Linguistic Knowledge and Unconscious Computations
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
RIVISTA INTERNAZIONALE DI FILOSOFIA E PSICOLOGIA ISSN 2039-4667; E-ISSN 2239-2629
DOI: 10.4453/rifp.2016.0035 Vol. 7 (2016), n. 3, pp. 338-349
STUDI
Linguistic Knowledge and Unconscious Computations
Luigi Rizzi(α)
Ricevuto: 24 gennaio 2016; accettato: 27 ottobre 2016
█ Abstract The open-ended character of natural languages calls for the hypothesis that humans are en-
dowed with a recursive procedure generating sentences which are hierarchically organized. Structural re-
lations such as c-command, expressed on hierarchical sentential representations, determine all sorts of
formal and interpretive properties of sentences. The relevant computational principles are well beyond
the reach of conscious introspection, so that studying such properties requires the formulation of precise
formal hypotheses, and empirically testing them. This article illustrates all these aspects of linguistic re-
search through the discussion of non-coreference effects. The article argues in favor of the formal linguis-
tic approach based on hierarchical structures, and against alternatives based on vague notions of “analog-
ical generalization”, and/or exploiting mere linear order. In the final part, the issue of cross-linguistic in-
variance and variation of non-coreference effects is addressed.
KEYWORDS: Linguistic Knowledge; Morphosyntactic Properties; Unconscious Computations; Corefer-
ence; Linguistic Representations
█ Riassunto Conoscenza linguistica e computazioni inconsce – Il carattere aperto del linguaggio naturale
avvalora l’ipotesi che gli esseri umani siano dotati di una procedura ricorsiva che genera frasi gerarchica-
mente organizzate. Relazioni strutturali come il c-comando, espresse su rappresentazioni frasali gerarchi-
che, determinano tutte le proprietà formali e interpretative delle frasi. I principi computazionali rilevanti
sono totalmente al di fuori della portata della coscienza introspettiva e così lo studio di tali proprietà ri-
chiede la formulazione di precise ipotesi formali e la loro verifica sperimentale. Questo articolo illustra
tutti questi aspetti della ricerca linguistica, esaminando gli effetti di non-coreferenza. Si argomenta in fa-
vore dell’approccio linguistico formale basato su strutture gerarchiche e contro alternative basate su va-
ghe nozioni di “generalizzazione analogica” e/o che impiegano il semplice ordine lineare. Nella parte fina-
le si affronta il tema dell’invarianza e della variazione cross-linguistica degli effetti di non-coreferenza.
PAROLE CHIAVE: Conoscenza linguistica; Proprietà morfosintattiche; Computazioni inconsce; Coreferen-
zialità; Rappresentazioni linguistiche
█ Introduction have not encountered in their previous linguis-
tic experience, and still they can easily integrate
ONE REMARKABLE PROPERTY OF HUMAN such new messages, understand them and
language is its unbounded character: speakers properly respond in dialogue. In fact, any
are constantly confronted with sentences they speaker potentially masters an unbounded
(α)
Dipartimento di Scienze sociali, Politiche e Cognitive, Università degli Studi di Siena, Complesso San Nic-
colò, via Roma, 56 - 53100 Siena (I)
E-mail: luigi.rizzi@unisi.it ()
Creative Commons - Attribuzione - 4.0 InternazionaleLinguistic Knowledge and Unconscious Computations 339
number of messages. This property caught the a mechanism for computing representations
attention of major thinkers in the past: in par- relevant for a specific cognitive domain, with
ticular, Descartes saw in it a test capable of dis- properties which may be in part domain gen-
criminating man and machine:1 no automaton, eral, and in part task-specific.
no matter how sophisticated, would be capa- Is there an alternative to this computational
ble of what any man can do («ainsi que les view? What is sometimes considered an alter-
hommes les plus hébétés peuvent faire»),2 native is the enrichment of the list model with
hear a new sentence and respond appropri- the idea that the unbounded character of the
ately to it, in a way that is neither determinis- system is due to our capacity for analogical
tic nor arbitrary. generalization: during language acquisition, the
Once the unbounded character of lan- child hears a finite number of messages and she
guage is properly recognized, this has immedi- memorizes them, so that the initial knowledge
ate implications for the nature of linguistic is “item-based”; at some later point, the child
knowledge. “Knowing a language” cannot be generalizes their properties to new massages
equivalent to “having memorized a list of mes- through analogy as in the “neuroconstructivist”
sages”, as the list would have to be infinite, approach.3 The problem with this (very popu-
and we would not have had enough time (or lar) view is that it is neither right nor wrong, if
space in neural circuitry) to memorize it. The expressed at this level of generality: it is simply
“memorized list” model may be essentially vacuous until when we give content to the no-
correct, at least for a first approximation, for tions of “analogy” and “analogical generaliza-
our knowledge of the lexicon: we hear a new tion” to capture the fact that certain “analogical
word, we figure out what it means, and we add generalizations” are unerringly made by all
it to our mental lexicon (it should be noted, speakers (and all language learners), while oth-
though, that a list-based model is to be ex- er conceivable “analogical generalizations” are
pressed with the proviso that the lexicon is not never made. But once we have properly struc-
an unstructured list, but a highly structured tured the concept, so much structure emerges
system). But the list idea is not appropriate to that the vague term “analogy” does not seem to
capture our mastery of sentences. Clearly, the be a felicitous terminological choice to address
secret of the unbounded character of language the “projection problem”, the fact that the child
is in its combinatorial nature: words can be projects her finite linguistic experience onto an
combined to form sentences, so knowing a unbounded set of possible messages.4
language means mastering the combinatorial I would like to illustrate these issues
laws that govern sentence formation. In other through a simple example: the constraints on
words, knowing a language means possessing referential dependencies between nominal ex-
a generative procedure, capable of computing pressions and pronouns. The discussion will
an unbounded number of messages; acquiring aim at making two points. On the one hand, a
a language means acquiring the lexicon, and point that directly bears on the topic of this
mastering the combinatorial procedure to workshop: the knowledge that speakers unerr-
generate new messages. ingly manifest is completely unconscious,
This is what is sometimes called the “com- there is simply no way to introspectively pene-
putational view” of linguistic knowledge, an trate the structural computations that we all
important legacy of the study of language to perform when we evaluate the possible inter-
the broader domain of cognitive neuroscience, pretation of a pronoun in context; so the only
as the idea of the “computational mind” thing to do to study this mental capacity is to
proved to be viable in the study of the human formulate precise structural hypotheses, let
cognitive capacities well beyond language, them generate predictions, and test the pre-
from vision, to reasoning, to motor control: dictions. This is true for this case, as well as for
having a cognitive capacity means possessing so many other cases of the study of non trivial340 Rizzi
mental capacities, in language and other cog- is abstract knowledge, in the sense that it is
nitive domains. On the other hand, I will try completely independent from any particular
to show that the knowledge that every speaker knowledge of the discourse situation, or of the
has about the possible referential dependen- states of affairs in the factual world: I may very
cies between nouns and pronouns obeys struc- well not know anyone named John, and still if I
tural constraints which appear to go well be- hear somebody utter (1) I will assume that the
yond the reach of an unstructured notion of speaker may mean that a certain guy John said
“analogical generalization”. A brief discussion that the same guy John was sick, while the
of the invariance and variation observed speaker uttering (2) does not intend to convey
across languages in the domain of coreference that meaning. Clearly, we dispose of a proce-
will conclude the paper. dure allowing us to evaluate coreference, and
the procedure discriminates between (1) and
█ A constraint on coreference (2), completely independently of any
knowledge of factual situations.
In certain sentences, a name (or other nom- What is this procedure? Here is the point of
inal expression, for instance a definite descrip- immediate relevance for our general theme. We
tion) and a pronoun can refer to the same indi- perceive in a crystal clear manner the result of
vidual, or “corefer”. For instance, in (1), the this mental procedure (coreference is possible
name John and the pronoun he can corefer: here and impossible there), but the procedure
itself is completely inaccessible to our conscious
(1) Johni said that hei was sick introspection.
In introductory courses, I always do a little
In other words, the sentence can mean informal experiment in class, first testing stu-
“John said that he, John, was sick”: we may ex- dents on the contrastive judgment between the
plicitly express coreference through the index- Italian equivalents of (1) and (2) (obviously ac-
ing notation, i.e., by assigning the same index cessible to everyone), and then asking them on
(i) to the two coreferential elements, as in (1). what basis they are able to differentiate coref-
Coreference is not obligatory (i.e., in (1) he erence possibilities in the two examples. Vari-
could refer to Bill, who was introduced in the ous hypotheses are made in the discussion in
previous discourse in a context like Nobody un- class, some rather complex, but one idea that
derstood why Bill had made such a mistake, but always emerges, and strikes everyone for its
then…), but it is an option in cases like (1). simplicity and plausibility, is that speakers may
Coreference is possible in certain environ- use a linear strategy:
ments and impossible in others. Consider (2)
for instance: (3) Linear strategy: A noun and a pronoun can
corefer when the noun precedes the pro-
(2) * Hei said that Johni was sick noun in the linear order of the words; if the
pronoun precedes the noun, coreference is
The sentence is fine, but he can only refer to impossible: i.e.,
an individual different from John, e.g., Peter,
introduced in previous discourse. Coreference a. ok … Ni … Proni …
(he, John) is excluded, and this is what the as- b. * … Proni … Ni …
terisk marking the sentence means: the sen-
tence is well-formed per se, but it is excluded This hypothesis has the appeal of simplicity,
with the interpretation expressed by the indi- and a considerable plausibility: after all, it
ces. makes perfect sense that we first introduce a
Notice that the contrast (1)-(2) is immedi- referent with a noun, and then we refer back to
ately clear to all the speakers of English, and it it thorough a pronoun. Nevertheless, the linearLinguistic Knowledge and Unconscious Computations 341
strategy is not the one speakers actually use █ The hierarchical nature of syntactic repre-
within a complex sentence: there is literally an sentations.
infinity of sentences (in English, Italian and
other languages) in which the pronoun pre- A fundamental property of syntactic struc-
cedes the noun, and coreference is still possible. tures is that words forming sentences are as-
Here are some examples: sembled into hierarchically organized units.
Over fifty years of discussion on the fundamen-
(4) People who know himi well say that Johni is sick tal combinatorial device led to the conclusion,
within the Minimalist Program,5 that it has the
(5) isi father said that Johni was sick simplest possible shape, an instruction which
says “take two elements and string them to-
(6) When hei is sick, Johni does not go to work gether to form a third element”. This is the op-
eration Merge:
So, (4) can naturally admit the interpreta-
(7)
tion in which the pronoun him refers to John, …A…B… A B
in (5) his father may well mean John’s father,
with the possessive his referring to John, he can Merge takes two elements, A and B,
naturally refer to John in (6), etc. There may be which can be items drawn from the function-
a moment of hesitation when coreference is al or contentive lexicon, or partial structures
evaluated in contexts in which the pronoun already built by previous applications of
precedes the noun (contexts of “backward co- Merge, and strings them together to form a
reference”), but the contrast between (4)-(6) new structural unit. Repeated applications of
and the sharp impossibility of coreference is (2) Merge give rise to hierarchical structures
is clear. which can be expressed by familiar syntactic
So, what should we conclude from these ex- trees like the following:6
amples? Speakers possess a procedure allowing
(8)
them to evaluate coreference, and they apply it
IP
very efficiently and automatically to the sen-
tences they hear: the judgment on the interpre- DP I’
tation of (1), (2) etc. is clear and quickly deter- John
mined by the speaker. Yet, speakers don’t have I VP
any conscious introspective access to the pro- will
cedure: they only “see” the result. Clearly, the V’
students who proposed the linear rule did not
have introspective access to the procedure they V DP
were using, they simply formulated a hypothe- meet Mary
Similar representations permit the defini-
sis on the nature of the procedure, based on the tion of fundamental structural relations
data they had access to, their own interpretive which govern the structuring of form and in-
judgments on these sentences. The hypothesis terpretation. One very fundamental relation
turned out to be incorrect, but this is, in fact, is c-command,7 which I will define as follows:
the only way to proceed: formulate a precise
hypothesis on the nature of the mechanism, (9) α c-commands β in the following con-
submit it to empirical testing, and revise it in figuration:
accordance with the empirical evidence.
In order to successfully address the prob-
lem, we now need to sharpen our assumptions α γ
about the structural organization of linguistic
representations. … β …342 Rizzi
i.e., c-command holds in a configuration in command is involved in interpretive process-
which β is contained in the node γ which has es. For instance, an anaphoric expression like
been merged with α; γ is also called the “sister a reflexive must be locally c-commanded by
node” of α. So, we can say that α c-command β an antecedent which determines its refer-
when β is contained in the sister node of α. ence. That is why a subject can bind an ana-
C-command is a formal way to express phoric object, but not vice-versa:
structural prominence. For instance, as is clear
from representations like (9), the subject c- (13) a Johni criticized himselfi
commands the direct object, but not vice-versa: b * Himselfi criticized Johni
the object is contained in I’, the sister node of
the subject DP John; the object does not c- Similarly, an anaphoric object can be
command the subject because the subject John bound by the subject DP, not by a possessive
is not contained in the sister node of the object DP contained within the subject. I.e., the fol-
DP Mary, which is V. lowing sentences mean “John’s brother criti-
C-command plays a pervasive role in the cized John’s brother”, and cannot mean
principles determining form and meaning of “John’s brother criticized John”, with the
syntactic representations. possessive acting as the antecedent (whether
Consider for instance agreement: the sub- the possessive is prenominal, as in English, or
ject, but not the object, agrees with the verb in postnominal, as in Italian):
person and number in English:
(14) [ [ John ]’s brother] criticized himself
(10) John has (*have) met the girls
(15) [ Il fratello di [ Gianni ]] ha criticato se stesso
This follows from the fact that the sub-
ject, but not the object, locally c-commands “The brother of John has criticized himself”
the node I, containing the inflected verb, in
representations like (8). Similarly, in All these properties follow from the fact
that anaphor binding requires c-command
(11) A picture of the girls was (*were) hanging on the from the binder: the subject asymmetrically
wall c-command the object, whence the facts of
(13); and a possessive DP does not c-
The copular verb agrees with the whole command anything outside the DP that con-
subject DP1 A picture of the girls, which local- tains it, whence the interpretive properties of
ly c-commands it, not with the adjacent DP2 (14)-(15). If the anaphor is DP internal, the
the girls, which does not c-command it:8 possessive can bind it because c-command
(12) IP holds:
DP1 I’ (16) [ [ John ]’s picture of himself]
was hanging from the wall
A picture PP I VP
was Here too, binding is fine because the
of DP2 hanging from the wall nominal specifier (the possessive) asymmet-
rically c-commands the nominal complement
the girls of picture:
In addition to being operative in all sorts
of morphosyntactic processes, such as prop- (16) [ [ John ]’s picture of himself]
erties of the case-agreement system, c- was hanging from the wallLinguistic Knowledge and Unconscious Computations 343
Here too, binding is fine because the
nominal specifier (the possessive) asymmet- (20) IP
rically c-commands the nominal complement DP I’
of picture: He
I VP
(17) DP
V’
DP D’
| V CP
John D NP said
‘s C IP
N’ that
DP I’
N PP John
picture I AP
of DP was |
himself sick
So, here too complement and specifier In a classical paper Howard Lasnik9 ob-
could not be reversed, because in that case served that the pronominal DP has the name
the anaphor would not be properly c- in its domain of c-command in (20), but not
commanded (hence bound): in (19), which led him to state the following
interpretive principle:
(18) * [ [Himself]’s picture of John ]
was hanging from the wall (21) Non-coreference: A pronoun and a name cannot
corefer when the pronoun c-commands the
We are now ready to go back to the non- name.
coreference pattern that we started with in
section 2, which also crucially depends on c- So, (21) precludes coreference in (20):
command. there the pronominal subject of the main
clause c-commands everything else, including
█ Non-coreference the name subject of the embedded clause. In
every other environment, coreference is an
Consider the tree representations of (1) option. For instance in (19) the c-domain of
and (2): the pronoun is limited to the embedded
clause, the name subject of the main clause is
(19) IP external to the c-domain of the pronoun,
DP I’
hence (21) does not apply and coreference is
John possible.
I VP
In (19) the name asymmetrically c-
V’ commands the pronoun, but coreference is
V CP
also possible in the environments in which
said neither element c-commands the other, and
C IP the pronoun and the name have disjoint c-
that domains.
DP I’
he
I AP
This happens, for instance, in (4), whose
was | structural representation is expressed with
sick some simplification by the following tree:344 Rizzi
(22) IP (23) IP
DP I’ DP I’
People CP I VP Poss N I VP
His father
who IP V’ V’
V CP V CP
I VP say said
C IP C IP
V DP that that
Know him DP I’ DP I’
John John
I AP I AP
is | was |
sick sick
Here neither him nor John contain the
(24) * Hisi picture of Johni’s father was hanging from
other element in their c-domain, hence (21)
the wall
does not apply and coreference is a viable op-
tion.
When the pronominal element is part of a Finally, in (6), the pronominal subject of
larger phrase, as the possessive his in (5), it the preposed adverbial clause When he is sick
can corefer with a following name because does not c-command the main clause and its
the possessive does not c-command the content, hence the name John and the pro-
name, as is clear from the following tree rep- noun he can corefer, as the non-coreference
resentation (the domain of c-command of principle does not apply here:
the possessive is the DP his father, which In conclusion, a complex array of inter-
does not include the name John): pretive facts involving coreference follows
from a simple principle such as Lasnik’s con-
(25) IP straint, applying on hierarchical structural
representations.
The subtle distinctions that we have ob-
CP IP
served in the complex non-coreference pat-
C IP DP I’
tern raise an interesting challenge for an un-
When John
structured notion of “analogical generaliza-
DP I’ I VP
tion”. The language learner will have access
he doesn’t
to a sample of sentences containing pro-
I AP V PP
nouns; she will figure out from context that
os sick go to work in some cases coreference with a nearby noun
is intended, and in other cases it is not. E.g.,
in connection with (1):
Notice that possessives are not exempted
from the non-coreference constraint: if the (26) John said that he was sick, and that’s why he
possessive c-commands the name, corefer- couldn’t come. He’s always looking for excuses!
ence is barred. So his cannot refer to John in (coreference between John and he is likely to be
the following, i.e., the sentence cannot mean intended)
“John’s picture of his (John’s) father is hang-
ing from the wall”: (27) Nobody had understood why Bill had not showedLinguistic Knowledge and Unconscious Computations 345
up, but then John said that he was sick (corefer- dependencies between linguistic expressions;
ence between John and he is likely not to be in- Reinhart11 proposed that the non-coreference
tended) effect may have its roots in the computation
at the interface between syntax and pragmat-
On the basis of this evidence, why doesn’t ics, and be amenable to principles of optimi-
the learner simply conclude, by analogical zation akin to Grice’s maxims. And other ap-
generalization, that coreference is always an proaches have been proposed.12 I will not try
option? But no one draws that conclusion, as to address the issue of the “further explana-
everyone systematically excludes coreference tion” here. The relevant point is that all these
in (2). 10 approaches (definitely including Reinhart’s
Even assuming that the child has some interface approach) crucially make reference
way of inferring that with structures like (2) to the hierarchical configuration of c-
coreference is never intended, hence that command, which just seems unavoidable if
some constraint must be assumed, why the approach aims at meeting empirical ade-
wouldn’t she assume a linear constraint and quacy, and capture the selective effects we
extend it by analogy to all cases of backward have reviewed.
anaphora, thus implicitly assuming the linear
rule (3) that is typically explicitly proposed in █ Invariance and variation
our toy experiment? But no language learner
does that, as speakers readily recognize the Comparative considerations become rele-
possibility of backward coreference in con- vant at this point. If non-coreference effects
texts like (4), (5), (6). It seems clear that, in follow from general principles of syntactic
order to reach empirical adequacy, an analo- organization and interpretation at the inter-
gy-based approach should build c-command faces, one expects to observe little or no
into the notion of analogical generalization cross-linguistic variation, under the assump-
(something like “in evaluating coreference, tion that such principles are shared by our
analogize only constructions involving iden- species. More precisely, one may expect vari-
tical c-command configurations between ation, but not of an unconstrained kind: if
nouns and pronouns”); but this move would general shared principles are involved, one
de facto assume the highly structured config- may expect cross-linguistic studies to show a
urational notions that the analogical ap- limited variation within a strongly invariant
proach is intended to avoid. In order to ade- architecture. In fact, non coreference effects
quately capture what speakers know and do, have been shown to hold in historically and
reference to the structured hierarchical no- typologically very different languages. A
tion of c-command just seems unavoidable. small sample:13
Then, the question arises of the “further
explanation” of constraint (21): should it be (28)
considered a primitive principle of the hu- a (Italian)
man language faculty, or could it be derived * ___ pensa che Gianni vincerà
___ thinks that Gianni will win
as a theorem from deeper and more general
principles? Obviously, the desirable option is b (M. Hebrew)
that a successful path of “further explana- * hu ma’amin she-John yenaceax
tion” may be identifiable. Here different ap- * he thinks that John will win
proaches have been proposed. Chomsky
c (Thai)
originally suggested that (21) may follow * khaw khit waa coon chálaát
from principle C, a component of the bind- * he thinks that John is smart
ing theory, the module expressing configura-
tional constraints on the possible referential d (Vietnamese)346 Rizzi
* no’ tin John se thang In other words, for some speakers of this
* he believes John will win language, the equivalent of “She said that
e (Gungbe) Mary was going to finish tomorrow” allows
*e vedo do Kofi na wa the reading “she = Mary” (while other speak-
* he believes that Kofi will come ers reject this reading, generally judged im-
possible by speakers of English).
f (Mohawk) What conclusion should be drawn from
* wa-hi-hrori-’tsi Sak ruwa-nuhwe’-s
* fact-1sS/MsO-tell-punc that Sak FsS/MsO-like-hab this element of variation for the general theo-
* I told him that she likes Sak ry of non-coreference effects? The rational
approach, here and elsewhere, is to study an
Nevertheless, the recent literature reports exceptional case to a very general pattern
that non-coreference effects of the simple with great care, in order to determine the ex-
kind considered so far are not, strictly speak- act scope and the fine structural properties of
ing, universal. Some languages seem to allow the “exception”. This is what Davies does,
coreference in a configuration in which the and the conclusion he reaches is that the ob-
pronoun c-commands the name in particular served variation is not “wild”, but highly con-
structural configurations. strained. The language offers clear evidence
Davies14 gives a comprehensive analysis of for a configurational organization, and a sen-
one such case, St’át’imcets (also known as sitivity of non-coreference effects from struc-
Lillooet), an American Indian language tural properties, such as c-command, e.g.:
member of the Salish family, spoken in the
southwestern interior of British Columbia, 1. Apparent violations of principle C in
Canada. The language manifests a certain St’át’imcets are limited to cases in which the
freedom in word order, with a tendency to pronoun and the name are in two distinct
have predicate initial clausal structures and clauses; if they are in the same clause, famil-
VOS order in transitive structures. Davis iar non-coreference effects are found, much
shows that for some speakers of the lan- as in English and the languages in (28).16 In
guage, a pronominal subject of a main clause other words, the St’át’imcets equivalent of
can be coreferential with a name or definite (31)a allows coreference, while the equiva-
description in an embedded clause: lent of (31)b does not, much as its English
counterpart:
(29)
Tsút=tu7 [kw=s=cuz’ nas ts’úqwaz’-am s=Mary (31)
natcw]. a She said Mary was going fishing tomorrow (* in
say=then [DET=NOM=going.to go fish-MID English, ok in St’át’imcets)
NOM=Mary tomorrow] b She smokes in Mary’s house (* in English, * in
“Mary said she was going fishing tomorrow”. St’át’imcets)
More literally: pro said Mary was going fishing tomor-
row.)15 2. Violations of non-coreference are found
for pronoun-name configuration but not
(30) for demonstrative – name (i.e., the equiv-
Skenkín [lh=w=as nmatq xát’em ti7 ku=qelhmémen’ alent of English “That one said that John
sqaycw áta7 tsítcw-s=a]. was sick” disallows coreference between
slow [COMP=IMPF=3CNJ walk uphill that DET=old That one and John, and much as the Eng-
man to house-3POSS=EXIS] lish equivalent does.)
“That old man walks slowly uphill to his house”.
(More literally: pro is slow when that old man walks 3. The language manifests Strong Crossover
uphill to his house.) effects, which are traditionally seen asLinguistic Knowledge and Unconscious Computations 347
cases of principle C violation. I.e., in Eng- archies and relations such as c-command (as
lish, while (33)a allows the bound reading in the classical “principles and parameters”
of the pronoun (for which x, x said that approach).19 Some properties appear to be
Mary likes x), (33)b does not: strictly universal (as the strong cross-over ef-
fects),20 others appear to allow a limited
(32) a Whoi __ i said that Mary likes himi? range of variation within a tightly con-
b * Whoi did hei say that Mary likes __i strained and otherwise uniform system.21
Davies shows that such strong cross-over █ Conclusion
effects hold in general in St’át’imcets as well,
with no observed variation across speakers. Humans constantly produce and under-
This is illustrated by the following pair: stand sentences they have not heard before.
This simple and crucial property of normal
(33) language use calls for a computational ap-
a. Swat ku=kw7íkwlacw e [kw=s=cuz’ melyíh pro ka- proach: knowing a language amounts to hav-
lál]? ing mastered a generative device capable of
who DET=dream [DET=NOM=going.to marry computing new structures. Studying even the
soon] most elementary formal and interpretive
“Who e dreamed [s/he was going to get married properties of linguistic representations, we
soon]?” quickly realize that the structural organization
of such representations is hierarchical, with
b. *Swat ku=s-kw7íkwlacw-s pro [kw=s=cuz’ melyíh e relations like c-command playing a crucial role
kalál]? in all sorts of syntactic processes, in morpho-
who DET=NOM-dream-3POSS [DET=NOM= syntactic properties like agreement, in proper-
going. to marry soon] ties at the interface with semantics and prag-
* “Who did s/he dream [e was going to get married matics such as the determination of referential
soon]?” 17 dependencies of pronouns and other nominal
expressions. Generic notions of analogy and
Once the empirical scope of the exceptional analogical generalization don’t even begin to
behavior is identified, Davies adopts the ap- capture the fine properties and role of such
proach to non coreference presented in Safir,18 hierarchical principles, which appear to be in-
and argues in detail for a formal parametrisa- timately related to the functioning of the fun-
tion of one of the principles of Safir’s theory damental combinatorial operations (recursive
(within the guidelines of Chomsky’s binding merge, in minimalist models).
theoretic approach) to capture the observed Mental computations extend well beyond
cross-linguistic difference. what is accessible to consciousness and intro-
Going into the details of Davis’s analysis spection. We have conscious access to the
is beyond the scope of the present article. Let representations which are computed, and
me just observe that the important point to this allows us to produce and understand new
be retained from his approach for our theme structures, and to express metalinguistic
is that non-coreference effects, like so many judgments of well-formedness and interpre-
other linguistic phenomena, show elements tation. But we have no introspective access to
of invariance and elements of variation. the underlying computational mechanisms.
Here, as elsewhere, the observed variation So, the only way to address and conduct the
never is “wild” and unconstrained: rather, scientific study of our cognitive capacities for
languages typically allow “local” points of language is to proceed as we would in the
variation in an otherwise invariant structural study of any other natural object: formulate
architecture, based on configurational hier- precise hypotheses, and submit them to em-348 Rizzi
6
pirical verification through techniques as di- Without addressing technicalities, let us just say
verse as possible. that DP, or “determiner phrase” is the label of a
We have only used a set of metalinguistic nominal expression, IP, or “inflectional phrase” is
judgments in different languages in our illus- the label of a clause, and VP, or “verb phrase” is
trative example, but there is no reason to put the label of the verbal predicate; for the sake of
any limitation to the kinds of linguistic evi- clarity I use here the familiar X-bar notation;
nothing substantially changes, for the points at
dence that we may want to consider: so, the issue, if Bare Phrase Structure (see N. CHOMSKY,
study of language acquisition (as in the refer- The Minimalist Program, cit.) is used, a more par-
ences mentioned above), language pathology, simonious notation adopted in much minimalist
psycholinguistic experimentation, brain im- research. We omit many details here, and we
aging studies can all be brought to bear on don’t discuss the algorithm through which the
precise models of linguistic knowledge. If elementary structures created by Merge receive
mental computations for language were ac- structural labels, a procedure which we leave at
cessible to consciousness, studying our com- the intuitive level: a verb and a nominal expres-
binatorial capacities would be straightfor- sion merged together form a verbal projection, a
ward; as linguistic computations are well be- verb phrase, etc. (see L. RIZZI, Cartography, Crite-
yond the reach of introspection, studying the ria, and Labeling, in: U. SHLONSKY (ed.), Beyond
Functional Sequence. The Cartography of Syntactic
properties of the system is a complex under- Structures, Oxford University Press, Oxford/New
taking which requires detailed formal hy- York 2015, pp. 314-338. For details on the label-
potheses and structured techniques of empir- ing algorithm see N. CHOMSKY, Problems of Pro-
ical testing, much as in the scientific study of jection, in: «Lingua», vol. CXXX, 2013, pp. 33-
any non-trivial aspect of the natural world. 49, special issue Core Ideas and Results in Syntax).
7
See T. REINHART, The Syntactic Domain of
█ Acknowledgments Anaphora, Ph.D. Dissertation, MIT, Cambridge,
MA 1976; T. REINHART, Anaphora and Semantic
Thanks are due to Giuseppe Varnier for Interpretation, University of Chicago Press, Chi-
helpful comments on a preliminary version cago 1983; T. REINHART, Coreference and Bound
of this article. Anaphora: A Restatement of the Anaphora Ques-
tions, in: «Linguistics and Philosophy», vol. VI,
n. 1, 1983, pp. 47-88.
█ Notes 8
If one assumes, as in many minimalist analyses,
that agreement is checked by establishing an “agree”
1
See R. DESCARTES, Discours de la méthode (1637), relation between the inflectional node and the sub-
Union générale d’éditions, Paris 1951; for a discus- ject DP in its thematic position, the agree relation
sion see N. CHOMSKY, Cartesian Linguistics. A reaches the subject DP1 and not DP2 because of lo-
Chapter in the History of Rationalist Thought, Har- cality. For Relativized Minimality see L. RIZZI, Rela-
per & Row, San Francisco 1966. tivized Minimality, MIT Press, Cambridge (MA)
2
R. DESCARTES, Discours de la méthode, cit., p. 86. 1990; for minimal search see N. CHOMSKY, Mini-
3
See, e.g., M. TOMASELLO, Do Young Children Have malist Inquiries: The Framework, in: R. MARTIN, D.
Adult Syntactic Competence?, in: «Cognition», vol. MICHAELS, J. URIAGEREKA (eds.), Step by Step. Essays
LXXIV, n. 3, 2000, pp. 209-253. in Minimalist Syntax in Honor of Howard Lasnik,
4
On some problematic aspects of the “item- MIT Press, Cambridge (MA) 2000, pp. 89-155. For
based” neo-constructivist approach to the acqui- evidence that agreement is checked both via the
sition of syntax see J. FRANCK, S. MILLOTTE, R. agree relation and in the local specifier-head relation
LASSOTTA, Early Word Order Representations: discussed in the text see M.T. GUASTI, L. RIZZI, On
Novel Arguments against Old Contradictions, in: The Distinction between T and Agr: Evidence from
«Language Acquisition», vol. XVIII, n. 2, 2011, Acquisition, in: G. CINQUE (ed.), The Structure of DP
pp. 121-135. and IP. The Cartography of Syntactic Structure, vol. I,
5
See N. CHOMSKY, The Minimalist Program, MIT Oxford University Press, New York/Oxford, pp.
Press, Cambridge (MA) 1995. 167-194; J. FRANCK, G. LASSI, U. FRAUENFELDER, L.Linguistic Knowledge and Unconscious Computations 349
RIZZI, Agreement and Movement – A Syntactic Anal- (ed.), 28th International Conference on Salish and
ysis of Attraction, in: «Cognition», vol. CI, n. 1, Neighboring Languages, University of Washington,
2006, pp. 173-216. Seattle 1993, pp. 217-232.
9 17
See H. LASNIK, Remarks on Coreference, in: H. DAVIES, Cross-linguistic Variation in Ana-
«Linguistic Analysis», vol. II, n. 1, 1976, pp. 1-22. phoric Dependencies: Evidence from the Pacific
10
Not only adults, but also children round age 3 Northwest, cit., p. 56. .
18
or earlier, as shown in experimental studies such See K SAFIR, The Syntax of (In)dependence, MIT
as S. CRAIN, C. MCKEE, Acquisition of Structural Press, Cambridge (MA) 2004.
19
Restrictions on Anaphora, in: S. BERMAN, J. CHOE, See N. CHOMSKY, Lectures on Government and
J. MCDONOUGH (eds.), Proceedings of the NELS Binding, Foris Publications, Dordrecht 1981.
20
XVI, GSLA, University of Massachusetts, Amher See P. POSTAL, Skeptical Linguistic Essays, Ox-
1986, pp. 94-110; M.T. GUASTI, G. CHIERCHIA, ford University Press, Oxford 2003.
21
Backward vs. Forward Anaphora in Child Gram- The special parametric property permitting
mar, in: «Language Acquisition », vol. VIII, n. 2, “upside down” coreference in cases like (29) may
1999-2000, pp. 129-170. See also S. CRAIN, R. well be linked to other properties of the language.
THORNTON, Investigation in Universal Grammar, Davies speculates that it may be connected to an-
MIT Press, Cambridge (MA) 1998. other special property of pronouns in the lan-
11
See T. REINHART, Anaphora and Semantic In- guage: their ability to be introduced in discourse
terpretation, cit.; T. REINHART, Coreference and contexts without a prior discourse referent. See
Bound Anaphora: A Restatement of the Anaphora also Vincent Homer (V. HOMER, Backward Con-
Questions, cit. trol in Samoan, in: S. CHUNG, D. FINER, I. PAUL, E.
12
See e.g. P. SCHLENKER, Non Redundancy. To- POTSDAM (eds.), Proceedings of the 16th Meeting of
wards a Semantic Reinterpretation of Binding The- the Austronesian Formal Linguistics Association
ory. in: «Natural Language Semantics», vol. XIII, (AFLA), University of California, Santa Cruz
n. 1, 2005, pp. 1-92. 2009, pp. 45-59), who proposes to link analogous
13
Notice that Italian is a Null Subject (or pro- “upside-down” coreference options in Samoan to
drop) Language, i.e., a language allowing phoneti- “backward control”, another “upside down” case
cally null pronominal subjects. With respect to of referential dependency (see M. POLINSKY, E.
the non-coreference effect, the null pronominal POTSDAM, Backward Control, in: «Linguistic In-
subject of Italian behaves exactly as the overt quiry», vol. XXXIII, n. 2, 2002, pp. 245-282).
pronominal subject of English. Whatever other property may be linked to “up-
14
See H. DAVIES, Cross-linguistic Variation in side-down” coreference, learnability considera-
Anaphoric Dependencies: Evidence from the Pacific tions suggest that the special parametric value
Northwest, in: «Natural Language and Linguist permitting it must be a marked property, as-
Theory», vol. XXVII, n. 1, 2009, 1-43. sumed to hold by the language learner only in the
15
Ivi, p. 20. presence of evidence directly justifying the choice
16
See L. MATTHEWSON, H. DAVIS, D. GARDINER, of this option, which is otherwise not entertained
Coreference in Northern Interior Salish, in: W. Shipley by the learner.You can also read
