The big fish in a small pond: Accommodation and the processing of novel definites
Lyn Frazier
University of Massachusetts, Amherst
Abstract
Definite phrases are often used to introduce new entities into discourse but the conditions
under which this is possible are not well understood. Evans (2005) argued that the definite
article may be used to introduce an entity if it is the most relevant of its type in its local frame of
reference, dubbing this the “Small world hypothesis.” Additional arguments will be given in
support of the Small world hypothesis, and the hypothesis will be developed in an attempt to
understand the role of the small frame in processing novel definites. It will be argued that the
definite article does not carry a familiarity presupposition, but only an existential and maximality
presupposition. Maximize presupposition is also important and plays a critical role in regulating
the impact of the Small world hypothesis: a definite will be favored under circumstances where it
may be used unless it brings along unwanted implicatures, typically implicatures concerning
number. Other types of presupposition accommodation are also considered with an eye to
evaluating the claim that distinct classes of presuppositions exist.
-1-

---

-2-

---

The big fish in a small pond: Accommodation and the processing of novel definites
Lyn Frazier
University of Massachusetts, Amherst
Preliminaries.
Speakers sometimes act as if the common ground includes a proposition that hasn’t been
explicitly introduced. Hearers may accommodate the proposition, or challenge it, as in (1).
(1)
A: John went to Paris again.
B: Again? I didn’t know he went once.
Presupposition triggers that imply a proposition is part of the common ground include the,
another, stop, more, again, too. It has been suggested that presupposition triggers may not all
behave alike (Abusch, 2005). Schwarz (2005), for example, showed that German readers will
perform costly syntactic reanalyses in order to satisfy the presupposition of auch (‘too’), but it is
unclear whether this is a general feature of presupposition triggers or only of particular ones that
are difficult to accommodate.1
The present paper will focus on the use of the definite determiner the to introduce novel or
unmentioned entities into discourse, as in (2).
(2)
Jason bought a Honda. The steering wheel was positioned oddly.
Section 1 considers various linguistic approaches to novel definites from a psycholinguistic
viewpoint. Section 2 provides an overview of psychological work on ‘bridging’ inferences and
the processing of novel definites. Section 3 introduces Will Evans “Small world hypothesis,”
which will be the starting point for my own proposal. The proposal will be developed in Section
1There is also the issue of whether there are cross-language differences in the implication
carried by a presupposition trigger. Matthewson (2005) claims Lillooet Salish don’t challenge
presupposition failures except in outrageous cases (‘nukw’=some but not all, in ‘A sun rose.’).
She tentatively suggests that the presupposition must only be contained in the objective
propositional context in Lilloet.
-3-

---

4. Section 5 explores the small world, and Section 6 presents some experimental evidence for
the proposal. In Section 7, we will return to the question of whether distinct presupposition
triggers differ in a principled way, and raise the question of whether special processing principles
govern the accommodation of unsatisfied presuppositions.
1. Linguistic theories from a psycholinguistic perspective
One approach to novel definites is to assume that a restrictor of the nominal has been
elided, e.g., in (2) the unelided form might be ‘the steering wheel of the Honda that John bought.’
(see discussion of syntactic, semantic vs pragmatic accounts in Stanley and Szabo, 2000, who
argue that syntactic and pragmatic accounts of implicit domain restrictions won’t work; instead
they argue for a semantic account where each nominal comes equipped with a context variable).
The problem usually raised for the ellipsis account is that it is often unclear what restrictor has
been elided (Abbott, 2004).
In general, it seems somewhat difficult to find an antecedent for an elided constituent more
than one sentence back, e.g., in the case of verb phrase ellipsis. But, as illustrated in (3), there is
no difficulty with a novel definite (the steering wheel) where the descriptive content of the
restrictor would come from a sentence (John bought a car) more than one back from the sentence
containing the novel definite.
(3)
John bought a car. He’d just inherited some money. But the steering wheel was placed in
an awkward position, so he plans to sell the car soon.
Similarly in general focused antecedents seem to be preferred for elided constituents (e.g.,
Carlson, 2002, Frazier, Clifton and Carlson, to appear, for focus and VPE). But in the case of
novel definites, there does not seem to be a preference for the material supplying the antecedent
to come from a focused constituent, as shown in (4) and (5) (and their plausibility ‘controls’
(4',5')). There does not seem to be a preference to interpret the official as the official of the
information booth in (4), but as a police official in (4'), or to interpret the first page as the first
page of the novel in (5) but as the first page of the magazine in (5').
(4)
Juan drove up to the INFORMATION booth; his friend went on to the police station. The
official was helpful.
-4-

---

(4')
John drove up to the POLICE station; his friend went on to the information booth. The
official was helpful.
(5)
Of course, Lyla opened a NOVEL right away. Maria picked up a magazine eventually.
The first page was ripped.
(5')
Of course, Lyla opened a MAGAZINE right away. Maria picked up a novel eventually.
The first page was ripped.
Although the evidence is not decisive, at the very least these observations suggest that processing
considerations do not particularly point in the direction of an ellipsis account of novel definites.
Another approach to novel definites is based on the assumption that definites carry a
familiarity presupposition. The ‘matching scorecard’ account of Lewis (1979) was based on the
claim that the carries a familiarity presupposition. The account emphasized that the speaker and
hearer’s assumptions should match. If the speaker acts as if a discourse entity is familiar, the
hearer should ‘accommodate’ the presupposition to make her common ground match that of the
speaker. As Abbott (2000, 2004) pointed out, the concern then is that accommodation should
vitiate the effects of familiarity altogether. Heim (1982,1983) takes up this challenge by
assuming that a familiarity approach could be coupled with limited accommodation. Specifically
she suggested that accommodation may only be possible where bridging to an explicitly
introduced discourse referent is possible.
This approach would be tempting if we had a suitably restricted theory of bridging
inferences. But to my knowledge, no such theory exits. A bridging inference is simply any
‘backwards’ inference to already processed material drawn in order to maintain discourse
coherence. Further, in many acceptable examples, such as (6) or (7), no explicitly introduced
discourse referent seems to support the introduction of a new discourse referent, such as the
tolltaker or the coffee. I suppose that in (6a), for example, one might say that the time introduced
by 6am counts as an explicitly introduced discourse referent and thus the coffee may be bridged
to it. But in this case it is not so clear what is being excluded by the condition that bridging
must be to an explicitly introduced discourse referent, since in most contexts, a time and location
will be available to be used for bridging purposes. Example (6b) suggests that a particularly
characteristic time isn’t needed in order to obtain successful accommodation of a novel definite,
-5-

---

at least assuming that noon is not a particularly characteristic time for drinking coffee. Example
(7) illustrates that a particularly characteristic location isn’t needed, assuming that most roads
don’t have toll takers.
(6)
a.
Maria got up at 6am. The coffee was already made, and Juan was leaving for work.
b.
Maria got up at noon. The coffee was already made, and Juan was leaving for work.
(7) Rex was already upset when the road narrowed from eight lanes into just four. Then the
toll taker was rude to him. He ended up in a terrible mood.
Further, it isn’t clear how a familiarity-based account, even one with limited accommodation,
would explain the acceptability of novel definites in quantificational contexts, such as (8), or in
‘telescoping’ examples like (9), if bridging is understood as introducing an entity into a mental
model representation (i.e., some post-semantic representation).
(8)
In most of John’s travels, he offends the customs official.
Indeed, in a written acceptability judgment with Chuck Clifton, we found no general cost for a
universal quantifier (every) as opposed to the definite article (the) in telescoping contexts where
the universal quantifier would induce a multiple interpretation of a definite phrase in a following
sentence. Whether the telescoped sentence was rated as less acceptable or more acceptable than
its control depended on the particular sentence. We could not convincingly pin down which
properties were most important for facilitating telescoping. Howewer, Anderssen (2004)
presented acceptability judgment data showing that the presence of a ‘generic’ relation, in the
sense of a causal connection that could support a general relation, is crucial for acceptable
telescoping examples but is not important for their definite description counterparts.
(9) Experiment with Clifton: Written rating study on scale from 1 (not very natural) to 5 (very
natural)
a.
Every truck driver will approach the tollbooths nervously.
The tolltaker will check a list of trucks with unpaid license fees. 3.706
b.
The truck driver will approach the tollbooths nervously.
The tolltaker will check a list of trucks with unpaid license fees. 3.693
One approach to these data is to assume that what is being quantified over is situations, not
entities of the type of the nominal. This would be very compatible with the view advocated in
-6-

---

Section 4.2
Another potential difficulty for the familiarity account of definites is the frequency with
which novel definites are encountered. In a spoken corpus, Spenader (2002) found that 40% of
definites are novel. If the figure were, say, 4%, then a familiarity account would seem more
tempting. Viewing accommodation as patching-up of discourse when the familiarity
presupposition is not satisfied is somewhat less attractive if it is not exceptional for definites to
be novel.
Another important account of definites is based on the assumption that the carries a
uniqueness presupposition. Hawkins (1991) proposed an insightful account of this type, based
on the idea of pragmatic sets (‘p-sets’) plus a uniqueness presupposition (‘inclusiveness’). He
argued that we must “postulate a rich pragmatic structuring of entities within our model” (p. 408)
and went on to give examples involving the immediate situation of utterance (“Pass the bucket”
if there’s a single bucket in the speaker’s visual field) and examples involving associative
relationships between entities (classroom, professor, student). The core idea was that uniqueness
holds within the pragmatic set. The notion of a more structured model is appealing and, in many
respects, Hawkins’ claim is related to the one proposed in Section 5 below. The problem with
the account, as with all accounts including the one offered here, is the problem of developing
sufficiently explicit constraints on the structuring of context, or on the definition of p-sets.
Further, examples like (10) indicate that there are counterexamples to the uniqueness claim
unless p-sets are defined strictly with respect to the structure imposed on the model in response
to the linguistic material uttered before the novel definite. Uniqueness does not hold in the larger
situation described by the discourse once we include the entire sentence containing the novel
definite.3
2Florian Schwarz pointed out to me that if one assumes that the small frame introduced in
Section 4 is identified as a topic situation, then one may unify the account of telescoping (as
quantifying over situations) with (most of) the observations concerning small frames. This
seems very promising to me. However, I am not a formal semanticist and thus must leave these
issues to others more qualified than myself. See in particular Kratzer (2005).
3Heim (1990) and Elbourne (2005) discuss the example in (a), which may be similar to
(10). (Thanks to Florian Schwarz for pointing this out to me).
-7-

---

(10) I called the Photo shop about their photography classes. The clerk asked me to hold for
another clerk who knew the schedule.4
In summary, an ellipsis approach to novel definites doesn’t readily fit with the observation
that, with novel definites, the anchor or supporting context can be more than one sentence back
and focus doesn’t seem to strongly influence the choice of anchor. Both properties differ from
what is expected for an ellipsis structure. A bridging inference approach doesn’t seem to place
any substantive constraints on when novel definites are acceptable, even if one adds the
requirement that bridging must be to an explicitly introduced discourse entity (examples like (7b)
suggest that either the requirement is incorrect, or it must be interpreted in a sufficiently weak
manner that it is not clear what is excluded). Novel definites in telescoping environments do not
necessarily create unacceptability or any obvious processing problem, suggesting that the local
situation of evaluation is important in licensing the accommodation of definites and suggesting
quantification in htese examples may be over situations.
2. Processing background
Perceivers construct inferences when a discourse contains repeated noun phrases, anaphoric
dependencies between phrases or causal relations between sentences (or, more accurately, causal
relations between the situations they describe). In their classic study, Haviland and Clark (1974)
found that a target sentence containing a definite (the beer) was read faster if the context sentence
explicitly mentioned some beer, as in (11b) than if it merely set up a likely scenario where beer
might be expected, as in (11a).
(a) If a bishop is in the same room as another bishop, he blesses him.
However, (a) seems to me to have a ‘generic’ quality which is entirely lacking in the episodic
sentence in (10). Whether this matters for accommodation, I’m not sure. But I could imagine
that (a) might involve quantifying over situations whereas (10) may not.
4One might try to account for (10) by assuming that the embedded clause introduces a
new frame or a new situation. But that does not seem to be a crucial property of examples like
(10). Example (a) below seems relatively acceptable even though the second clerk is introduced
in the same clause as the novel definite.
(a) I waited in line at the Photo Shop. The (overworked) clerk asked another clerk to help me.
-8-

---

(11) a.
We checked the picnic supplies.
(Given but bridging inference required)
b.
We got some beer out of the trunk.
(Given and explicitly mentioned)
c.
Andrew was especially fond of beer.
(Control for lexical repetition, inference
required)
Target: The beer was warm.
In general, elaborative inferences (forward inferences, explicit inferences about predictable
events) are not drawn in advance, see especially McKoon and Ratcliff. Consequently, reading
time is longer for the final sentence (containing the shovel) in (12) following (12b), which
doesn’t mention a shovel, than in (12a), which does, as shown by Singer (1979).
(12) a.
The boy cleared the snow with a shovel.
b.
The boy cleared the snow from the stairs.
The shovel was heavy.
Note: The instrumental (shovel) inference is elaborative if it occurs before the definite (the
shovel) is encountered; it is anaphoric or ‘bridging’ (Clark, 1977) if it occurs afterward.
However, the data from eye movement recording studies suggest that the degree of contextual
constraint determines the processing time for novel definites. In highly constraining contexts,
definites take no longer to read with implicit antecedents than with explicit ones, but in low
constraint contexts they do. In an eye movement recording study, O’Brien et al. (1988)
investigated the processing of examples like (13), where a definite (the knife in the final
sentence) was mentioned (13a,c) or not, and the context was highly constrained (13a,b) due to the
verb (stabbed) or less constrained (13c,d, containing assaulted).
(13) All the mugger wanted was to steal the woman’s money. But when she screamed, he
stabbed [assaulted] her with his knife [weapon] in an attempt to quiet her. He looked to see
if anyone had seen him. He threw the knife in the bushes and ran away.
a.
Highly constrained context, explicit referent: stabbed her with a knife
b.
Highly constrained context, implicit referent: stabbed her with a weapon
c.
Less constrained context, explicit referent: assaulted her with a knife
d.
Less constrained context, implicit referent: assaulted her with a weapon
O’Brien found that reading times were longer following sentence (13d) than (13c), but not longer
-9-

---

following (13b) than (13a), indicating that the novel definite incurred a processing cost only with
the less constraining context. In principle this could be explained either as a counterexample to
the generalization that forward inferences aren’t drawn or as indicating that backward inferences
can be so cheap in sufficiently constrained contexts that their cost cannot be measured with
current methods. Interestingly, in speech, there appears to be a penalty even in the highly
constraining context if there’s not a pitch accent on the noun introducing an unmentioned entity,
i.e., a penalty when there’s no pitch accent on the novel definite (Birch and Clifton, 2004).
In the ERP literature, the N400 (a negativity peaking 400ms after the onset of the critical
word) has often been taken to be an index of ‘semantic’ integration since it is correlated with the
predictability of a word in its context (van Berkum et al., 2003). A late positivity (P600) has
been associated with syntactic reanalysis and other complex sentence-level processing
operations. In an ERP study of bridging inferences, Burkhardt (2005) tested sentences like those
in (14) where a critical definite (the conductor) appeared in a context which either explicitly
mentioned the corresponding entity, implied the existence of the entity (a ‘bridging’ context) or
did neither.
Materials presented 400ms/word with 100ms ISI.
(14) a.
Given context
Tobias besuchte einen Dirigenten in Berlin. ‘Tobias visited a conductor in Berlin.’
b.
New context
Tobias unterhielt sich mit Nina. ‘Tobias talked to Nina.’
c.
Bridging context (concert-conductor)
Tobias besuchte ein Konzert in Berlin. ‘Tobias visited a concert in Berlin.’
Target:
Er erza”hlte, dass der Dirigent sehr beeindruckend war. ‘He said that the conductor
was very impressive.’
She found that given DPs showed reduced negativity (N400) and no positivity. New DPs
elicited an enhanced negativity (N400) followed by left posterior positivity (P600). Bridged DPs
(14c) patterned first with given DPs (exhibiting a reduced N400), then with new DPs, as
-10-

---

Document Outline

• Page 1
• Page 2
• Page 3
• Page 4
• Page 5
• Page 6
• Page 7
• Page 8
• Page 9
• Page 10
• Page 11
• Page 12
• Page 13
• Page 14
• Page 15
• Page 16
• Page 17
• Page 18
• Page 19
• Page 20
• Page 21
• Page 22
• Page 23
• Page 24
• Page 25
• Page 26
• Page 27
• Page 28
• Page 29
• Page 30
• Page 31
• Page 32
• Page 33
• Page 34
• Page 35
• Page 36
• Page 37
• Page 38
• Page 39

---