. 'Solely'means
that any morphological structure the word may have is not used additionally
to its entry-order in accounting for the position of the word. Indeed, we may
accept that certain operators are amxes (e.g. -enl-ed) entering into affixal position
upon their second argument.
The different elements of language structure in terms of the present theory
fit the different universalities of linguistic features. The system of argument-
requirements and of differences in relative likelihood (selection) for words in a
single entry class seem to be universal, although some languages, for example,
have few primitive arguments (concrete nouns and demonstrative pronouns).
The main types of reduction seem to be found everywhere - referential zeroing
and pronouning, and some kinds of'expectable' zeroing. Indeed, the intertrans-
latability of language relates in large part to the similarity of entry-structures and
r6
 GRAMMAR ON MATHEMATICAL PRINCIPLES

of the main reduction types, with Onn operators in one language being translated
largely by On, operators in the other, and so on. The specific reductions are
more nearly language-specific, and in several cases the most special and difficult-
to-analyse transformations are peculiar to the one language in which they are
found.
The special grammatical patterns of each language or family of languages
arise largely from characteristics of the reductions: either from special shapes
(such as the compound-word stress), or from similarities in shapes (such as the
similarity of noun-modifiers to the subject and object of a nominalized sentence),
or from having similar domains for various reductions (this creates such gram-
matical categories as nouns), or from complementary domains (which create
conjugations and the like) - and finally from the frequency of use of some
reductions (as in the auxiliaries). But these patterns may be affected also bv
analogic extensions of sentence features, by systematic borrowing, etc'
Nlany of the source forms reached in this analysis reconstruct known or
possible historical sources. There are cases of a word moving to a new grammatic-
ul ,t.trr., replacing its previous occupant, asinwill becoming a future tense, and
in the compound (periphrastic) tenses of various European languages. But many
other extensions and specializations of use of a word do not get as far away from
their earlier uses. They may differ locally, as in the case of the auxiliaries, but in
their farther environment or in their complementary words (variants) they may
still belong to their class of origin: can c n still be considered a variant of a
tensed O,o verb. And the development of really new grammatical relations is
rare and slow indeed. The status of modifier may seem to promise such a
development; but there are many transformational connexions which keep the
set of all modifiers closely tied to being an entry-and-reduction resultant'
namely to being residues of a secondary sentence one of whose arguments is the
same as one of those in the primary sentence.
There are, however, other situations which exhibit not so much the tenacity
of history as the inherent limitations of what language can or cannot express.
Thus morphology may seem to present a degree of freedom independent of
syntax, one which would make morphology-rich languages able to say things
that morphology-less languages cannot. This turns out not to be the case
(witness their intertranslatability, for one thing). The affixes are equivalent to
reductions of operators, reaching their affix status by established reductions'
This may be due to the fact that many of them were indeed such operators.
For the others, granted that they developed outside syntax, they were domesti-
cated, upon entering a sentence, by the entry relation which creates syntax'
Whatever can be expressed with the aid of morphology can be expressed (given
the availability of the words) without morphology, by operator-argument rela-
tions among simple words, as though the affixes were reductions of such words.
Since the entry-structure of language is so simple, the bulk of what is called
t7
 ZELLIG HARRIS
grammar is created bv the reductions. But the reductions do not alter the entry-
content of sentences. In all cases, allor,ving for degeneracies, when an entry
lras been zeroed, its presence can be reconstructed from the structure of the
sentence in its post-zeroing form, so that we do not even have to say that the
word has been dropped but only that its phonemic shape has been changed to
zero.If the zeroing is referential we know (up to ambiguity) from the anrecedent
which word has been zeroed. If the zeroing is of a non-specific or of an ex-
pectable word, its meaning-contribution to the sentence was small and can be
found in the new grammatical relation that results (e.g. the verb-less milhman,
the object-less He reads). This means also that if a sentence contains covert
meanings, not explicitly contained in any of its u'ords, thesc are clue to zeroed.
words in the source form.
The result of this is that the meaning of each entry, and the meaning of its
operator and argument relation to prior and next entries, is preserved in the
sentence as it grows with new entries and as it is transformed. A sentence is a
particular ordering of entries, and its meaning, the meaning of the ordering of the
entries, is the ordering of the meaning of the entries. In making a syntactic
analysis we therefore obtain the meaning of the sentence directly in terms of its
entering words; and the components of the syntax are the components of the
meaning. The present method enables us to reconstruct a sentence in such a
way that no component says anything which is not said in the final sentence.
Thus ly'o man came is derived not from A man came; the man is none but from
I deny that a man came. And Frozenfood lasts long is taken not fron Food lasts
long; that food is frozen but from something lasts long; that something is
food;
that food is frozen.
Since even the nuances brought in by transformations are brought in by
zeroable words, this means that all the information expressed in the sentences
of the language is expressed in the subset consisting of their unreduced, possibly
reconstructed, source sentences which have only the entry structure. And since
the entry relation has little or no grammatical restriction on the words of an
entry class, but only differences in likelihood, this means that the information in
language is expressible in these unrestricted entry constructions. Thus it is not
any complexities of the real world of information that are responsible for the
complexities so characteristic of grammar. For the information alonc, a very
simple system, the entry-order, suffices.
one sees from the entry and reduction analysis that the structure of language
is an information-carrying system; if we take its discreteness into account it is
even better understandable as an information-transmission system. If we
exclude gestural, non-grammatical, features such as intonations of irony,
language has no structural provision for the expression of feeling; one only
makes statements, possibly false, that one has this or another feeling. The use of
language as a vehicle for feeling and art, as in poetry, is done by a secondary

r8
 GRAMMAR ON MATHEMATICAL PRINCIPLES
manipulation of an existing language and grammar. And where the structure of
music, certainly a vehicle for feeling and art, includes such elements as sequences
of notes and variations defined on the sequence, the word-sequences of language,
as the symbol-sequences of mathematics, are onlv projections of the partially
ordered entries.
The informational character of language comes out even more sharply in
the specialized languages of individual sciences, where one can see not only a
somewhat different grammar, but even something new: a somewhat different
type of grammar, made possible by the consistency of subject-nratter and by the
fact that whereas language has no external metalanguage, the sublanguage of a
science can draw upon the whole language for its metalanguage.
From the nature of the separate structural features which together create
language, one can see a naturallv growing instrument for transmitting informa-
tion. The entry requirement is an adequate form for predication, which is
seen, e.g. in the derivation of the interrogative, to be the source information-
device for all language; if a word of class Y is said only provided one has said.a
word of class X, then Y exists in sentences onlv as something said on -Y - as
close a structure for predicating Y of I as one might expect.
Language is also reasonablv efficient when looked upon as a naturally developcd
instrument for transmitting information. F-irst, it is the least structure needed
for the entry relation: the arguments are only.Ay', O, and not arbitrary subclasses
of O, or even Od as vs. Oo, etc. And the number of arguments goes up to onlv
two or three - longer arguments could always be paraphrased by operators on
operators. Second, the reductions do not take place upon sentences as a whole
(..g. by judging their redundancy) but upon the entering words (by judging
their infornrational contribution in respect to each other). Third, the reductions
make for a great shortening of the sentence without changing its information.
'lhe degeneracies which result from some of them could have been avoided by
having different reductional shapes. That they do not avoid this is as mucir a
result of the unplanned character of language as are the homonyms that result
from certain sound changes.
One item of efficiencv, or of unplanncdness, is that many reductions have
become rules rather than preferences. For example, the dilTerence between
verbs and adjectives can be stated as the tense being attached directly in the
case of the less durative operators (verbs), and via a carrier 6e in the case of the
more durative operators (adjectives). Borderline cases could have remained a
matter of preference, depending either upon the speaker or upon the meaning
in the context: e.g. sleep,i//could have been sometimes said as verbs and some-
times as adjectives. Instead, although different languages make different decisions
about such words, within a language each word is fixed - sleep a verb, ill an
adjective - so that to use i// with the less-durative, verb-like, placing of tense
constitutes not merely a delicate variation in meaning but an outright 'mistake',
r9
 ZELLIG HARR
a flouting of convention, albeit a very old convention, and not not maintained by
institutional interests. Whether one sees this as efficiency or as an unnecessary
restriction upon freedom of action may depend upon one's views of what is
desirable in society.
Given all this, the structure of language can be seen as an outgrowth of the
uses in which language was developed. There is no need to appeal to some
independently existing structure or structuralism - whose existence before
language or independent of language could in any case not be explained - to
see how the structure of language came to be what it is,

RDFERENCES
Harris, Z. S. (rSSZ). Co-occurrencc and transformation in linguistic structure. Zg 33.
283-34o.
Harris,Z. S.(r968), Mathematicalstructuresof language, (Intersciencetractsinpureand
applied mathematics zr.) New York: Wiley.
Harris, Z. S. (rgZo). Papers in structural and transformational linguisrrcs. Dordrecht:
Reidel.
Hartis, Z. S. (r976a). Notes du cours de syntat( (ed. Gross, Maurice). Paris: Editions du
Seuil.
Harris, Z. S. (r976b). A theory of language structure. American Philosophical Quarterly
t3. 237-55.
Harris, Z. S. (r976c). On a theory of language. JPh ZS. 2fi-76.
Harxis, Z. S. (to appear). Grammar of English on mathematical principles. New York:
Wiley.