1
817
21
22
23
25
24
26
27
28
29
30
31
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
70
69
74
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1
1
7
WHEN SEMANTICS MEETS PHONETICS: ACOUSTICAL STUDIES OF
SECOND-OCCURRENCE FOCUS
DAVID BEAVER
University of Texas at
Austin
FLORIAN JAEGER
University of Rochester
BRADY ZACK CLARK
Northwestern University
EDWARD FLEMMING
Massachusetts Institute of
Technology
MARIA WOLTERS
University of Edinburgh
A second-occurrence (SO) focus is the semantic focus of a focus-sensitive operator (e.g. only),
but is a repeat of an earlier focused occurrence. We report on the first systematic production and
perception experiments to show that SO foci occurring after a nuclear accent are, as Rooth (1996b)
has claimed, prosodically marked. We find that (i) there is no mean pitch rise on SO foci, (ii)
SO foci are marked by longer duration and greater energy, and (iii) listeners are able to detect
the difference between SO foci and nonfoci. On the basis of these results, we argue that SO focus
is compatible with theories of focus interpretation that it has been claimed to contradict.*
1. INTRODUCTION. In studying information-structural notions like focus, topic, and
contrast, semanticists depend on phoneticians and phonologists to categorize the types
of marking available for interpretation. If only it were that simple! Despite enormous
progress in the last few decades, the development of theories of prosody has been
hindered by an absence of clear evidence as to what the underlying semantic distinctions
are. As Ladd (1996:102) writes: ‘For the present, proposals about intonational meaning
are not a reliable source of evidence on intonational phonology’.1 Our point of departure
is that both the phonological study of prosody and the semantic study of information
structure stand to benefit from collaboration between semanticists and laboratory phonologists.
That prosodic marking affects truth-conditional meaning was observed by Paul (1888:
312ff.). Consider our examples in 1, where small caps indicate prosodic prominence
(on either money or Bill), and this emphasis is standardly described as marking focus.
In English, focus is typically marked by a nuclear pitch accent, that is, the last pitch
accent in a phonological phrase (see Cohan 2000 and Ladd 1996:45–46). The two
different prosodic realizations of the same string of words yield truth-conditionally
different semantic interpretations. In a situation where Jan gives Bill and Malachi
money, and gives nobody anything else, 1a is true while 1b is false.
(1) a. Jan only gave Bill MONEY.
b. Jan only gave BILL money.
* This work was partially funded by a Stanford University Office of Technology Licensing grant held by
David Beaver, and by the Scottish Enterprise Stanford-Edinburgh Link grants ‘Sounds of Discourse’ and
‘Synthesis’. This article has benefited from commentary of Language referees and editors, from discussions
with Katy Carlson, Bob Ladd, and Mark Steedman, from audiences at conference presentations at the Linguistic Society of America annual meetings at Atlanta in 2003 and Boston 2004, the Phonologie und Phonetik
Workshop at Potsdam in 2004, and the Texas Linguistic Forum at Austin in 2004, and from numerous
seminar audiences. Our greatest intellectual debt is to Mats Rooth, whose insightful observations motivated
us to perform an experimental study.
1
The reader should set Ladd’s comments against a background of significant advances in the study of
prosody in the previous twenty years. These advances are exemplified by the development of prosodic
transcription standards (e.g. the ToBI system) and the development of several corpora with prosodic transcriptions (e.g. the BRN corpus; Ostendorf et al. 1995). Note that the term INTONATION (and, similarly, INTONATIONAL PHONOLOGY) is often used with a broader meaning than the etymology would suggest, and includes
suprasegmental features other than pitch, such as duration and energy. The term PROSODY reliably includes
all suprasegmental features.
251
1
312
252
77
Many standard approaches to grammar separate phonology and semantics into components that cannot exchange information directly. In such approaches, the relationship
between prosody and meaning is mediated by syntax. Therefore, many authors, from
Halliday (1967) and Chomsky (1972) on, have concluded that there are syntactic features that encode prosodic prominence. Chomsky (1976), for example, suggested that
all focused phrases move outside their base position in syntax. Related ideas are present
in recent work in the principles-and-parameters framework; see, for example, Kayne
1998 and Tancredi 1990. Many other authors also postulate some syntactic effect of
prosodic focus, though they might use features or derivations in a quite different way;
see, for example, Krifka 1992, Rooth 1985, Steedman 2000, and von Stechow 1985/
1989.
What all of these accounts have in common is that they accept the premise that there
is a grammatically mediated interaction between focus and meaning, albeit that this
linkage may involve several stages. Yet recently some have suggested that the interaction between focus and meaning is not grammatically mediated at all; see Büring 1997,
Dryer 1994, Kadmon 2001, Roberts 1996, Schwarzschild 1997, Rooth 1992, and Williams 1997. These authors have suggested instead that many effects of focus on meaning
are pragmatic. This view strikes such a radically discordant note with the approaches
listed earlier that it is worth exploring whether the claims of the pragmaticists are
justified. The current article deals with one particular claim that the pragmaticists have
used as the cornerstone of their argument. Specifically, they have claimed that in certain
environments the interpretational effects normally attributed to focus are found even
without the phonological prominence that marks focus, and thus that those interpretational effects need to be accounted for by a mechanism that is independent of such
focus marking.
Pragmatic theories of focus could be seen as part of an enterprise of keeping the
interfaces between phonology, morphosyntax, and semantics as narrowly circumscribed
as possible. These pragmatic theories lean on extragrammatical mechanisms to make
up the shortfall, much as in the model of Grice (1989). If the interpretive effects of
focus could be explained pragmatically, then the phenomenon of focus would provide
us with little insight into how suprasegmental information is represented at the
morphosyntax/phonology and semantics/morphosyntax interfaces. But if there is a
grammaticized connection between focus and meaning, then that places minimal constraints on the suprasegmental information that must be represented at the
morphosyntax/phonology interface. It also places a lower limit on what information
must be passed between syntax and semantics. What we show here is that the major
argument that has been put forward in favor of pragmatic theories of focus is empirically
flawed. As far as we can tell, there is no evidence that suprasegmental information like
that marking focus is treated differently from segmental information as regards the
architecture of the grammar: both contribute to meaning, and there is no a priori reason
to view them as contributing in substantially different ways.
The specific phenomenon at issue here has been studied extensively in the semantic
literature and involves expressions that are FOCUS-SENSITIVE. An expression is focussensitive if its interpretation is dependent on the placement of focus, keeping in mind
that the linguistic realization of focus (e.g. by prosody, syntactic position, or morphology) varies crosslinguistically. All theories of focus-sensitivity agree that focus interacts
with expressions like only in linguistic contexts like that exemplified in 1. A special
case of the controversy over whether the interpretation of focus is mediated via syntax,
however, involves disagreement as to how grammaticized the relationship between only
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
1125
1
7
LANGUAGE, VOLUME 83, NUMBER 2 (2007)
1
231
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
144
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
WHEN SEMANTICS MEETS PHONETICS
and its associated focus is (Partee 1999:215ff.). Does the lexical entry of only stipulate
association with a focused constituent in its syntactic scope, or are there contexts in
which the interpretational effects illustrated by 1 occur without focus being present
either phonologically or syntactically?
Examples are given in the literature that involve an apparent dislocation between
the interpretation of only and the position of prosodic prominence. Most of these examples fall under the rubric of SECOND-OCCURRENCE (SO) focus, where a repeated focused
item apparently lacks a pitch accent.2 Example 2, adapted from Partee (1999:215),
illustrates SO focus. The two sentences are to be read as a dialogue. The convention
of indicating the focused item with a subscripted ‘F’ is widespread in the syntactic and
semantic literature, and, as we discuss shortly, is usually taken to indicate marking at
a syntactic level, while our ‘SOF’ marking allows us to remain (temporarily) agnostic
as to whether the item described as a SO focus is in fact focused at a syntactic level
or any other.
(2) A: Everyone already knew that Mary only eats [vegetables]F .
B: If even [Paul]F knew that Mary only eats [vegetables]SOF , then he should
have suggested a different restaurant.
Partee (1999) summarizes the problem succinctly. If only is a focus-sensitive operator
(i.e. needs a prosodically prominent element in its scope), then the two occurrences of
only eats vegetables in 2 should have the same analysis. But if there is no phonological
reflex of focus in the second occurrence of vegetables, then this leads to the notion of
‘phonologically invisible focus’. The notion of inaudible foci ‘at best would force the
recognition of a multiplicity of different notions of ‘‘focus’’ and at worst might lead
to a fundamentally incoherent notion of focus’ (Partee 1999:215–16).
Following up on observational evidence presented by Rooth (1996b), we present
experimental data concerning precisely the cases where the pragmaticists have claimed
interpretational effects without phonological marking. We show that in these cases the
claim that there is no phonological marking is wrong. Specifically, we present the
results of multisubject production and perception experiments in which we examine
the acoustic correlates of SO focus in the scope of operators only and always, which
are both assumed to be focus-sensitive in the bulk of prior literature.3 Our results
confirm that there are acoustic correlates involving duration and energy of the focused
item, and show that the acoustic correlates are perceptible. These conclusions make
clear that the phenomenon of SO focus does not support an argument that current
syntactic and semantic theories of focus are ‘fundamentally incoherent’.
For semantics and pragmatics, our results contribute to a long-running debate. But
the phenomenon of SO focus is not well known in phonetics and phonology. The results
are of relevance to these fields because they shed light on the acoustic realization of
focus and the acoustic properties of the POSTNUCLEAR domain, the part of an intonational
phrase that follows the nuclear accent. As becomes clear, the results (i) imply an increased role for phrasal STRESS in marking focus, while showing that pitch accents
are not required, and (ii) suggest that although postnuclear deaccenting restricts the
appearance of pitch accents, phrasal stress distinctions must be permitted in the postnu-
1428
2
1429
1430
1431
11432
3
1
7
253
See Gussenhoven 1984 and Hajičová 1973, 1984 for related discussion.
But see discussion in §4: work conducted since our original production study indicates that only may
be more prototypical as a focus-sensitive operator than always. To look ahead, the experiments reported
here did not yield significant effects in support of this distinction, but this may be because our experimental
design was not intended to reveal a distinction between the two.
1
312
254
174
clear stretch. This latter point buttresses arguments Ladd (1996) has put forward for
the existence of postnuclear prominence, although our conclusions differ in detail from
his.
The article is organized around three hypotheses: first, that SO focus is prosodically
distinguished from nonfocus; second, that SO focus is marked differently from other
foci; and, third, that SO focus marking is perceptible. We thus first provide the technical
background of the article as regards the syntax, semantics, pragmatics, and phonetics/
phonology of focus. The production and perception experiments are then presented,
followed by a more general discussion that connects the results to wider theoretical
concerns and possibilities for future work.
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
LANGUAGE, VOLUME 83, NUMBER 2 (2007)
2. BACKGROUND. The following discussion involves a range of phonetic, phonological, syntactic, and semantic notions of focus. These notions include PROMINENCE (a
psychoacoustic notion), PHONOLOGICAL FOCUS MARKING (which in prior literature has
sometimes been referred to as PROSODIC or INTONATIONAL focus), F-MARKING (which is
syntactic), and SEMANTIC FOCUS (an aspect of the representation of meaning).4 We must
now explain why the absence or presence of certain acoustic properties has been claimed
to be a decisive factor for theories of focus.
208
209
210
211
212
213
214
215
2.1. THE PHENOMENON OF FOCUS-SENSITIVITY. Focus-sensitivity can be thought of as
a mechanism that allows two-place operators to select their arguments. Consider the
English exclusive only, which is sometimes treated as a universal quantifier. The parallel
with the canonical universal every is seen in examples like 3a,b, which involve only
within an NP (hence ‘NP-only’). The semantic interpretation of 3a to 3b is related in
4.
(3) a. Only felines are immortal.
b. Every immortal is feline.
(4) ONLY(FELINE)(IMMORTAL) ⬅ EVERY(IMMORTAL)(FELINE)
In 4, ONLY represents the meaning of the word only. Let us call the first argument
of ONLY its RESTRICTOR, that is, FELINE, and the second its SCOPE, that is, IMMORTAL.
Then only is just like every except that the operator we use to represent its meaning
has its arguments reversed.5 In the case of NP-only, configuration fully determines the
restrictor and scope, but, as discussed below, the situation is more complex for VPonly.
The standard mechanisms for argument selection are (i) syntactic configuration and
(ii) morphology (e.g. case marking). In English, the burden of selection falls primarily
to configuration. However, focus-sensitive operators like VP-only are unusual in that
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
11445
4
Note that the term FOCUS has been used to refer to a variety of semantic phenomena in addition to the
association with a focus-sensitive operator (as in 1 above), including contrastive focus, focus in questionanswer pairs, and focus in clefts or left dislocations. These foci are sometimes taken to share phonological
features, but this is far from obvious (cf. Bartels & Kingston 1984, Hedberg 2003, and Hedberg & Sosa
2001.
5
For evidence that only and every are related in this way semantically, see, for example, de Mey 1991
and Horn 1996. Note that on some approaches, the focus of only would appear in only’s specifier at some
point in the syntactic derivation (Kayne 1998, Krifka 2006). In this case, NP-only might combine first with
the argument we call here its scope, and second with the argument we call here its restrictor, making our
terminology less than transparent for adherents of these approaches. It is presumably because of the nonstandard way in which only combines with its arguments that terminology has not become standardized. Here we
stick to a terminology that takes the analogy between NP-only and every as basic, and labels arguments in
other cases by extrapolation from this case.
191
192
193
194
195
196
197
201
204
207
1
7
1
231
WHEN SEMANTICS MEETS PHONETICS
255
265
their arguments are not determined configurationally, at least not in surface form. Indeed, many analyses do not even assume that the arguments of focus-sensitive operators
need to be continuous surface constituents. Consider 5a,b (repeated from 1a,b with
explicit F-marking) and their intended meanings 6a,b.
(5) a. Jan only gave Bill [money]F .
b. Jan only gave [Bill]F money.
(6) a. ONLY(MONEY)(␭x[JAN GAVE BILL x])
i.e. everything Jan gave Bill was money.
b. ONLY(BILL)(␭x[JAN GAVE x MONEY])
i.e. everyone Jan gave money was Bill.
If 5b is analyzed as in 6b, then the restrictor is the interpretation of Bill, while the
scope is derived by interpretation of Jan gave . . . money, which is discontinuous.6
Common approaches to argument selection neither allow for discontinuous arguments,
nor explain how focus could play any role. The puzzle of focus-sensitivity is this: what
mechanisms explain the correlation between the differing placement of focus in 5a,b
and the differing interpretations in 6a,b?
A wide range of English expressions has been analyzed as focus-sensitive, including
exclusives (only, as in the above examples, but also just, merely, solely), additives (too,
also), scalar additives (even), particularizers (in particular, for example), intensifiers
(really, totally), quantificational adverbs (usually, always), certain quantificational determiners (many, most), comparatives, some sentence connectives (because, counterfactual if-then), negation, questions, and emotive predicates (regret, be glad that)
(Hajičová et al. 1998 Rooth 1996a).7
Clearly not all focus-sensitive expressions are quantifiers, which makes it inappropriate to label the arguments of those expressions as restrictor and scope. All of the
expressions can be thought of semantically as multi-argument operators, however, such
that one argument position is filled by a focused expression. We term this argument
the SEMANTIC FOCUS of the operator.8 Thus, according to the terminology we introduced
previously, the semantic focus of only is its restrictor, and the puzzle of focus-sensitivity
may be stated more broadly than above: how do language users associate phonologically
focused constituents with the semantic foci of focus-sensitive operators?
Let us summarize just what is in need of explanation. First, sometimes an operator
can select for an argument (its semantic focus) even when the argument is separated
from the operator. Second, there is no morphological case marking or agreement to
facilitate the link between operator and argument, and neither is there any canonical
configurational relationship in surface form that determines what the argument of the
operator is. But there is a correlation between the selection of one of the operator’s
arguments and the placement of phonological focus marking.
1446
1447
1448
1449
1450
1451
1452
1453
1454
11455
6
It might be argued that JAN should be kept out of the argument of only in 6a,b, in which case we could
think of the meaning of 5a, for example, as underlyingly consisting of separate NP and VP meanings combined
conjunctively: ONLY(MONEY)( ␭x[SUBJ GAVE BILL x]) ⵩ SUBJ ⳱ JAN. The interpretation of the subject in
sentences with VP focus-sensitive operators is not a concern of this article. See Beaver & Clark 2003 for
an analysis.
7
It must be stressed that this is not a phenomenon peculiar to English: see, for example, the many Dutch
particles discussed by Foolen (1993).
8
Note that semantic focus is logically distinct from the phonological notion of focus marking, although
the experiments reported here suggest that a large class of semantic foci must be focus-marked. See the
introduction to §2.
216
217
218
219
220
224
227
231
233
236
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
1
7
1
312
256
266
2.2. GRAMMATICAL MECHANISMS FOR FOCUS-SENSITIVITY. It is natural to think of focus
as having a function analogous to that of a morphological case marker: it signals which
argument goes where. Thus, many theorists have assumed that a focused constituent
is marked at the level of syntax, normally with a special ‘F’ feature, and have then
postulated that a further grammatical mechanism explains focus-sensitivity. The details
of the mechanism depend on whether the F-marked constituent is moved or interpreted
in situ.
Early proposals suggesting that F-marking triggers movement were those of Anderson (1972) and Chomsky (1976). In the minimalist program (Kayne 1998), the focussensitive operator only can attract the F-marked Bill in 5b to the specifier of the phrase
that only heads.9
In situ approaches became dominant in the semantics community after Rooth (1985)
observed a range of problems with movement accounts.10 But in order to achieve in
situ interpretation, the semantics must depart quite radically from the now-canonical
compositional account of Montague (1974). Loosely, we may say that in situ approaches, while they do not move the F-marked constituent itself, do move the meaning
of the F-marked constituent, passing that meaning from node to node on the syntactic
tree until it becomes accessible to a focus-sensitive expression like only. Rooth’s (1985,
1992) ALTERNATIVE SEMANTICS adds to every single constituent a second semantic value
for the focus in addition to the standard one, while yet another modification to Montague
grammar is made in the STRUCTURED SEMANTICS of von Stechow (1985/1989) and Krifka
(1992).11 For current purposes, the details of the modified semantics are not important.
What is important is that in some way the modifications mean that the semantics encodes
which parts of the meaning come from focused constituents, and which parts of the
meaning come from nonfocused constituents.
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
11466
1
7
LANGUAGE, VOLUME 83, NUMBER 2 (2007)
2.3. STRONG THEORIES OF FOCUS-SENSITIVITY. The in situ and movement models described above explain the interpretational effects which we term focus-sensitivity in
terms of grammatical mechanisms conditioned on syntactic F-marking. Two questions
arise with regard to the predictions of a model of this sort. First, if a constituent in an
appropriate position relative to a focus-sensitive operator is F-marked, does it necessarily become the semantic focus of that operator? Second, and more relevant to this
article, is the semantic focus of a focus-sensitive operator necessarily F-marked?
It is hard to find explicit discussions of these questions in the early literature on
focus-sensitivity. Nonetheless, we are not aware of any syntactic proposal that would
allow material with no associated F-marking to undergo focus movement. As regards
in situ approaches, it seems fair to say that prior to Rooth 1992, it had been widely
assumed both that F-marked constituents were obligatorily interpreted as semantic foci,
and that semantic foci were obligatorily F-marked. And such assumptions seem hardly
9
In Kayne 1998, the entire VP gave Bill a book moves into the Spec of only. It is unclear what Kayne’s
preferred analysis of the semantic association with a focus within the VP would be. Note that Krifka’s
(2006) hybrid association with focus account is comparable to Kayne’s but assumes that a separate semantic
mechanism of focus association operates in tandem with the syntax.
10
Rooth’s (1985) empirical arguments against movement-based accounts include the interaction of focussensitive expressions with multiple foci and syntactic islands. Krifka (2006) suggests that focus-sensitive
operators associate through movement of the syntactic island that contains the associated focus, thus sidestepping some of Rooth’s objections to movement-based accounts.
11
Note that the analysis of focus given by Steedman (2000), although it involves structured semantics,
does not depend on syntactic F-marking. For Steedman, intonation controls derivation rather than syntactic
form.
1
231
WHEN SEMANTICS MEETS PHONETICS
257
341
surprising: if there are some environments where the semantic focus is not what is
F-marked, then yet another mechanism must be postulated to explain focus-sensitivity.
The very fact that models of focus like those described above require significant revisions to syntax and/or semantics makes it unattractive to suppose that yet further mechanisms are used in some cases.12
Rooth 1992 introduced a more nuanced view, one that allows for F-marking to be
dissociated from the semantic focus. Rooth discussed a spectrum of possible theories
of focus, ranging from weak to strong. The idea is that WEAK theories are those that
involve the most stipulation and are thus only weakly explanatory. Specifically, a weak
theory would involve stipulating item by item which words are members of the class
of focus-sensitive operators, and item by item exactly how each member of the class
operates on the focus. Both the movement-based and in situ models sketched above
would be classified by Rooth as weak, since no general principle constrains which
expressions can be focus-sensitive.
A STRONG theory, in Rooth’s sense, would not involve any grammaticized mechanism
relating focus to argument selection or semantic interpretation. Instead, the theory would
have to rely on general principles of pragmatics and interpretation to derive its predictions about the effects of focus. If, in spite of this handicap, and in spite of not allowing
lexical specifications to make any reference to focus, a theory could still predict focussensitive behavior, the theory would clearly deserve to be called strong.13 But is a
strong theory in this sense really a possibility? In fact, a number of candidate proposals
for strong theories have been put forward, although none of of them have been applied
to a wide range of focus-sensitive expressions.
The general template Rooth presented for strong theories allows that operators that
have been termed focus-sensitive have a free variable; the argument that we have been
calling the semantic focus is introduced lexically as a free variable. An effect of focus
is to make certain discourse objects salient, or to show that an already salient discourse
object is to remain so. For example, I only like [cheese]F would be appropriate when
the set of things the speaker likes is under discussion, and hence this set is salient. In
a strong theory, there is no grammatical rule forcing such a salient discourse entity to
become associated with the free variable, but this can optionally happen. On such a
view, free variables involved in focus-sensitivity are resolved pragmatically in much
the same way that a pronoun optionally becomes associated with a salient discourse
entity: Martı́ (2003) explores this parallel. The tricky task that proponents of strong
theories set themselves is then to make sure that the right discourse object gets tied to
the right free variable at the right time. We do not discuss the details of how pragmatic
models might achieve this.14 What is pertinent to the current enterprise is that such
models would predict that an F-marked constituent might not become the semantic
1467
1468
1469
1470
1471
12
Of course, it is clear that in written language F-marking usually cannot be derived from surface form
alone, which might lead us to wonder to what extent language users really need a special grammatical
mechanism to mediate focus-sensitivity. But then again, there are many obligatory distinctions in spoken
language that are not available in standard orthography. An example is the stress pattern on some bisyllabic
words which are ambiguous between noun and verb in standard written form (to invı́te vs. an ı́nvite).
1472
1473
1474
1475
11476
13
Rooth also considers the possibility of what he terms an INTERMEDIATE theory, one that allows for lexical
stipulation, but only of a very limited sort. Von Fintel’s (1994, 1997) account of quantificational adverbs
may well be an intermediate theory, though it could also be taken as strong.
14
For pragmatic models that Rooth would classify as strong, the reader is referred to Geurts & van der
Sandt 2004, Kadmon 2001, Martı́ 2003, Roberts 1995, 1996, Schwarzschild 1997, and Vallduvı́ 1990.
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
1
7
1
312
258
342
focus, and conversely that the semantic focus could be a constituent that is not
F-marked.
It should now be broadly clear how the strong/weak split relates to the question of
whether there are environments in which semantic foci are not syntactically F-marked.
Weak theories, in the form they have been stated, would lead us to expect obligatoriness
of F-marking.15 Thus they predict that that there are no environments in which semantic
foci are not syntactically F-marked. To complicate matters, it is possible to conceive
of a theory in which the relationship between F-marking and interpretation is grammaticized but optional. We do not attempt to describe such a model in detail. It suffices
to say that the more evidence we find for non-F-marked semantic foci, the more plausible strong, pragmatic theories are going to become, while contrarily, evidence of a
compulsory link between F-marking and interpretation provides support for grammaticized models, despite Rooth’s description of them as weak.
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
383
LANGUAGE, VOLUME 83, NUMBER 2 (2007)
2.4. RELATING THE STRONG/WEAK DISTINCTION TO PHONOLOGICALLY UNMARKED SEWe are almost at the point where we can tie this discussion in with the
main empirical phenomenon to be examined here. But first, there is one further subtle
distinction that we need to amplify: that between phonological focus marking and
syntactic F-marking. We have described above how models may differ with regard to
whether the link between interpretation and F-marking varies according to the environment or is compulsory, but we have not said much about how F-marking relates to
phonological focus marking (and, ultimately, to psychoacoustic features such as prominence). The relationship is not straightforward. Often, a phonologically focus-marked
constituent appears to be F-marked, and that constituent may become the semantic
focus of a focus-sensitive operator. But it is also possible for the F-marked constituent
not to correspond directly to expressions bearing phonological focus marking. For
example, the VP have a dog may be F-marked if dog is the only word that is phonologically focus-marked.
Relating phonological focus to F-marking are theories of FOCUS PROJECTION (e.g.
Selkirk 1995 and Truckenbrodt 1995). The question that we care about here is this: is
the relationship between F-marking and phonological focus optional or compulsory?
If there was considerable optionality in this relationship, then it would be very hard to
ever get data on whether the relationship between F-marking and semantic focus was
itself optional or compulsory. In effect, all we would know would be that somewhere
between phonological focus and semantic focus, some optionality crept in, but we could
not tell whether this should be thought of as occurring in the phonetics-phonology map,
the phonology-syntax map, or the syntax-semantics map. But for our purposes it is
enough that in models of focus projection the principle given in 7 is usually adhered
to.
(7) FOCUS PROJECTION
An F-marked constituent must contain at least one phonologically focusmarked subpart.16
MANTIC FOCI.
1477
1478
1479
15
Beaver and Clark (2003) provide syntactic and semantic criteria for obligatoriness of F-marking and
show that F-marking is obligatory in the scope of some expressions that are usually taken to be focussensitive.
1480
1481
1482
11483
16
Note that what we call F-marking corresponds to what some would describe as undominated F-marking.
Selkirk (1995) distinguishes F-marked constituents that are not dominated by any other F-marked constituent
by the label Foc. Schwarzschild (1999) introduces a constraint FOC (‘A Foc marked phrase contains an
accent’), which corresponds to the principle termed focus projection here.
1
7
1
231
386
387
388
389
390
391
392
393
394
395
396
397
401
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
430
434
438
441
444
445
446
447
11484
1
7
WHEN SEMANTICS MEETS PHONETICS
259
Given that F-marked constituents contain a phonological focus, we can now say
that many weak, grammaticized theories of focus should lead us to expect not only a
compulsory relationship between F-marking and interpretation, but also a compulsory
relationship between phonological focus marking and interpretation. Specifically, such
theories predict that the semantic focus of a focus-sensitive operator should correspond
to a constituent that contains a phonological focus (Partee 1999:217). Of all the predictions made by theories of focus, it is this prediction that has come under heaviest attack
in the semantics literature, and by far the most common basis for the attack is the
phenomenon of SO focus.
2.5. THE ARGUMENT FROM SO FOCUS. We have already seen (an adaption of) an
example of SO focus from Partee, repeated here.
(2) A: Everyone already knew that Mary only eats [vegetables]F .
B: If even [Paul]F knew that Mary only eats [vegetables]SOF , then he should
have suggested a different restaurant.
Let us repeat the analysis of SO focus examples given in §1, but using some of the
concepts now introduced. Sentence B contains two focus-sensitive operators, even and
only. The semantic focus of even is Paul, which is also a phonological focus. The
semantic focus of only is vegetables. But Partee and others note that vegetables in its
second occurrence, that is, the occurrence in sentence B, can be felicitously uttered
without any pitch movement, and hence, they suppose, it cannot contain a phonological
focus. By focus projection, it follows that this occurrence of vegetables is not syntactically F-marked. Thus we appear to have a case of a dissociation between semantic
focus and F-marking, contrary to what is expected on the movement and in situ accounts
of focus sketched above.
It is this type of analysis that has allowed theorists to argue that SO-focus data
counterexemplifies weak theories of focus interpretation. As mentioned previously,
many authors hold that SO focus data provides evidence against the major in situ
focus accounts (alternative semantics and structured meaning semantics). These include
Büring (1997), Dryer (1994), Kadmon (2001), Martı́ (2003), Roberts (1996), Rooth
(1992), Schwarzschild (1997), Vallduvı́ (1990), and Williams (1997), all of whom
interpret SO-focus data as favoring strong accounts of association with focus phenomena over weak accounts.
We now present in a maximally explicit form the argument from SO-focus data
against grammaticized theories of focus-sensitivity—most arguments in the literature
from SO focus have basically the same structure.17
(i) Weak, grammaticized theories of focus-sensitivity require the semantic focus
to be F-marked.
(ii) By focus projection, these theories also predict that the semantic focus should
contain a phonological focus.
(iii) a. SO foci are semantic foci,
b. but SO foci contain no phonological focus-marking.
(iv) Therefore we should prefer a strong, pragmatic account of focus.
This article addresses step (iii)b of the above argument. This step could be attacked
in two ways. First, one might argue, purely theoretically, that observations of a lack
of acoustic prominence cannot possibly justify a claim of a lack of phonological focus
marking, so that even if there were no acoustic prominence, step (iii)b would be unjusti17
See for example Rooth (1996b:206) for a similarly organized argument.
1
312
260
448
fied. Yet we seek to undermine the standard argument even more completely. We seek
to show that it is irrelevant whether the absence of acoustic prominence could, in
principle, evidence a lack of phonological focus. For what we test is whether there is
an acoustic prominence in the first place. If there is, then (iii)b is false. To set this
up, we now consider the nature of acoustic prominences that might potentially signal
phonological focus marking.
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
481
482
485
486
487
488
489
490
491
492
493
494
495
496
497
1485
1486
11487
1
7
LANGUAGE, VOLUME 83, NUMBER 2 (2007)
2.6. THE PHONETICS AND PHONOLOGY OF FOCUS IN ENGLISH. In order to characterize
the phonetics and phonology of focus marking it is necessary to briefly review the
basic elements of English prosody. We adopt here the model developed in Pierrehumbert
1980 and subsequent works (especially Beckman & Pierrehumbert 1986 and Ladd
1996), which also forms the basis for the ToBI system for the transcription of prosody
(Beckman & Elam 1997). In this model, the prosody of an utterance involves three
interrelated components: intonation, phonological phrasing, and stress. An utterance is
divided into phonological phrases, each of which is characterized by a tonal melody
and is usually marked by lengthening of the final syllable or syllables (Shattuck-Hufnagel & Turk 1996).18 Phonological phrasing is related to, but distinct from, syntactic
phrasing (see Shattuck-Hufnagel & Turk 1996 for a review).
The intonational melody is represented as a sequence of high and low tones. These
tones are of two types: pitch accents and boundary tones. A pitch accent is ‘a local
feature of a pitch contour—usually but not invariably a pitch change, and often involving a local maximum or minimum’ (Ladd 1996:45–46) and is associated with a stressed
syllable (cf. Bolinger 1958 and Pierrehumbert 1980). Pitch accents are represented
phonologically as high or low tones (H*, L*) or combinations of tones (e.g. LⳭH*),
and are annotated with an asterisk to differentiate them from boundary tones, which
associate to the edges of phonological phrases. For example, a sentence like 1 can be
produced as a single phrase 8a, or the subject can be phrased separately from the verb
phrase as in 8b. In the latter case, the sentence-medial phrase boundary is marked by
a low boundary tone (Lⳮ).
(8) a. Jan only gave Bill money.
[H*
H* LⳮL%]
b. Jan
only gave Bill money.
[H*Lⳮ] [
H* LⳮL%]
The stress pattern of an utterance specifies the relative prominence of syllables.
Phonetically this prominence is realized by ‘a complex of properties that can be related
to greater force of articulation, including increased intensity and duration, and shallower
spectral tilt’ (Ladd 1996:58). Pitch accents can only be associated with the syllables
that receive the strongest stresses in a phrase (Pierrehumbert 1980, Selkirk 1995).
Focus marking typically involves the alignment of a pitch accent with the primary
stressed syllable of the focused item. Although focus is most frequently marked by an
H* accent (e.g. Hedberg 2003, Hedberg & Sosa 2001), there are many exceptions to
this generalization. For example, in yes-no questions, focus is often marked by an
LⳭH* accent (Hedberg & Sosa 2002). More specifically, focus is usually marked by
a nuclear pitch accent, that is, the last pitch accent in a phonological phrase (see Cohan
2000 and Ladd 1996:225ff.). The nuclear accent is perceived as more prominent than
preceding accents in the same phrase. For example, in 8a there is a pitch accent on
18
The ToBI system distinguishes two levels of phrasing: utterances are divided into INTONATIONAL PHRASES,
which in turn consist of one or more INTERMEDIATE PHRASES. This distinction is not important here, and we
make no claims as to the number of levels prosodic phrasing can take.
1
231
WHEN SEMANTICS MEETS PHONETICS
261
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
Jan, but this first accent is perceived as less prominent than the nuclear accent that
marks the focus, Bill. This observation corresponds to Jackendoff’s (1972:230) generalization that focus is marked by ‘the main stress . . . in the sentence’, since the main
sentence stress receives a nuclear accent. It is possible for a sentence to contain more
than one nuclear accent if it contains multiple phrases, as in 8b, so it is more accurate
to say that focus is marked by the strongest stress in a phonological phrase. It is also
possible for a sentence to contain more than one focus (Jackendoff 1972:258ff.), although the presence of multiple nuclear accents does not necessarily imply the existence
of multiple foci (Ladd 1996:248ff.)—that is, not every nuclear accent marks focus.
At the phonetic level, a pitch accent is primarily a pitch event, as the name suggests.
Pitch is properly speaking an auditory property, but it corresponds fairly directly to
fundamental frequency (F0), which is a measurable property of the speech signal. So
the main acoustic correlates of a pitch accent are to be found in the F0 contour. For
example, an H* pitch accent is typically realized by a local maximum in F0. It is
well established however, that accented syllables also differ from lexically stressed but
unaccented syllables in having greater duration and intensity (a key acoustic correlate
of perceived loudness; Sluijter & van Heuven, 1996a, Turk & White 1999). There is
also evidence that accented vowels tend to have more extreme formant frequencies,
and thus better differentiated vowel quality (de Jong 1995, Harrington et al. 2000).
It is not clear that the greater duration and intensity of accented syllables are correlates
of the pitch accent per se; rather, they may be correlates of stress since pitch-accented
syllables also have the highest levels of stress in a phrase. Treating duration and intensity
as correlates of stress rather than pitch accent is consistent with evidence that longer
duration and higher intensity correlate with stress even in the absence of pitch accent
(Beckman 1986, Sluijter & van Heuven 1996a,b).
Bolinger (1958) has argued that phrasal prominence is marked by pitch accent alone,
so there is no role for a separate notion of phrasal stress (see Ladd (1996:45ff.) for a
detailed discussion of this dispute). In §6, we argue that the distinction between phrasal
stress and pitch accent is helpful in analyzing the differences between the marking of
SO focus and regular focus, but for now all that matters is that pitch accents are generally
accompanied by greater duration and intensity, whether these properties are direct correlates of the pitch accent or of the accompanying stress. Accordingly, duration and
intensity can also play a role in marking focus. Ladd (1996:226ff.) has gone further,
suggesting that certain foci (e.g. in the answer to a question) can be marked without
a pitch accent, so that nonpitch measures may become the primary indicator of focus.
It is thus natural to examine the role of nonpitch measures in second-occurrence focus,
and, as discussed below, we are not the first to do so.
535
536
537
538
539
540
541
542
543
544
545
1546
2.7. THE PHONETICS AND PHONOLOGY OF SO FOCUS. We now turn from the general
issue of how focus is marked to the more specific issue of how SO focus is marked.
It is generally accepted that SO foci do not contain a pitch movement, although we
know of only one systematic experimental study of pitch movement in SO focus—
Bartels 1997, discussed below. But it remains possible that a phonological correlate of
SO focus corresponds acoustically to other measures discussed above, such as duration,
intensity, and vowel quality.
We presented in §2.5 an explicit argument from the phenomenon of SO focus, an
argument that would be undermined if it could be demonstrated that SO focus has
phonological correlates. Rooth (1996b) presented suggestive evidence indicating just
that, and his observations provided the impetus for the more systematic experimental
studies that we report here.
498
499
500
501
502
503
504
1
7
1
312
262
547
562
563
564
Rooth recorded himself uttering minimal pairs of dialogues differing with regard to
which item was the SO focus. He found that the pitch track was flat in SO-focus position,
but that he could auditorily detect the marking of SO focus in his own productions.
Furthermore, he noted that this marking of SO focus is visible in the small sample of his
productions for which he presents waveforms and spectrograms. In these productions,
although no major pitch movement is visible on SO-focus expressions, these expressions
have greater duration and absolute intensity than their nonfocused counterparts.
In related work, Bartels (1997) studied nonpitch correlates of prominence (relative
syllable lengthening and amplitude) on second-occurrence expressions in a multisubject
production study. She found that SO foci differed from ordinary foci with regard to
not only pitch, but also amplitude and duration. While Bartels demonstrated that a
systematic experimental approach to SO-focus phenomena was possible, her conclusions concerned the different realization of SO focus from regular focus. She did not
add controls in which the test words were not in focus at all, that is, in which the
test words were neither foci nor second-occurrence foci (Bartels 1997:24). Thus her
experiments do not determine whether SO focus is marked prosodically or what its
correlates might be, but merely establish that if SO focus is marked, then it is marked
differently from ordinary focus on several acoustic dimensions.
565
566
567
568
569
570
571
572
573
3. A PRODUCTION EXPERIMENT. We now report on a production experiment designed
to test the hypothesis that SO focus is acoustically distinguished from nonfocus. This
hypothesis competes with the null hypothesis implicitly assumed in the argument from
SO focus (see §2.5) that SO focus is not prosodically marked. Earlier, we alluded to
a possible explanation for the strong intuition of some researchers that SO focus is NOT
focus-marked: SO focus is not realized acoustically in the same way as ordinary focus.
A secondary goal of the experiment is to confirm the hypothesis that SO focus is marked
differently from first-occurrence focus by singling out respects in which SO foci differ
acoustically both from nonfocused expressions and from ordinary foci.
574
575
3.1. DESCRIPTION OF THE EXPERIMENT. We ran a production experiment in which
naive subjects read prepepared written materials.
576
577
SUBJECTS. Twenty US-born native speakers of English (ten female, ten male) were
recruited, none with any training in linguistics.
578
579
580
581
582
STIMULI. An example of a minimal pair of discourses used as stimuli is given in 9
and 10. These three-sentence discourses are designed to probe SO-focus effects in the
scope of a focus-sensitive operator in the third sentence. In this case, the relevant
operator is only, while in other test pairs the operator was always. Note that another
operator, even, is also present in 9c and 10c, typically causing the initial subject NP
to bear nuclear accent, and leaving only and its associate (Sid or court) in a postnuclear
position.
(9) a. Both Sid and his accomplices should have been named in this morning’s
court session.
b. But the defendant only named Sid in court today.
c. Even the state prosecutor only named Sid in court today.
(10) a. Defense and Prosecution had agreed to implicate Sid both in court and
on television.
b. Still, the defense attorney only named Sid in court today.
c. Even the state prosecutor only named Sid in court today.
548
549
550
551
552
553
554
555
556
557
558
559
560
561
583
584
585
587
590
591
593
594
596
598
601
602
604
1
607
605
1
7
LANGUAGE, VOLUME 83, NUMBER 2 (2007)
1
231
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
1488
1489
1490
1491
1492
1493
1494
11495
1
7
WHEN SEMANTICS MEETS PHONETICS
263
In examples 9 and 10, the relevant potential foci are Sid and court. For all of the
stimuli we used, the material following the second focus-sensitive operator in the third
sentence (e.g. 9c and 10c) does not differ between the two members of the discourse
pair. Thus, the phonological context for the segment of text containing the two potential
foci should not differ between the two elements of the minimal pair. We can therefore
attribute acoustic differences between the potential foci in the two pairs to SO-focus
effects.19
Crucially, the focus in each of the (b) sentences in 9 and 10 was different, so the
textually identical VPs in the (c) sentences should differ only in so far as they contain
different expressions that are the SO focus. In 9c, Sid is the second occurrence of a
focus, and court is nonfocal, whereas in 10c, Sid is nonfocal, and court is the second
occurrence of a focus. Thus the hypotheses above can be operationalized in terms of
the acoustic differences between the realizations of Sid in each of the two final sentences, and the acoustic differences between the realizations of court in each of the
two final sentences. If an expression is the second occurrence of a focus in the final
sentence of a discourse, we say that the expression is in the SO-FOCUS CONDITION, and
otherwise we say it is in the NON FOCAL CONDITION.
Note that our stimuli, like those of Rooth (1996b), are designed so that each minimal
pair of discourses gives us two distinct probes of the hypotheses, since the conditions
are reversed for the two potential SO foci Sid and court in the two discourses. Finally,
note that we chose our stimuli to be maximally conservative with regard to the null
hypothesis. In all stimuli, the SO focus follows another first-mentioned focus, which
typically receives a nuclear accent (see §2.2). In other words, the SO-focus expressions
investigated here not only are repeated but also appear in the postnuclear domain,
thereby making accenting of the SO-focus expression very unlikely. We chose stimuli
of this type to bias heavily against our hypothesis. We are interested in whether foci
are acoustically marked even if they occur in prosodic environments that prevent marking by pitch accents.
We used a total of fourteen discourse stimuli, made up of seven minimal pairs like
those above. In each case, the repeated focus-sensitive operator operator was only or
always, while the focus-sensitive even was used to induce the nuclear accent in the
subject of the final sentence of the discourse. In all cases the potential second-occurrence
foci in the final sentence are not sentence-final, this being contrived by the use of an
additional adverb, for example, today in 9c and 10c. The reason for adding an adverb
is to prevent features of a potential focus expression from being combined with, and
perhaps masked by, pitch movements marking the end of an intonational phrase.
PRESENTATION AND PROCEDURE. Following standard procedure, the twenty-eight experimental stimuli were arranged in a pseudo-random order and intermingled with
sixteen unrelated filler stimuli, also discourses.20 To control for recency effects that
might occur when a subject encounters both members of a minimal pair of discourses,
19
A referee points out that in all of our stimuli, the choice of SO-focus placement is between two NPs,
as opposed to an NP and some other constituent. The referee observes that conceivably the presence of the
two NPs might induce the speaker to use a contrastive intonation. It is important to realize that such contrastive
marking, in and of itself, would not be expected to yield a net effect on our measures, since we consider
relative differences between the two NPs. Because of this design feature, however, it might be argued that
our conclusions do not strictly cover SO focus marking in general, but only SO focus marking in the presence
of potential contrast.
20
Fillers were drawn from a separate experiment, reported in Wolters & Beaver 2001.
1
312
264
648
we ensured that paired discourses were always separated by at least four other discourses, either other experimental stimuli or fillers. In this way, four different stimuli
lists were constructed. Each participant in the study received one of these lists, and
list was included as a factor in the statistical analysis. The twenty-eight experimental
discourses consisted of seven minimal pairs repeated twice (for an example of a minimal
pair, see 9 and 10 above). After recording all of the speakers, the word boundaries
were marked on the two probes taken to be potential foci in the target sentence (i.e.
the third sentence of each discourse), for example, Sid and court in 9c and 10c above.
This procedure yielded 28 (discourses) ⳯ 20 (subjects) ⳯ 2 (probes per target sentence)
⳱ 1,120 probes.
The word boundaries were labeled by a subset of the authors (DB, BC, and MW),
based primarily on examination of spectrograms. For each word, boundaries were
marked at acoustic landmarks (such as stop closures) near the word onset and offset.
The landmarks were selected (i) to be consistently identifiable across utterances of a
word, and (ii) to include the vowel of the test word, since this is the expected locus
of pitch and duration effects.
After all of the relevant word boundaries had been hand-labeled, seven phonetic
measures were taken for each target word. The measures were: duration of the target
word (in seconds), root-mean-square intensity of the word (in decibels (dB); henceforth
rms intensity), F0 range, and maximum, minimum, and mean F0 (all in Hz).21 A pseudoenergy measure was created by multiplying the standardized rms intensity value by the
duration of the target. These measures were selected based on previous research on the
acoustic correlates of focus, reviewed in §2.6.
For each phonetic measure taken on the target items we conducted both a by-subjects
(henceforth F1) and a by-items (henceforth F2) analysis of variance (ANOVA). Each
ANOVA corresponded to a two-way crossed 2 ⳯ 2 repeated measures design of SO
Focus (whether the preceding context identified the target phrase as SO focus or not)
and Word Position (whether the target phrase was the direct or the indirect object).
Following Clark 1973 and Raaijmakers et al. 1999, a minimum F was calculated from
the F1 and F2 analyses. This maximally conservative approach was taken because the
results reported below provide evidence against a long-standing opinion that SO focus
is not phonetically marked. Although we report F1, F2, and minF values below, our
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
11514
1
7
21
LANGUAGE, VOLUME 83, NUMBER 2 (2007)
Intensity and F0 information were extracted using standard procedures in Boersma and Weenink’s Praat
program. All pitch measures and the intensity measure were normalized using standard deviation from their
mean in that single utterance and therefore are analogous to z-scores. This was done to prevent confounding
effects from differences in the speakers’ characteristic pitch ranges. Distributional normality of test measures
can be affected by such operations and is addressed in Appendix B.2.
Note also that all F0 measures are across a section strictly including the vowel. Thus there is potential
interference from obstruent perturbations. This could have been avoided by analyzing only a subpart of the
vowel. However, (i) we performed an analysis of foci (involving the same focused words) in the second
sentences of the discourses using the same methodology and found that pitch accents were easily detectable,
so there is reason to think that our approach is sensitive to F0 effects; (ii) obstruent effects, if present, would
have been revealed by our item analysis, but we observed no such effects; and (iii) we hand-corrected pitch
measures for a subset of the data, and found no change in results. Furthermore, analyzing only a subpart of
the vowel might have lead to other problems, for example, the risk of missing late pitch accents, which do
occur in our data.
A related issue arises with respect to variation of intrinsic F0 in the stimuli according to vowel height.
We did not control explicitly for vowel height, although the target words in the stimuli have a range of
vowel heights. We can make a similar comment here as with respect to segmental effects of obstruents: if
intrinsic F0 effects due to vowel height had systematically affected SO-focus expressions differently from
nonfocal expressions, then that would have shown up in our analysis, but it did not.
1
231
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
WHEN SEMANTICS MEETS PHONETICS
265
conclusions are based solely on the conservative minF analysis. If not mentioned otherwise, F-values are reported as significant if p ⬍ 0.05, and as marginal if p ⬍ 0.1. Word
position was included in the design in order to test for interactions between word
position and SO focus.
A blind review process was performed prior to the statistical analysis. Overall, 0.7%
of the data was excluded from the analysis of duration, intensity, and energy and 2.8%
of the data was excluded from the pitch analysis (for a detailed description of the
exclusion criteria, see Appendix B.1).22 The higher data loss for the pitch analysis is
due to the well-known problems of pitch-tracking algorithms.23
3.2. RESULTS. The analyses revealed a significant main effect of SO focus on duration
(F1(1,19) ⳱ 6.9, F2(1,6) ⳱ 15.3, minF(1,24) ⳱ 4.8) and energy (F1(1,19) ⳱ 8.2,
F2(1,6) ⳱ 17, minF(1,24) ⳱ 5.5). SO-focus phrases were on average 6 ms longer than
nonfocused phrases and were realized with 3.4% more energy. There was also a marginal effect of SO focus on standardized rms (root-mean-square) intensity (F1(1,19)
⳱ 7.1, F2(1,6) ⳱ 8, minF(1,19) ⳱ 3.8), the standardized F0 range (F1(1,19) ⳱ 8.8,
F2(1,6) ⳱ 8, minF(1,17) ⳱ 4.2), and the standardized minimum F0 (F1(1,19) ⳱ 5.2,
F2(1,6) ⳱ 9.1, minF(1,23) ⳱ 3.3). SO-focus phrases were realized with 0.31 dB higher
intensity, a 10.5% higher relative F0 range (38.3 vs. 34.9 Hz), and a 6.9% lower
relative minimum pitch (121.4 vs. 124.5 Hz). No main effect of SO focus was found
for maximum F0 and mean F0. The means for all measures in both conditions of SO
focus and the mean differences between the two conditions of SO focus are given in
Table 1. Significant results are marked with an asterisk, marginal results with a plus
sign.
12
SO FOCUS
MEAN DIFFERENCE
(␮nf )
(␮sof )
⌬(␮sof ⳮ ␮nf )
0.267
0.273
*0.006
Ⳮ
ⳮ27.28
ⳮ26.97
0.31
0.0106
0.0117
*0.0011
162.2
163.0
0.8
143.1
142.5
ⳮ0.6
Ⳮ
ⳮ3.4
127.5
124.1
Ⳮ
34.6
38.7
4.1
differences for nonfocal and SO-focus expressions.
NON FOCAL
21
14
23
27
33
29
35
39
41
45
47
51
57
53
59
63
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
11532
1
7
N
Duration (s)
521
Std. rms intensity (dB)
521
Rel. energy
521
Max F0 (Hz)
483
Mean F0 (Hz)
440
Minimum F0 (Hz)
473
Range F0 (Hz)
457
TABLE 1. Summary of means and mean
DEPENDENT VARIABLE
22
Most exclusions were due to problems tracking F0 in creaky voice. This could be problematic in an
experimental study of prosody, since creaky voice is presumably correlated with prosodic factors. For example, creaky voice is less likely to obscure the F0 of a nuclear accent than to obscure the F0 of material in
the postnuclear tail. Since F0-tracking problems affected less than 5% of the data, we did not attempt to
build creaky voice in as a factor in our statistical analysis.
23
We also checked for possible effects of the prescriptivist dogma that expressions such as exclusives
should not be separated from what the modify, and thus in ‘good English’ are not focus-sensitive at all. We
know of no basis for such a claim, since focus-sensitivity is not only widespread, but also not a historically
recent development. For detailed historical consideration of the focus-sensitivity of English exclusives, see
Nevalainen 1991. It is conceivable that prescriptive rules may have affected how subjects in our experiments
reacted to the stimuli we used, which involve exclusives that are separated from what they modify. We
cannot see how this would produce any consistent effect, however. Furthermore, the use of three-sentence
discourses in which only the third sentence provides our main data set means that we have a baseline measure
of how subjects produce sentences with focus-sensitive expressions. If prescriptive rules had any effect, we
would expect that effect to arise with regard to both the second and the third sentences of our discourses.
On the basis of informal sampling, however, the huge majority of the second sentences in the stimuli were
produced with the type of intonation that would be expected in the theories of focus we consider here. It
seems that prescriptive rules did not interfere with the subjects’ ability to produce foci naturalistically.
1
312
266
703
Unsurprisingly, word position had a main effect on rms intensity, energy, and all standardized F0 measures. All of these main effects were in the F1 analyses and only the
two main effects of standardized mean F0 (F1(1,19) ⳱ 11, F2(1,6) ⳱ 5.7, minF(1,13)
⳱ 3.8) and standardized minimum F0 (F1(1,19) ⳱ 17.1, F2(1,6) ⳱ 11.4, minF(1,15)
⳱ 6.8) were significant in the minF analyses.
For reasons addressed in §3.1 (under the heading ‘Presentation and Procedure’), we
also tested for interaction effects between SO focus and word position. We found a
marginal interaction effect of SO focus and word position on the standardized mean
F0 in both the F1 and the F2 analyses. Whereas the relative mean F0 was higher for
SO focus than for nonfocused indirect objects, this effect was reversed for the direct
object. The effect, however, turned out to be neither significant nor marginal for the
minF analysis (F1(1,19) ⳱ 4.4, F2(1,6) ⳱ 5.3, minF(1,20) ⳱ 2.4). We also found a
marginal interaction effect in the F1 analysis for standardized rms intensity (F1(1,19)
⳱ 3.5, F2(1,6) ⳱ n.s., minF ⳱ n.s.) and in the F2 analysis for the standardized
minimum F0 (F1(1,19) ⳱ n.s., F2(1,6) ⳱ 6.7, minF ⳱ n.s.). Whereas the standardized
minimum F0 was the same for SO-focused and for nonfocused indirect objects, it was
lower for SO-focused than for nonfocused direct objects. The inverse was observed
for intensity. Relative rms intensity on direct objects was the same in both SO-focus
conditions, but was higher for SO-focused than nonfocused indirect objects. This means
that the main effect of SO focus on intensity is only due to intensity differences on the
indirect object. The main effect on minimal F0 is only due to F0 differences on the
direct object. The choice of which list (out of the four available with different stimulus
orders) was read by the subject was not significant in any of the analyses.
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
1750
1
7
LANGUAGE, VOLUME 83, NUMBER 2 (2007)
3.3. DISCUSSION. As expected on the basis of prior literature, we found no evidence
for systematic pitch marking of SO focus. Instead, we found a small but significant
lengthening of the SO focus in comparison with the same expression in a sentential
context that differs only in lacking focus. Across all trials, this lengthening averaged
6 ms. There is also a statistically significant increase in energy, as well as marginally
significant increases in intensity and the FO-range in the SO-focus condition. Finally,
there was a marginally significant drop in minimum FO in the SO-focus condition.
The nature of interaction effects between word position and SO focus for minimum
FO and intensity suggests a more complex relation between these measures and SOfocus marking.
The statistical significance of the duration and energy results provides strong evidence
that SO foci are focus-marked at some level of linguistic representation, since we have
been careful to ensure that there is no difference between the SO-focus and nonfocused
condition except for the position of focus. This corroborates what Rooth observed in
his own productions and is sufficient to undermine any argument for a strong (pragmatic) theory of focus that relies crucially on SO focus not being formally marked. As
far as SO-focus phenomena are concerned, weak (syntactic/semantic) theories of focus
are quite defensible.
Accented words are known to differ from unaccented words in duration and intensity,
as discussed in §2.6. The results we have presented show that SO focus is marked by
the same cues. By contrast, fundamental frequency plays a central role in the realization
of pitch accent, but appears to play only a minor role in the marking of SO focus: the
only identifiable effect on FO measures is a marginally significant lowering of minimum
FO in SO-focus expressions in direct object position. We have no convincing explanation for this limited effect. It does not correspond to any local pitch movement in the
1
231
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
11545
1
7
WHEN SEMANTICS MEETS PHONETICS
267
examples that we have examined, and in the few cases where SO focus is marked by
a clear pitch accent, the accent is high (H*), which would lead us to expect an increase
in FO measures (at least the maximum and minimum FO measures) on the SO-focus
expression (see §5.1 for further discussion).24 In any case, the perception experiment
reported in §4 provides evidence that this FO difference does not play a significant
role in the perception of the marking of SO focus. So we conclude that greater duration
and energy are the primary correlates of SO focus.
While the existence of a difference between SO foci and their nonfocal counterparts
in production indicates that speakers are making a distinction between the two, the
mean differences between SO foci and their nonfocal counterparts on the measured
parameters are small. This raises the question of whether the difference is actually
perceptible. We address this question in the experiment reported in the next section.
4. A PERCEPTION STUDY. We conducted a forced-choice perception experiment to
test the hypothesis that hearers can distinguish between the prosodic realizations of SO
foci and nonfoci.
4.1. DESCRIPTION OF THE EXPERIMENT.
SUBJECTS. Fourteen native speakers of English (ten female, four male; age range:
23–28) were recruited. Ten were linguistically naive (eight female, two male; age
range: 19–26) and lacked prior experience in linguistic perception studies, while the
remaining four subjects (two female, two male; age range: 24–28) had prior training
in linguistics.
STIMULI. Forty minimal discourse pairs with the focus-sensitive operator only were
quasi-randomly selected from sentences produced in the experiment described in the
previous section. Three (out of the four available) minimal discourse pairs with the
focus-sensitive operator only were used (the pairs stimuli consisted of sixteen tokens
each of the pairs in Appendix A, 2 and 3, and eight tokens of the pair in 4).
Discourses that contained stuttering, unusually long pauses, or other signs of ‘reading
effects’ (recall that the production experiment elicited the stimuli in a reading task)
were excluded from the experiment in a blind review process prior to the random
selection of the discourse pairs. Of the selected discourses, only the final sentence was
used. Both parts of a pair always were taken from the same three-sentence discourse
pair and were uttered by the same speaker. The two elements of each pair differed only
in their (second-occurrence) focus assignment.
The list of stimulus pairs was balanced so that two pairs were derived from each of
the twenty speakers in the production experiment. For each of the three minimal dis24
Katy Carlson (p.c.) suggests that some of our recordings should be transcribed with a low pitch accent
(L*) on the SO focus. If correct, this might explain the observed F0 effect. We did not identify any L*
accents in our own transcriptions of a subset of the data, but it can be difficult to distinguish an L* accent
from absence of accent in a context where F0 is expected to be low in any case. In support of the unaccented
interpretation, we argue below (§5.2) that analyzing the usual realization of SO focus as unaccented enables
us to relate the occurrence of this otherwise anomalous pattern of focus marking to the phenomenon of
postnuclear deaccenting. By contrast, current generalizations about the use of L* accents (e.g. Pierrehumbert &
Hirschberg 1990) do not provide any basis for expecting an L* accent in SO-focus environments, and in
fact the few cases in our data where SO focus unambiguously receives a pitch accent involve high accents
(H*). Note also that Hedberg and Sosa (2001) provide corpus-based distributional evidence that, for ordinary
focus marking, H* accents are much more common (61.9%) than L* accents (19.0%). Nothing we currently
know about H* and L* accents would lead us to expect that the relative frequencies of H*s and L*s are
reversed in the case of SO-focus marking.
1
312
268
786
course pairs used, there were an equal number of instances in which the first NP was
a SO focus and in which the second NP was a SO focus.25
To avoid systematic effects from our presentation of the stimuli, a number of order
constraints were observed when the list of stimuli was constructed: (i) the same internal
order of a pair (the first sentence has a SO-focus-marked first NP or a SO-focus-marked
second NP) was not allowed to occur more than three times in a row; (ii) the same
pair was not allowed to occur twice in a row. Furthermore, two versions of the experiment with inverse order were produced, and half of the subjects were given each version.
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
LANGUAGE, VOLUME 83, NUMBER 2 (2007)
PRESENTATION AND PROCEDURE. The acoustic stimuli were incorporated into an HTML
page. Subjects first read through the instructions and were allowed to ask the experimenter questions before they started the experiment. Subjects were asked to judge ‘in which
of the two sentences (A or B) the speaker wished to make the second word (which is
given in bold face; see Figure 1) more prominent than the first’. Subjects were not
allowed to make any changes once they reached a decision for a pair.
Each pair of acoustic stimuli was accompanied by two words that were displayed
in the same row, as exemplified in Fig. 1.
6
7
8
9
FIGURE 1. Excerpt of HTML page displayed to subjects during the experiment.
802
803
804
805
806
807
808
809
Ten pairs of stimuli were displayed at a time. After every tenth pair of sentences,
subjects were asked to take a break of thirty seconds to two minutes, which they could
time themselves, before they continued on to the next page. Subjects were asked to
proceed from pair to pair and always to listen to the first stimulus in a pair first. After
that, they were allowed to listen to each of the stimuli at will. Subjects required fifteen
to twenty-five minutes for the experiment and approximately two minutes to read the
instructions. To test whether subjects understood the instructions, after the experiment
all subjects were asked to repeat in their own words what they had been asked to do.
810
811
812
4.2. RESULTS. All subjects performed above chance (mean performance ⳱ 63%
correct answers; range ⳱ 52.5–77.5%). The one-sample t-test against an expected
mean of twenty correct answers out of forty reveals that subjects on average perform
significantly above chance (t(13) ⳱ 7.7, p ⬍ 0.001) and that subjects performed ‘alike’
(i.e. the effect was not just driven by a few subjects; n2 (13) ⳱ 9.6, p ⬎ 0.7). The average
effect size is small to medium ( ␻ ⳱ 0.28). All subjects showed clear understanding of
the task in the postexperiment interview.
None of the between-subject factors (i.e. gender, which of the two stimulus lists was
used, or linguistic education) had a significant effect. Neither did ordering (i.e. when
813
814
815
816
817
818
1546
1547
11548
1
7
25
Recall also that subjects in the production study read every stimulus twice (see §3). So the stimuli used
in the perception experiment controlled for repetition: if one stimulus in a pair was a token of a repeated
production of the same discourse, then the other stimulus in that pair was also a repeat.
1
231
819
820
821
822
823
824
825
826
827
828
829
830
WHEN SEMANTICS MEETS PHONETICS
269
in the list a stimulus was presented) have a significant effect, and this indicates that
the task was not too tiresome for our subjects. Crucially for pooling the data from all
subjects, the level of linguistics education did not matter (n2(1) ⳱ 1.2, p ⬎ 0.25).
Furthermore, all effects that hold for all subjects together also hold for linguistically
naive subjects alone. There was also an item effect (n2(39) ⳱ 163.9, p ⬍ 0.001). That
is, subjects did not perform equally well on all sentence pairs. Interestingly, there were
items that nobody answered as expected and items that everyone answered as expected
by our hypothesis that SO foci are perceptible. A planned comparison of the discourse
pairs used in the experiment revealed a significant effect (n2(2) ⳱ 10.9, p ⬍ 0.01):
one of the three minimal discourse pairs on average lead to better performance than
the other two pairs. It is important for the current purpose that subjects performed
significantly above chance for all three discourse pairs.26
862
863
4.3. DISCUSSION. The results confirm that SO foci are perceptually distinguishable
from nonfoci. Even though the phonetic correlates of SO focus observed in the production experiment are small, subjects were able to identify SO-focus-marked expressions
correctly in 63% of all cases in the perception experiment. This shows that subjects
are able to identify SO-focus marking without contextual clues, but it also raises the
question of why people did not perform even better (or in other words, why the effect
size is relatively small). For one thing, overall performance decreased due to four items
that were almost always judged contrary to our hypothesis of acoustically prominent
SO foci (judged correctly by only three or fewer subjects of the fourteen total) and
four other items that performed below chance (correctly judged by four to six subjects).
The existence of items judged below chance is neither what is expected according to
our hypothesis nor what would be expected in the absence of prosodic cues to SO
focus. Since subjects should perform at chance in the absence of phonetic cues, these
answers should have been judged correctly by 50% of all subjects.
Prior to the perception experiment, we had already examined the prosody on the
second sentences in the read discourses, which contained one ordinary focus expression
and no SO focus expression. While the majority of the stimuli were well-formed in
their second sentence, as judged by the authors in an informal listening experiment,
the rate of intonationally ill-formed sentences was high enough to explain why some
stimuli in the perception study were consistently judged counter to expectations. In
several stimuli, the second sentence contained a strong pitch accent on what was intended to be the unfocused expression but not on the focused expression, indicating
that the subject had problems reading the stimuli (either because the stimulus was
textually unnatural or because the subject was tired or not paying sufficient attention).
The wrong placement of the pitch accent sometimes seemed to carry over to the third
sentence containing the SO-focus expression. The decision not to exclude those supposedly ill-formed stimuli a priori was made because this may have seriously confounded
the study. It is clear, however, and, indeed, unsurprising, that the existence of occasional
unnatural productions in the original reading task, perhaps due to the artificial experimental scenario, is one of the factors limiting performance on the perception task.
Finally, it is significant that the perception task that subjects were asked to perform
referred to intended prominence rather than directly to SO-focus marking. The main
purpose of the perception experiment was to establish whether the small but reliable
1549
1
1550
26
Stimuli from pair 2 in Appendix A elicited 60.7% correct answers and from pair 3 59.4% correct, while
pair 4 yielded 76.8% correct answers.
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
1
7
1
312
270
864
SO-focus marking we had previously observed is above the threshold of what can be
humanly perceived. In that sense, participants were used as (very intelligent) minimaldifference detectors. That they performed significantly above chance means that the
difference between the phonetic realization of SO focus and that of unfocused repeated
expressions is humanly detectable, not that it is reliably detectable in natural situations.
In addition, it is impossible to know from this experiment whether listeners actually
use phonetic marking to infer SO focus: it is logically compatible with our findings
that hearers disregard the phonetic differences we have observed and rely entirely on
context to identify second-occurrence foci.
With regard to the latter issue, it should be borne in mind that the point of this article
is not to falsify pragmatic theories of focus or to show that pragmatic and discourse
clues are irrelevant to the determination of focus assignments. Rather, our goal is to
provide evidence that, contrary to claims in the literature, phonetic cues are available
to infer the focus assignments of a sentence even in the case of SO focus. Our production
results demonstrated that cues are reliably present, and the perception results show that
such cues are in principle accessible to hearers. But since hearers did not identify
SO foci as prominent with anything close to 100% accuracy under our experimental
conditions, it must remain open to what extent SO-focus marking is used in online
speech processing. Note here that relatively little is known about the processing of
even first-mention focus marking, but see, for example, Watson et al. 2006 and Welby
2003.
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
11551
1
7
LANGUAGE, VOLUME 83, NUMBER 2 (2007)
5. GENERAL DISCUSSION. In this section we combine data from the production and
perception experiments to study in more detail the important differences between SO
focus and ordinary focus marking. At that point, we are in a position to consider why
SO focus is realized differently from regular focus, and in what other environments
similar effects should be expected.
Our data indicates that SO focus is made prominent by greater duration and energy
(and perhaps other prosodic correlates not considered in this study) but is usually not
pitch-accented. Below, we suggest that SO-focus marking is an example of phrasal
prominence without pitch accent. This realization can be understood as resulting from
the fact that items in SO focus occur after the nuclear accent of their phrase, the
environment of postnuclear deaccenting. We show that this interpretation has implications for the general theory of phrasal prominence (stress and accent) and the marking
of focus. We then relate this analysis to prior discussion in the semantics literature of
example types closely related to the SO-focus paradigm (§5.3).
5.1. WHAT IS NONSTANDARD ABOUT SO FOCUS-MARKING? We conducted a comparison
between SO-focus marking and the marking of ordinary, nonrepeated focus, based on
two questions.27 First, does standard SO-focus marking lack a pitch accent, and, second,
can SO focus optionally be marked using (ordinary) focus marking, complete with
pitch accent?
As discussed in §2, focus is usually marked by a pitch accent, perhaps specifically
a nuclear pitch accent, and the primary correlates of pitch accent are to be found in
the fundamental frequency contour. So a striking feature of SO-focus marking is the
minor role played by fundamental frequency. The production study revealed no significant main effect for F0 differences between words in SO focus and unfocused words.
Auditorily based transcription of a subset of the data indicates that SO-focus expressions
27
See Jaeger 2004:§4 for further details.
1
231
WHEN SEMANTICS MEETS PHONETICS
271
941
942
943
944
945
946
947
948
are occasionally marked with high pitch accents (H*), but in most cases there is no
perceptual indication of a pitch-accent on the SO-focus expression. For cases without
a pitch accent on the focus expression, two representative pitch tracks along with their
transcriptions are shown in Figures 2 and 3. The focus expression associated with even
is marked by a clear F0 peak, indicating that the focus of even is marked by a high
pitch accent. Then the fundamental frequency falls to a low level and remains relatively
low, or slightly declining, to the end of the utterance. That is, there is no evidence for
a pitch accent on any words after the focus expression associated with even.28
Figure 4 is a representative example of a SO-focus expression marked by an H*
pitch accent (though a comparatively subtle one).29
So the results of the production study show not only that second-occurrence focus
is consistently marked by greater duration and energy, but also that it is occasionally
marked by F0. To put it in the phonological terms we used to set up this issue: SO
foci can optionally be marked with a pitch accent. Given the existence of optional pitch
accents, it is logically possible that the perception subjects’ ability to distinguish SO
focus from lack of focus is based on a few stimuli in which second-occurrence focus
is marked by a pitch accent. To show that this is not the case, we now turn to a brief
analysis of the acoustic properties of the stimuli used in the perception experiment.30
A logistic regression model was used to predict the subjects’ responses in the perception task based on measurements of the acoustic properties of the stimuli. Note that
this model predicted the subjects’ actual responses rather than the ‘correct’ answers,
in order to try to identify the stimulus parameters that listeners attended to in making
their judgments. The best predictors of subjects’ judgments were measures based on
the duration and intensity of the target words, and the best single predictor was the
energy measure derived by multiplying duration by intensity. Adding fundamentalfrequency information did not significantly improve the fit of a model based on the
energy factor. This indicates that subjects relied primarily on duration and intensity in
identifying the more prominent word, and fundamental frequency did not play an important role.
In addition to the logistic regression analysis, we examined the ‘best stimuli’ from
the perception test, that is, those stimuli for which all subjects gave the same response.
We would expect these stimuli to exhibit a particularly clear distinction between SO
focus and unfocused items. Careful inspection of the pitch measurements in the target
words and the overall pitch contours revealed no apparent way in which F0 could have
been used to distinguish the prominence of the words.
Based on both the logistic regression analysis and the sample of best stimuli, we
conclude that the primary correlates of SO focus are duration and energy in the sense
that these are the measures that most reliably distinguish SO focus from lack of focus
in production, and that they are the properties that listeners attended to in the perceptual
1552
1553
1554
1555
1556
1557
1558
1559
1560
11561
28
The small perturbations mostly at the syllable onsets and syllable codas of some words are consonant
effects and are not intonationally significant. For an overview of such effects, see for example Silverman
1987 or Beckman & Elam 1997. Note also that the pitch tracks in Figs. 3 and 4 contain instances where
the pitch-tracking algorithm erroneously halved the fundamental frequency for a short period (cf. Ladefoged
2003:83–84). The actual F0 is more or less level.
29
In the utterance pictured in Fig. 4, doctor, the focus of even, is highly prominent auditorily. However,
it is notable that (aside from onset effects) there is no pitch movement on doctor. It is conceivable that this
is related to the fact that a pitch movement does occur on the later SO focus Pete, but we have not studied
this possibility further.
30
More details are given in Jaeger 2004:§4.4–4.5.
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
1
7
FIGURE 2. First sample with a SO focus lacking any pitch accent.
312
1
1
7
272
13
14
15
12
1
11
10
LANGUAGE, VOLUME 83, NUMBER 2 (2007)
FIGURE 3. Second sample with a SO focus lacking any pitch accent.
231
1
1
7
20
21
22
19
1
18
17
WHEN SEMANTICS MEETS PHONETICS
273
FIGURE 4. Sample with a pitch-accented SO focus.
312
1
1
7
274
27
28
29
26
1
25
24
LANGUAGE, VOLUME 83, NUMBER 2 (2007)
1
231
WHEN SEMANTICS MEETS PHONETICS
275
949
950
951
952
953
954
955
956
957
task. Fundamental frequency is only optionally used in marking SO focus and is not
required for perception of the prominence of SO focus. Our conclusions are supported
by new evidence from a study on SO foci in German. Féry and Ishihara (2005) found
that in German, too, postnuclear SO foci are marked by longer durations but not by
pitch accents. Of course, if a speaker chose to mark SO focus with a pitch accent, it
is quite possible that the associated F0-related cues would be more significant than
duration and energy in giving rise to a percept of prominence, but this is not the typical
realization of SO focus, nor is a pitch accent necessary for listeners to detect the
prominence associated with SO focus.
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
5.2. WHY ARE SO FOCI MARKED WITHOUT A PITCH ACCENT? Phonologically, we interpret the confirmation that pitch suppression occurs in terms of a distinction between
phrasal stress and pitch accent: focus is usually marked by phrasal stress and a nuclear
pitch accent, but SO focus is marked by phrasal stress only. As discussed in §2.6,
theories that distinguish stress and pitch accent generally identify greater duration and
intensity as primary correlates of stress, while pitch accents are realized in terms of
the fundamental-frequency contour. So the greater duration and energy of SO focus
are attributed to phrasal stress, and the absence of F0 differences between SO-focus
and nonfocused material is attributed to the absence of pitch accents in both conditions
(with a few exceptions in the case of SO focus, as noted above).
Positing that SO focus is marked by stress but not accent helps explain why SO
focus is realized differently from regular focus, by relating SO focus to the wellestablished phenomenon of postnuclear deaccenting. As mentioned in §2.6, a nuclear
accent is the last pitch accent in a phrase, so placing a nuclear accent early in a phrase,
as in 11, implies that all following words in that phrase must be unaccented. This is
referred to as DEACCENTING, because in an example like 11 the word Sandy would be
accented in a more neutral (or broad-focus) reading of the sentence, but is not if the
preceding book receives a nuclear accent (e.g. because it is narrowly focused).
(11) Pat gave a [book]F to Sandy.
In all of the examples used in our experimental study, the second-occurrence focus
is preceded by an expression that we expect to be realized with a nuclear pitch accent,
marking the focus associated with even. So, as we have indicated, the absence of a
pitch accent on the SO focus could be regarded as a case of postnuclear deaccenting
(Bartels 1997:12, Rooth 1992). That is, the SO focus could not be accented without
supplanting the accent of the focused subject as nuclear accent, or initiating its own
phrase, in which case the sentence would contain two phrases, each with a nuclear
accent.31 The predominance of pitch-accentless realizations should be related to the
fact that the material following the subject (e.g. only named Sid in court today in Fig.
3) is GIVEN information (in the sense of Prince 1981) and structurally identical to a
portion of the preceding sentence (i.e. the same words bear the same grammatical
functions and are in the same syntactic positions). These two factors favor unaccented
realizations (Ladd 1996:66, Terken & Hirschberg 1994), so speakers may be disinclined
to place this material in its own phrase where it must receive a nuclear accent.32
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
1562
1563
1564
1565
1566
1567
1568
1
1
7
31
See Jaeger 2004:§5.3 for discussion.
Note, however, that given material can in principle be accented (Hedberg 2003). It is really the postnuclear occurrence of given material that leads to deaccenting and prosodic subordination (phrasing of the
material following the nuclear accent into the same phrase as the nuclear accent (Wagner 2005)). Hence,
the account suggested here predicts that pitch accent should more frequently be a valid option for prenuclear
SO foci (i.e. repeated foci in a prenuclear position) than for postnuclear SO foci. A recent study on German
(Féry & Ishihara 2005) suggests that this prediction is borne out.
32
1
312
276
993
The picture we have arrived at is one according to which the focus of a focussensitive operator should receive the strongest phrasal stress in the scope of the operator
(cf. Jackendoff 1972), which is preferentially achieved by assignment of a nuclear
accent to the focus. In the canonical postnuclear SO-focus examples this conflicts with
a dispreference for nuclear accents on repeated material, and an independent preference
for placing the nuclear accent on a separate focused constituent that occurs earlier in
the sentence. This conflict is generally resolved in favor of deaccenting, but the SO
focus can still be marked by the strongest stress in the scope of the item with which
it is associated.
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1034
1032
1035
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
11580
1
7
LANGUAGE, VOLUME 83, NUMBER 2 (2007)
5.3. WHAT ENVIRONMENTS YIELD PITCH-ACCENTLESS FOCI? Our results demonstrate
that in one specific type of environment focus is marked using duration and energy,
and the earlier discussion in this section addressed features of the environment that
might be relevant in producing such realizations. The semantics literature includes
discussions of a number of variants in the basic second-occurrence-focus paradigm. In
this section we consider these variants in the light of what we have learned.33
In the experimental materials, the SO focus is preceded by a nuclear pitch accent,
marking the focus associated with even. In looking through the literature on apparent
dissociation between focus and pitch-accenting, we observe that the great majority of
examples involve an expression that semantically one would expect to be accented,
but that occurs in postnuclear position.
Consider the classic example in 12.
(12) People who GROW rice generally only EAT rice.
(Rooth 1992:109)
Like SO-focus examples, this case involves an apparent mismatch between the prosodic
focus of a VP following a focus-sensitive operator (only) and the semantic focus of
the operator. The prosodic focus, at least as regards pitch movement, is eat, but the
semantic focus of only is rice (as evidenced by the meaning of the sentence). As for
the second-occurrence data we have examined, the semantic focus follows a nuclear
accent, so neutralization of pitch movement is to be expected. It is natural to wonder
whether the semantic focus of only is lengthened in the same way as we found for
deaccented second-occurrence foci. We leave examination of this question for further
work.34
Are there any cases in which a semantic focus lacks accent, but does not follow the
nuclear accent? Dryer (1994) claims that there are and presents a candidate example.
In sentence B of 13, the SO-focus expression a book precedes the nuclear accent on
many people.
(13) A: I hear that John only gave [a book]F to Mary.
B: True, but John only gave [a book]SOF to [many people]F .
33
In the discussion in this section we concentrate on the importance of the relative positioning of SO (or
other) foci and nuclear-accented foci. The role of discourse status or textual repetition has also come up in
prior literature, specifically in the work of Krifka (1997:270–71). Krifka considers variants of SO-focus
examples in which the SO focus is pronominalized, and argues that such examples require pitch accents on
the pronominal foci. If so, then textual identity of the SO focus with the first occurrence may be crucial for
pitch suppression of foci. This accords with Ladd’s (1980) observation that there is a strong tendency not
to accent repeated material in English, mentioned in §5.2 above.
34
Roberts (1996) introduced a more complex type of example, in which a VP (invited Lyn for dinner)
is the semantic focus, one element (Lyn) is contrastive, and the rest of the VP is old. The crucial claim is
that the semantic focus is not marked, since this would tend to produce a pitch accent on dinner, yet there
is (or need not be) any such accent. Once again, we may ask whether in this case the word in question
(dinner) is marked as prominent in some other way, for example, by lengthening and energy effects.
1
231
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1581
1582
11583
1
7
WHEN SEMANTICS MEETS PHONETICS
277
Dryer is certainly right to suggest that the most prominent accent in B’s utterance
can be on many people rather than a book. But it is not clear to us whether this sentence
can be felicitously uttered with no pitch accent on a book. As a matter of fact, Féry
and Ishihara (2005) present evidence from similar examples in German suggesting that
prenuclear SO foci (a book in 13) are marked by pitch accents. These pitch accents
are on average ‘weaker’ (e.g. lower maximum F0 for an H* accent) than pitch accents
on first-mention focus (e.g. many people in 13). Although the new data is not from
English, it is suggestive of a reason why prenuclear SO foci might APPEAR to be unaccented, that is, because the accenting is relatively weak. While we are unable to draw
a firm conclusion about Dryer’s examples at this stage, the Féry and Ishihara results
support our methodological view: questions like those Dryer raises are best studied in
the laboratory.
6. CONCLUSION. The focus of a focus-sensitive operator receives the strongest
phrasal stress in the scope of the operator, usually realized as a nuclear accent. SO focus
provides an extreme test case for this generalization since it is typically deaccented. Yet
our results indicate that SO focus is still marked with the strongest stress in the scope
of the operator.35
This analysis has two implications for theories of focus marking and accentuation.
First, it implies that phrasal stress plays an important role in marking focus, and that
pitch accent is not required, although it is usual. Second, postnuclear deaccenting is
purely a restriction on the appearance of PITCH ACCENTS following a nuclear accent,
while phrasal stress distinctions are permitted following a nuclear accent. This is compatible with analyses of phrasal stress and intonation such as those of Selkirk (1984:
154ff.) and Hayes (1995:396), but is inconsistent with analyses such as that of Bolinger
(1958) which attribute all phrase-level prominence to pitch accents, implying that there
are no prominence distinctions in the postnuclear domain. Ladd (1996) has also proposed a number of arguments for the existence of postnuclear prominence distinctions,
and also argues that focus is marked primarily by metrical stress. The only previous
experimental investigation of postnuclear prominence of which we are aware is Huss
1978, but this study examines lexical stress, not phrasal stress, showing that stress
minimal pairs such as éxport (noun) vs. expórt (verb) are distinguished in postnuclear
position (although the difference is subtle).
As discussed in §5, studies of our production data indicate that pitch-accenting is
optional on SO foci, although it occurs in only a minority of utterances. A more complicated question is whether the majority of SO foci are just like ordinary foci, but with
pitch movement suppressed. Our answer is a qualified yes: it is clear that pitch is
usually suppressed in the relevant examples, while other parts of SO-focus marking
are similar to ordinary focus marking, but we cannot conclude that pitch is the ONLY
suppressed feature.36 If we hypothesize that pitch suppression forces (most) SO foci
to be marked without a pitch accent, so that they are realized instead with phrasal stress
only, we arrive at a testable prediction for future work: ALL prosodic features that are
shown to correlate with phrasal stress rather than pitch accents should be employed in
SO focus-marking. Vowel quality, for example, may be such a prosodic feature.
35
For a recent theoretical elaboration using our results, see Büring 2006.
Jaeger (2004:§5) presents evidence that intensity marking, unlike durational marking, is subject to a
relatively strong reduction on SO-focus expressions.
36
1
312
278
1080
As regards the semantics and pragmatics of focus, we have demonstrated that an
argument from SO focus found in the literature is invalid. The original argument implied
a need for strong, pragmatic theories of focus rather than weak theories in which the
interpretative effects of focus are mediated via the syntax-semantics interface. The
argument took as its empirical basis a claim that certain semantic foci are not phonologically focus-marked, and, under further assumptions, hence not F-marked. We have
shown, however, that there are no grounds for denying that these semantic foci are
phonologically focus-marked: SO foci have acoustic properties that, in the absence of
phonological focus marking, would be unexplained. So the original argument fails.
Does it follow that pragmatic theories of focus are incorrect? Not at all. Indeed, our
data shows that though SO focus is consistently marked and is typically above the
threshold of what is perceptible, it may be that hearers can retrieve that marking with
far less than perfect accuracy. Pragmatic theories could then describe an important
source of information available in the processing of SO focus (as suggested to us by
Jaye Padgett, p.c.). But such a processing claim would be independent of whether, as
required by syntactic and semantic theories, association with focus is grammaticized,
a claim about competence, not performance. It is now clear that standard arguments
from SO focus cannot settle this competence issue.
We began this article with Ladd’s pessimistic observation ‘that proposals about intonational meaning are not a reliable source of evidence on intonational phonology’. We
have shown that work on intonational (or, perhaps better, PROSODIC) meaning can provide a useful source of evidence for suprasegmental phonology, but we have also sought
to demonstrate that study of suprasegmental phonology is a necessary component of
work on intonational meaning. In the process, we have revealed SO focus to be a
phenomenon of considerable phonetic and phonological subtlety.
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1108
1111
1113
1114
1117
1119
1121
1123
1125
1126
1128
1130
1132
1135
1137
1138
1141
1143
1145
1148
1150
1151
1154
1156
1158
1161
1
1
7
LANGUAGE, VOLUME 83, NUMBER 2 (2007)
APPENDIX A: STIMULI
(1)
a.
b.
(2)
a.
b.
(3)
a.
b.
(4)
a.
b.
(5)
a.
Twins Kate and Jane usually get lots of cards from their friends on their birthday. But Jim
only sent Kate a card today.
Even Jack only sent Kate a card today.
Kate usually gets lots of nice presents on her birthday. But her brother only gave Kate a
card today.
Even her mother only gave Kate a card today.
Pete really needed an injection to ease his pain. But the nurse only gave Pete a pill today.
Even the doctor only gave Pete a pill today.
Both Pete and Edward are suffering from the flu. But the nurse only gave Pete a pill today.
Even the doctor only gave Pete a pill today.
Both Sid and his accomplices should have been named in this morning’s court session. But
the defendant only named Sid in court today.
Even the state prosecutor only named Sid in court today.
Defense and Prosecution had agreed to implicate Sid both in court and on television. Still,
the defense attorney only named Sid in court today.
Even the state prosecutor only named Sid in court today.
The family cat either stays in the tent or caravan. But mom only let the cat in the tent
today.
Even the kids only let the cat in the tent today.
The cat and the dog usually stay in the tent. But mom would only let the cat in the tent
today.
Even the kids would only let the cat in the tent today.
At the San Francisco zoo, the chimps love nuts and fruit. But tourists always throw nuts
to the chimps there.
Even the guides always throw nuts to the chimps there.
1
231
1163
1164
1167
1169
1171
1174
1176
1177
1181
1183
1185
1188
1190
1191
1194
1196
WHEN SEMANTICS MEETS PHONETICS
b.
(6)
a.
b.
(7)
a.
b.
279
At the Los Angeles zoo, both chimps and baboons love nuts. But tourists always throw
nuts to the chimps there.
Even the guides always throw nuts to the chimps there.
You might think that in the prestigious Clark company of architects, all drafts were done
on the computer. But the intern always uses a pen for drafts there.
Even the chief architect always uses a pen for drafts there.
In some architecture companies, final versions of floor plans are drawn with pens, but this
is different at Flemming Associated Architects. The intern always uses a pen for drafts
there.
Even the chief architect always uses a pen for drafts there.
You might think that Texas drugstores sell both small toys and sweets to kids. But they
always sell sweets to kids there.
Even Walgreens always sells sweets to kids there.
You might think that Texas drugstores sell sweets to both adults and kids. But they always
sell sweets to kids there.
Even Walgreens always sells sweets to kids there.
APPENDIX B: FURTHER
DETAILS ON THE PRODUCTION EXPERIMENT
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
B.1. EXCLUSION CRITERIA. As mentioned in §3.2, not all elicited sentences could be used in the phonetic
analysis. Some items were lost through experimental error, though this did not affect overall balance of the
study, and some items were qualitatively not acceptable. In this section, we elaborate on our exclusion
criteria.
A blind review of the data was performed and all experimental sentences that contained hesitations,
stuttering, or other significant reading effects were removed. Whenever a sentence was removed from the
analysis, the matching sentence (with the inverted SO-focus assignment) was also removed. This resulted
in the removal of 17 ⳯ 2 ⳱ 34 experimental items (corresponding to 6% of all items). A lost sound file
and two targets on which no measures could be obtained resulted in another 0.9% data loss. We then averaged
over the two repetitions of each stimulus (or used only one if the other had been excluded), yielding 556
targets of a possible 560 (as calculated by taking two targets from each of seven dialogue items, each under
two different focus structures, multiplied by twenty speakers, thus 2 ⳯ 7 ⳯ 2 ⳯ 20 ⳱ 560). These 556
targets were available for the duration, energy, and intensity analyses, corresponding to a data loss of 0.7%,
as cited in the main text. The pitch analysis was performed on 544 targets (2.8% data loss). This was due
to further exclusion of items for which either (i) the pitch tracking failed to find a pitch value, (ii) the
observed pitch value was likely to be incorrect (i.e. the maximum pitch on the target deviated more than
two standard deviations from the mean pitch of the whole utterance), (iii) the pitch on the target had a range
of 0, or (iv) an otherwise usable item was removed due to problems (i) to (iii) on its matching item.
1215
1216
1217
B.2. TESTS OF NORMALITY. The standard normality tests were conducted, and in cases where a dependent
variable was not sufficiently normally distributed the usual transformations were used to meet the normality
conditions (ln(x)) for duration data; x2 for the energy measure; 兹x for root-mean-square intensity).
1218
REFERENCES
ANDERSON, STEPHEN. 1972. How to get even. Language 48.4.893–906.
BARTELS, CHRISTINE. 1997. Acoustic correlates of ‘second occurrence’ focus: Towards an
experimental investigation. In Kamp & Partee, 11–30.
BARTELS, CHRISTINE, and JOHN KINGSTON. 1984. Salient pitch cues in the perception of
contrastive focus. Focus and natural language processing, vol. 1: Intonation and syntax, ed. by Peter Bosch and Rob van der Sandt, 11–28. Heidelberg: IBM Working
Papers of the Institute for Logic and Linguistics.
BEAVER, DAVID, and BRADY CLARK. 2003. Always and only: Why not all focus sensitive
operators are alike. Natural Language Semantics 11.4.323–62.
BECKMAN, MARY E. 1986. Stress and non-stress accent. Dordrecht: Foris.
BECKMAN, MARY E., and GAYLE A. ELAM. 1997. Guidelines for ToBI labelling (version
3.0). Columbus: The Ohio State University, MS.
BECKMAN, MARY E., and JANET B. PIERREHUMBERT. 1986. Intonational structure in Japanese
and English. Phonology Yearbook 3.255–309.
BOLINGER, DWIGHT. 1958. A theory of pitch accent in English. Word 14.3.109–49.
BÜRING, DANIEL. 1997. The meaning of topic and focus—The 59th Street Bridge accent.
London: Routledge.
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
11235
1
7
1
312
280
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
11293
BÜRING, DANIEL. 2006. Been there, marked that—A tentative theory of second occurrence
focus. Los Angeles: University of California, Los Angeles, MS.
CHOMSKY, NOAM. 1972. Studies on semantics in generative grammar, 62–119. The Hague:
Mouton.
CHOMSKY, NOAM. 1976. Conditions on rules of grammar. Linguistic Analysis 2.4.303–51.
CLARK, HERBERT H. 1973. The language-as-fixed-effect fallacy. Journal of Verbal Learning
and Verbal Behavior 12.4.335–59.
COHAN, JOCELYN. 2000. The realization and function of focus in spoken English. Austin:
University of Texas, Austin dissertation.
DE JONG, KENNETH J. 1995. The supraglottal articulation of prominence in English: Linguistic
stress as localized hyperarticulation. Journal of the Acoustical Society of America
97.1.491–504.
DE MEY, SJAAK. 1991. ‘Only’ as a determiner and as a generalized quantifier. Journal of
Semantics 8.1–2.91–106.
DRYER, MATTHEW S. 1994. The pragmatics of association with only. Paper presented at the
annual meeting of the Linguistic Society of America, Boston, January.
FÉRY, CAROLINE, and SHINICHIRO ISHIHARA. 2005. Interpreting second occurrence focus.
Potsdam: University of Potsdam, MS.
FOOLEN, AD. 1993. De betekenis van partikels: Een dokumentatie van de stand van het
onderzoek met bijzondere aandacht voor ‘maar’. Nijmegen: Catholic University of
Nijmegen dissertation.
GEURTS, BART, and ROB VAN DER SANDT. 2004. Interpreting focus. Theoretical Linguistics
30.1.1–44.
GRICE, H. PAUL. 1989. Studies in the way of words. Cambridge, MA: Harvard University
Press.
GUSSENHOVEN, CARLOS. 1984. On the grammar and semantics of sentence accents. Dordrecht: Foris.
HAJIČOVÁ, EVA. 1973. Negation and topic vs. comment. Philologica Pragensia 16.81–93.
HAJIČOVÁ, EVA. 1984. Presupposition and allegation revisited. Journal of Pragmatics
8.2.155–67.
HAJIČOVÁ, EVA; BARBARA H. PARTEE; and PETR SGALL. 1998. Topic-focus articulation,
tripartite structures, and semantic content. Dordrecht: Kluwer.
HALLIDAY, MICHAEL. 1967. Notes on transitivity and theme in English (part 2). Journal of
Linguistics 3.2.199–244.
HARRINGTON, JONATHON; JANET FLETCHER; and MARY E. BECKMAN. 2000. Manner and place
conflicts in the articulation of accent in Australian English. Papers in laboratory phonology 5: Acquisition and the lexicon, ed. by Michael B. Broe and Janet B. Pierrehumbert, 40–51. Cambridge: Cambridge University Press.
HAYES, BRUCE. 1995. Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press.
HEDBERG, NANCY. 2003. The prosody of contrastive topic and focus in spoken English.
Pre-proceedings of the workshop on information structure in context, 141–52. Stuttgart:
Institut für Maschinelle Sprachverarbeitung.
HEDBERG, NANCY, and JUAN M. SOSA. 2001. The prosodic structure of topic and focus in
spontaneous English dialogue. Paper presented at Topic and Focus: A workshop on
intonation and meaning, Linguistic Society of America Summer Institute of Linguistics,
University of California, Santa Barbara.
HEDBERG, NANCY, and JUAN M. SOSA. 2002. The prosody of questions in natural discourse.
Proceedings of Speech Prosody 2002: The first international conference on speech
prosody, 375–78. Aix-en-Provence: ISCA.
HORN, LAURENCE. 1996. Exclusive company: Only and the dynamics of vertical inference.
Journal of Semantics 13.1.1–40.
HUSS, VOLKER. 1978. English word stress in post-nuclear position. Phonetica 35.2.86–105.
JACKENDOFF, RAY. 1972. Semantic interpretation in generative grammar. Cambridge, MA:
MIT Press.
JAEGER, T. FLORIAN. 2004. Only always associates audibly, even if only is repeated: The
prosodic properties of second occurrence focus in English. Stanford, CA: Stanford
University, MS.
1
7
LANGUAGE, VOLUME 83, NUMBER 2 (2007)
1
231
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
11351
1
7
WHEN SEMANTICS MEETS PHONETICS
281
KADMON, NIRIT. 2001. Formal pragmatics: Semantics, pragmatics, presupposition and
focus. Oxford: Blackwell.
KAMP, HANS, and BARBARA H. PARTEE (eds.) Context-dependence in the analysis of linguistic
meaning: Proceedings of the workshops in Prague and Bad Teinach. Stuttgart: Institut
für Maschinelle Sprachverarbeitung, University of Stuttgart.
KAYNE, RICHARD. 1998. Overt vs. covert movement. Syntax 1.2.128–91.
KRIFKA, MANFRED. 1992. A compositional semantics for multiple focus constructions. Informationsstruktur und Grammatik 4, ed. by Joachim Jacobs, 17–54. Opladen: Westdeutscher.
KRIFKA, MANFRED. 1997. Focus and/or context: A second look at second occurrence expressions. In Kamp & Partee, 253–76.
KRIFKA, MANFRED. 2006. Association with focus phrases. The architecture of focus, ed. by
Valerie Molnar and Susanne Winkler, 105–36. Berlin: Mouton de Gruyter.
LADD, D. ROBERT. 1980. The structure of intonational meaning: Evidence from English.
Bloomington: Indiana University Press.
LADD, D. ROBERT. 1996. Intonational phonology. Cambridge: Cambridge University Press.
LADEFOGED, PETER. 2003. Phonetic data analysis: An introduction to fieldwork and instrumental techniques. Oxford: Blackwell.
MARTÍ, LUISA. 2003. Contextual variables. Storrs: University of Connecticut, Storrs dissertation.
MONTAGUE, RICHARD. 1974. The proper treatment of quantification in ordinary English.
Formal philosophy: Selected papers of Richard Montague, ed. by Richmond Thomason,
247–70. New Haven: Yale University Press.
NEVALAINEN, TERTTU. 1991. But, only, just: Focusing adverbial change in modern English
1500–1900. Helsinki: Société Néophilologique.
OSTENDORF, MARI; PATTI J. PRICE; and STEFANIE SHATTUCK-HUFNAGEL. 1995. The Boston
University radio news corpus. (Boston University technical report ECS-95-001.) Boston, MA: Boston University.
PARTEE, BARBARA H. 1999. Focus, quantification, and semantics-pragmatics issues. Focus:
Linguistic, cognitive, and computational perspectives, ed. by Peter Bosch and Rob van
der Sandt, 213–31. Cambridge: Cambridge University Press.
PAUL, HERMANN. 1888. Principles of the history of language. London: Swan Sonnenschein,
Lowrey, & Company. (Translation of Principien der Sprachgeschichte.)
PIERREHUMBERT, JANET B. 1980. The phonology and phonetics of English intonation. Cambridge, MA: MIT dissertation.
PIERREHUMBERT, JANET B., and JULIA HIRSCHBERG. 1990. The meaning of intonational contours in the interpretation of discourse. Intentions in communication, ed. by Philip
Cohen, Jerry Morgan, and Martha Pollack, 271–311. Cambridge, MA: MIT Press.
PRINCE, ELLEN. 1981. Toward a taxonomy of given-new information. Radical pragmatics,
ed. by Peter Cole, 223–55. New York: Academic Press.
RAAIJMAKERS, JEROEN G. W.; JOSEPH M. C. SCHRIJNEMAKERS; and FRANS GREMMEN. 1999.
How to deal with ‘the language-as-fixed-effect fallacy’: Common misconceptions and
alternative solutions. Journal of Memory and Language 41.3.416–26.
ROBERTS, CRAIGE. 1995. Domain restriction in dynamic semantics. Quantification in natural
languages, ed. by Emmon Bach, Eloise Jelinek, Angelika Kratzer, and Barbara H.
Partee, 661–700. Dordrecht: Kluwer.
ROBERTS, CRAIGE. 1996. Information structure in discourse: Towards an integrated formal
theory of pragmatics. OSU Working Papers in Linguistics: Papers in semantics
49.91–136
ROOTH, MATS. 1985. Association with focus. Amherst: University of Massachusetts, Amherst
dissertation.
ROOTH, MATS. 1992. A theory of focus interpretation. Natural Language Semantics
1.1.75–116.
ROOTH, MATS. 1996a. Focus. The handbook of contemporary semantic theory, ed. by Shalom
Lappin, 271–97. London: Basil Blackwell.
ROOTH, MATS. 1996b. On the interface principles for intonational focus. Proceedings of
Semantics and Linguistic Theory (SALT) 6, ed. by Teresa Galloway and Justin Spence,
202–26. Ithaca, NY: Cornell University.
1
312
282
LANGUAGE, VOLUME 83, NUMBER 2 (2007)
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1401
1400
1399
SCHWARZSCHILD, ROGER. 1997. Why some foci must associate. New Brunswick, NJ: Rutgers
University, MS.
SCHWARZSCHILD, ROGER. 1999. GIVENness, AvoidF and other constraints on the placement
of accent. Natural Language Semantics 7.2.141–77.
SELKIRK, ELIZABETH. 1984. Phonology and syntax: The relation between sound and structure. Cambridge, MA: MIT Press.
SELKIRK, ELIZABETH. 1995. Sentence prosody: Intonation, stress, and phrasing. The handbook
of phonological theory, ed. by John A. Goldsmith, 550–69. London: Basil Blackwell.
SHATTUCK-HUFNAGEL, STEFANIE, and ALICE E. TURK. 1996. A prosody tutorial for investigators of auditory sentence processing. Journal of Psycholinguistic Research
25.2.193–247.
SILVERMAN, KIM. 1987. The structure and processing of fundamental frequency contours.
Cambridge: Cambridge University dissertation.
SLUIJTER, AGAATH M. C., and VINCENT J. VAN HEUVEN. 1996a. Acoustic correlates of linguistic stress and accent in Dutch and American English. Proceedings of the fourth International Conference on Spoken Language Processing, vol. 2, 630–33.
SLUIJTER, AGAATH M. C., and VINCENT J. VAN HEUVEN. 1996b. Spectral balance as acoustic
correlate of linguistic stress. Journal of the Acoustic Society of America 101.1.503–13.
STEEDMAN, MARK. 2000. Information structure and the syntax-phonology interface. Linguistic Inquiry 31.4.649–89.
TANCREDI, CHRISTOPHER D. 1990. Not only even, but even only. Cambridge, MA: MIT, MS.
TERKEN, JACQUES, and JULIA HIRSCHBERG. 1994. Deaccentuation of words representing given
information: Effects of persistence of grammatical function and surface position. Language and Speech 37.2.125–45.
TRUCKENBRODT, HUBERT. 1995. Phonological phrases: Their relation to syntax, focus, and
prominence. Cambridge, MA: MIT dissertation.
TURK, ALICE E., and LAWRENCE WHITE. 1999. Structural influences on accentual lengthening.
Journal of Phonetics 27.171–206.
VALLDUVÍ, ENRIC. 1990. The information component. Philadelphia: University of Pennsylvania dissertation.
VON FINTEL, KAI. 1994. Restrictions on quantifier domains. Amherst: University of Massachusetts, Amherst dissertation
VON FINTEL, KAI. 1997. A minimal theory of adverbial quantification. In Kamp & Partee,
153–93.
VON STECHOW, ARNIM. 1985/1989. Focusing and backgrounding operators. (Technical report.) Konstanz: University of Konstanz.
WAGNER, MICHAEL. 2005. Prosody and recursion. Cambridge, MA: MIT dissertation.
WATSON, DUANE; CHRISTINE GUNLOGSON; and MICHAEL TANENHAUS. 2006. Online methods
for the investigation of prosody. Methods in empirical prosody research, ed. by Stefan
Sudhoff, Denisa Lenertová, Roland Meyer, Sandra Pappert, Petra Augurzky, Ina
Mleinek, Nicole Richter, and Johannes Schließer, 259–82. Leipzig: Mouton de Gruyter.
WELBY, PAULINE. 2003. Effects of pitch accent position, type and status on focus projection.
Language and Speech 46.1.53–81.
WILLIAMS, EDWIN. 1997. Blocking and anaphora. Linguistic Inquiry 28.4.577–628.
WOLTERS, MARIA, and DAVID BEAVER. 2001. What does he mean? Proceedings of the twentythird annual meeting of the Cognitive Science Society, ed. by Johanna Moore and Keith
Stenning, 1176–80. Mahwah, NJ: Lawrence Erlbaum.
Beaver
Department of Linguistics
The University of Texas at Austin
Calhoun Hall 501
1 University Station B5100
Austin, TX 78712-0198
[[email protected]]
1
1
7
[ Received 27 December 2004;
revision invited 10 July 2005;
revision received 19 April 2006;
accepted 9 October 2006]