| EDITORS: Fabricius-Hansen, Cathrine; Ramm, Wiebke
TITLE: 'Subordination' versus 'Coordination' in Sentence and Text
SUBTITLE: A cross-linguistic perspective
SERIES: Studies in Language Companion Series 98
PUBLISHER: John Benjamins
Rui P. Chaves, Department of Linguistics, University at Buffalo, State
University of New York
This book focuses on the study of Coordination and Subordination at both the
sentence and the discourse levels. This makes this volume a particularly
interesting one, since it brings together avenues of research that would
otherwise be separate. This publication results from a workshop held during the
28th Annual Meeting of the Deutsche Gesellschaft für Sprachwissenschaft, in
February 2006. The book is divided in four parts. Part I is entitled ''General
and theoretical issues'' and is dedicated to central theoretical questions. Part
II is entitled ''Cross-linguistic approaches'' and contains corpus-based
cross-linguistic research about clause combination and discourse structure
(spanning various languages such as English, German, Dutch, French and
Norwegian). Part III is entitled ''Monolingual studies'' and contains papers
addressing specific issues pertaining to German, English or French. Finally,
Part IV is entitled ''Diachronic perspectives'' and offers a diachronic
perspective of subordination, coordination and rhetorical relations in early
Germanic languages. The book aims to contribute to a better understanding of
This volume starts off with a rather lengthy introduction produced by the
editors (20 pages, larger than some of the papers), which goes far beyond
introducing the volume and laying out the goals and contents of the publication.
The editors' introduction describes the linguistic issues under discussion in
some detail and provides a brief introduction to the perspectives and
theoretical frameworks available in the literature, while at the same time
referring the reader to the collected papers that make a contribution to that
particular linguistic issue and/or framework. The introduction thus paints an
interesting research landscape, spanning empirical issues mainly pertaining to
syntax, semantics, pragmatics, and discourse.
The introduction is valuable and it would have been very difficult to provide a
more detailed overview of the phenomena without making it longer than the rest
of the book. However, the non-expert reader should be aware that the
introduction does not come near to exhausting the issues and theoretical
frameworks pertaining to the analysis of coordination/subordination, nor does it
cite the standard, original literature on these topics. For example, when
discussing the concept of asymmetry in coordination the introduction cites
recent work from 2000 and 2005, while the issue (and the terminology) has been
much debated since the 60s (see for example Ross (1967), Schmerling (1972,
1975), Goldsmith (1985), Lakoff (1986), and Levin and Prince (1986), inter
alia). A second issue concerns that more emphasis is given to semantics rather
than syntax, but this is more likely to be a reflection of the collected papers.
All in all, the introduction does a good job at interweaving a pre-theoretical
exposition of the phenomena, the collected papers, and the frameworks discussed
therein. The introduction ends with a brief exposition of each paper,
cross-referencing terms and issues across chapters, which is also very useful.
PART I. General and theoretical issues
Manfred Stede (''RST revisited: Disentangling nuclearity'') draws attention to a
number of problems concerning the central notion of nuclearity in Rhetorical
Structure Theory (RST). Stede shows that previous accounts positing linear
correlations between nuclearity on the one hand, and syntactic patterns and
coherence relations on the other, are excessively strong. A number of
counter-arguments are provided to show that the relevant generalization is being
missed, suggesting that the very notion of nuclearity should not be a primitive
concept of RST. Stede's proposal consists in adopting a broader and more
flexible concept -- salience -- and abandoning the requirement that coherence
and discourse organization are necessarily represented in a tree structure.
Stede argues convincingly that what is needed is a multi-level representation
framework, with distinct dimensions of coherence working in parallel. The paper
goes on to outline a multi-level annotation (MLA) schema, which was applied to
the Potsdam Commentary Corpus with the help of a number of RST-based annotation
The paper, however, does not present a very detailed discussion of the
annotation levels, and so the real benefits of the proposal are not easy to
assess objectively. This may be partly due to the fact that there are many
possible configurations of annotation levels, and that such possibilities can
only be explored via future work. Still, some of this research is work in
progress, and Stede provides a number of hyperlinks as further sources of
valuable information about this enterprise.
Hardarik Blühdorn (''Subordination and coordination in syntax, semantics and
discourse'') argues against drawing too many theoretical parallels between
syntactic and discourse structure, and claims that syntactic hierarchy should
not be considered a general model for the conceptualization of discourse
hierarchy, nor vice versa. Blühdorn resorts to various data from German in
support of the cautionary claims.
Most of Blühdorn's data are not new, and have been discussed since Ross (1967),
but I agree that syntactic structure and discourse structure are different
domains altogether, and that the combinatorial semantic processes that are
observed in these domains are also different. However, Blühdorn goes further in
claiming that the relation between syntax, semantics and discourse is a matter
of rhetorical option, and that the concepts of subordination and coordination in
language are terminological metaphors.
These claims are mainly based on data that are given to show that symmetrical
(non-hierarchical; typically coordinate) connections and asymmetrical
(hierarchical; typically subordinate) connections can be encoded either by
coordinators or subordinators. The data are given in German, but have an
identical behavior in English:
1) a. The penguins were yellow-brown, and the giraffes were black and white. b.
The giraffes were black and white, and the penguins were yellow-brown.
2) a. The penguins were yellow-brown, while the giraffes were black and white.
b. While the giraffes were black and white, and the penguins were yellow-brown.
Blühdorn observes that the pairs of clauses in (1) and in (2) can be interpreted
'symmetrically', and concludes that they are equivalent counterparts. He uses
the term 'symmetric' because reordering the embedded clauses in each sentence
does not yield a meaning contrast. It is therefore argued that the concepts of
subordination and coordination in language are terminological metaphors, given
that both subordination and coordination can give rise to semantically symmetric
and asymmetric structures.
There are problems with the notion of symmetry that is being alluded to here for
non-hierarchical structures. It is claimed that the daughters in these exhibit
equal semantic functions and equal semantic weight. I'm unsure about how to
evaluate the claim about weight since one can certainly conjoin sentences of
varying size and semantic content (e.g. ''A gentle breeze stirred the air and a
silverback gorilla who had for far too many years roamed the jungle alone
drowsed under a large and ancient tropical tree''). But a deeper problem has to
do with a number of well-known symmetric properties that the coordination
examples in (1) have, and that the (allegedly symmetric) subordination sentences
in (2) lack. For example, only the examples in (1) are subject to Ross's
constraints on coordinate extraction. That is, no conjunct can be extracted and
if one conjunct contains an embedded extraction site, so must all other
conjuncts contain an embedded extraction site:
3) a. *And the giraffes were black and white, the penguins were yellow-brown.
b. *And the penguins were yellow-brown, the giraffes were black and white.
4) a. *This is the animal that [[Kim photographed _ ], [and Sue fed the
giraffes]]. b. *This is the animal that [Kim photographed the penguins], [and
Sue fed _ ]]. c. This is the animal that [Kim photographed _ ], [and Sue fed _ ]].
This symmetry between conjuncts is lost in the case of subordination, as shown
in the examples in (2). The subordinate clause can be moved, and can contain an
extraction site only if the head contains one too:
5) a. While the giraffes were black and white, the penguins were yellow-brown.
b. While the penguins were yellow-brown, the giraffes were black and white.
6) a. This is the animal that [[Kim photographed _ ], [while Sue fed giraffes]].
b. *This is the animal that [[Kim photographed the penguins], [while Sue fed
_]]. c. This is the animal that [[Kim photographed _ ], [while Sue fed _]].
If the data were truly symmetrical, then one would not expect such a difference,
which the literature usually attributes to semantic parallelism properties
specific to coordination. It is not clear what is at stake here, when one deems
(1) and (2) as equally symmetric.
There are also different properties with regard to non-clausal tense
constraints. One of the trademarks of coordination is that different tenses can
be conjoined as long as they are finite. This is not the case in subordination:
7) Kim photographed the penguins while feeding hay to the giraffes.
8) *Kim photographed the penguins and feeding hay to the giraffes.
This evidence suggests that the two structures that compose the examples in (2)
are not truly symmetric on syntactic and semantic grounds. Other differences
could be enumerated here, but a simpler explanation for the apparent syntactic
symmetry in (2) is that the semantics of 'while' is symmetric, while the
construction itself is not. More specifically, to say that the time interval t1
in which proposition P1 occurs intersects the time interval t2 in which P2
occurs is the same as saying that the time interval t2 in which P2 occurs
intersects the time interval t1 in which P1 occurs. Regardless of this symmetry,
the clause introduced by 'while' is not nuclear as the evidence in (5) and (6)
shows, which in turn indicates that it plays a different semantic role than the
one played by the head clause. Hence, they are not only asymmetric syntactically
but also, in a way, semantically.
These data cast doubts on the paper's claim that discourse connections are
independent from the syntactic distinction between coordination and
subordination. Matters are not that flexible, as the evidence shows. Syntactic
subordination and syntactic coordination do trigger different semantic
properties. I suspect that the correct generalization lies somewhere in the
middle: syntax, semantics, pragmatics, and discourse all give their contribution
for the communicative goal of a sentence. The latter is closer to the position
advocated in Kehler (2002), for example, where discourse coherence is argued to
be responsible for the kinds of extraction and ordering patterns -- as well as
certain deletion phenomena, such as Gapping -- that are known to be only
observed in symmetric coordinate structures. Blühdorn also proposes that while
the purpose of discourse structure is to present information in order to achieve
certain communicative goals, the purpose of syntactic structure is ''to arrange
formal expressions in order to facilitate parsing'' (81). I find this hypothesis
difficult to accept, as it seems to entail that all languages should be
structurally identical, non-ambiguous, and rather rigid with regard to word
order. It is unclear to me how word order and syntactic structure can aid
parsing without these having anything to do with semantics and communicative goals.
PART II. Cross-linguistic approaches
Christelle Cosme (''A corpus-based perspective on clause linking patterns in
English, French and Dutch'') provides a cross-linguistic corpora study where
various clause-linking patterns are compared. The study is based on comparable
(authentic) corpora as well as on translated corpora. There are two major claims
that the paper puts forth. First, that the three languages differ quantitatively
in how clausal coordination and paratactic clause combination are used, and
therefore, that they differ with regard to how information packaging is
preferentially encoded (more specifically, it is claimed that Dutch uses more
coordination than English, and that English uses more coordination than French;
this is dubbed the 'Dependency Hypothesis'). The second claim is that clausal
coordination and subordination are two extremes in a gradient clause combination
continuum. Cosme presents a study based on a trilingual corpus from which 300
random sentences were extracted, in each of the three languages. The comparison
method was partially automatic, and very much contingent on the annotation
scheme. The results suggest that Dutch employs fewwer subordinate clauses than
English and French, but that no major difference is noted between French and
English. In order to refine the study and provide support for the Dependency
Hypothesis, Cosme undertakes a finer-grained follow-up study of coordination
that takes into consideration both clausal coordination and VP-coordination. The
results suggest that French indeed uses less overall coordination (S and VP)
than English does, in support of the Dependency Hypothesis.
At this point, a number of concerns arise. First, one can argue that the sample
size of 300 sentences is too small (causing the total number of coordinations in
the French sample to be only 28). It would be useful to repeat the study with
larger samples, or alternatively, to repeat the study and measure variance in
the overall results. A related concern is that the cross-linguistic differences
in the frequency of VP coordination are in fact statistically not significant
(although come to a close p <= 0.07). This raises some doubts about the
annotation choices, the sampling, and the comparison methodology.
In a second stage of the study, translation corpus data were used to show that
translators more often shift a coordinate structure into a subordinate structure
when translating from English to French, and that translators more often shift a
subordinate structure into a coordinate structure. It is not clear what is
behind these results. The author notes that some shifts are caused by necessity
and others by style, but since the analysis cannot tell these apart, it is
unclear if style plays a larger role than syntax, for example, or if the role of
style is significant.
No discussion is devoted to possible confounding factors that are particularly
relevant with small samples, and that may interfere with the results. For
example, there can be one translator overrepresented in the sample, or there can
be biases caused by the native language of the translator, or by some
translation strategy that these professionals develop over time. Finally, how
natural do these translations sound to a naive native speaker who is not a
translator by profession?
The paper goes on to argue that the aforementioned statistical data supports the
theoretical claim that there is a continuum in clause linking.
I agree with the idea of a continuum, but not with the particulars of Cosme's
conclusion. Cosme places on one end of the scale the coordination of full
clauses, then VP conjunction, then hypotaxis, and finally, clause embedding. I
find that this scale is objectionable and not sufficiently supported. For
example, there is no reason why VP-conjunction is closer to embedding than
clausal coordination (other than the existence of an external object, but the
same can be said about S coordination and dangling modifiers, topicalized
phrases and the like): it has all the same syntactic and semantic properties
that are observed in clausal conjunction (with regard to extraction,
reversibility, ellipsis, symmetric/assymmetric readings, etc.). When discussing
the role of the rightmost conjunct in a VP coordination, Cosme claims that it
''exhibits a higher degree of dependency, because it depends on the first clause
for the expression of one argument, namely the subject, (...)'' (109). Thus,
Cosme seems to be assuming a syntactic analysis for 'VP conjunction' where [S
[and VP]], which is a fairly non-standard view of the structure of VP
conjunction. It has long been recognized that VP conjunction exists as such and
that something closer to [VP [and VP]] is more adequate. Not only does this
provide an account that can scale to all other conjoinable categories, but also
it can cope with classical VP coordination examples like ''You can't [drive a car
and talk on the phone]'' and ''She both [loved her subjects and was loved by
them]''. See for example McCawley (1988).
The support for Cosme's continuum hinges on particular theoretical assumptions,
rather than a more finer-grained empirical basis. It seems more likely that
there is such a continuum with regard to the connectors used in clause-linking:
'and', 'but', 'as well as', 'just as', 'while', 'because' (see for instance
Huddleston and Pullum (2002,Ch.15)).
Kare Solfjeld (''Sentence splitting - and strategies to preserve discourse
structure in German-Norwegian translations'') presents a corpora study about
syntactic and semantic patterns in the translation from German prose to
Norwegian, and examines the relation between syntactic
subordination/coordination and discourse interpretation. Various studies have
observed that German adjuncts at the NP or VP level are often translated as full
sentences in the Norwegian target texts. Solfjeld's data indicate that the
choice of target structure (when adjuncts at the NP and VP levels in German are
split off into separate sentences in the Norwegian targets) depends on
maintaining the same discourse structure: if the information is backgrounded in
the source then it is typically realized to the left, but if realized on the
focal part then typically the target translation is located where clear
anaphoric links can be made, which is usually to the right side of the target
utterance. Clause coordination with 'og' (and) is used when grammatical options
are lacking for expressing backgrounding via subordination.
Syntactic subordination often signals informational backgrounding or
'downgrading', and when adjuncts are translated as full sentences it may be hard
to recover said informational backgrounding/downgrading. This poses a challenge
for translators, and Solfjeld's study examines the strategies that Norwegian
translators adopt in order to prevent subordinate - and therefore downgraded -
information in the German source to be upgraded in the Norwegian translation.
Solfjeld's study considers 274 cases of sentence splitting, where a NP or VP
adjunct in the German source results in a separate sentence or a conjunct in the
German translation. These consist of 13 excerpts of German texts where each
translator is represented only once. The results indicate that the surface order
of realization of the adjuncts in German and corresponding adjuncts/conjuncts in
the target Norwegian is relatively similar. In about half of the cases where the
German source contains a preposed adjunct, the Norwegian target is rendered via
a coordination. Solfjeld argues that the reason that these structures can be
rendered as coordinations is that having a 'coordinate unit' prevents the
possibility of each conjunct relating to the context separately. Moreover, since
coordinations can be interpreted assymetrically, the adjunct turned conjunct can
have a lesser role with regard to information structure.
Solfjeld's observation is in fact a well-established property about
coordination, even though historically the first accounts of it attempted to
view these kinds of coordination as a form of subordination (e.g. Ross (1967)).
The study goes on to suggest that strategies to preserve discourse structure
differ when the adjuncts that trigger sentence splitting are in the focus part
of the source sentence. Solfjeld argues that adjuncts relatively far to the
right in the source sentence are often in the focused part of the utterance, and
that in the splitting stage these adjuncts often become separate sentences to
the right. More generally, this study indicates that the relative position of
the information elements are typically kept intact, which suggests that there is
a fairly linear mapping of the information structure of the monosentential
version onto the multisentence version.
An important follow-up for this research would be a quantitative comparison
study comparing target Norwegian prose and non-translational Norwegian prose. If
Solfjeld is right about the order patterns found in translational texts being
contingent on the German sources, then one would expect different quantitative
results when comparing these texts with non-translational Norwegian.
Wiebke Ramm (''Upgrading of non-restrictive relative clauses in translation'')
investigates whether the discourse-functional effect of the distinction between
appositive/discontinuative relative clauses has an effect on German to Norwegian
translations. It is known that German non-restrictive relative clauses (hf.
NRRCs) can have various different discourse functions
(discontinuative/appositive or a continuative discourse function), and the
question that the paper addresses is to what extent does clause linkage in the
target Norwegian makes a difference for the discourse interpretation of the
original vs. translated versions of the text. Ramm thus focuses on the upgrading
of NRRCs to independent main clauses in the target Norwegian translations, and
studies the impact that the upgrading to main clause has on the interpretation.
It is not unusual for German NRRCs to be translated as independent clauses in
Norwegian (in some cases there is no alternative translation), but Ramm
concludes from the observed data that the upgrading is only problematic in the
case of relative clauses with a discontinuative/appositive status. German
continuative relative clauses (a subtype that can only occur in sentence-final
position) are not syntactically embedded and thus do not yield rise to major
interpretation differences when split and translated as main clauses. Ramm
considers the hypothesis that the closer the relative clause is to an
independent main clause, the less translational upgrading will be available to
alter the interpretation in the target version. In turn, this leads to the
prediction that translators avoid upgrading discontinuative/appositive relative
clauses in Norwegian translations and use other strategies instead, in order to
produce a less coherent text or one with a significantly different interpretation.
In her corpus study, Ramm finds some quantitative evidence consistent with the
stated hypothesis. However, the quantitative study is potentially confounded
with the difficulty of identifying the individual meanings of the various
relative adverbs that come under scrutiny. A more robust study about the
reference usage of translators would be psycholinguistic in nature, rather than
corpora-based. In a controlled translation experiment, one could manipulate the
texts and more objectively identify the factors that trigger the various
translation choices. Corpora frequency may result from a number of different
interacting effects, including translator-specific biases, stylistic
preferences, cognitive processing factors, and grammatical differences between
German and Norwegian.
Ramm concludes by arguing that the trade-off between two discourse organization
strategies is language-specific, and that coherence is a concept best measured
against language-specific preferences.
This issue is discussed again in the next paper in the book. This kind of work
is important to further the understanding of what coherence is, and to
eventually obtain a testable theory about how coherence works in the fragment
Mary Carroll, Antje Rossdeutscher, Monique Lambert (''Subordination in narratives
and macro-structural planning - a comparative point of view'') present a
comparative study for English, French and German, where groups of 20 subjects
from each language are asked to describe a story portrayed in a nine minute long
silent animation movie. The protagonists of the story are animate and
non-animate, and for each language, various thousands of utterances were
produced and analyzed. Since the story is the same, this kind of controlled
experiment allows one to compare the narrative preferences for how the
information structure is construed across speakers from the different languages.
It is not clear, however, what was the experimental design used, and how the
subject pool is constituted. No mention is made about the age, sex, or
geographical location of the subjects of the experiment. All of these factors,
among others, potentially have an impact on the results that are obtained.
The results indicate that German speakers center the narrative sequence on a
temporal frame of reference, structured around temporal shifts, assign to the
main character a higher status than the one assigned to other agents in the
plot. Various strategies are used for this purpose, including downgrading and
passivization of structures where agents other than the main character is
referred. For the majority of English speakers, on the other hand, all agents
are recipients of the same status, and they can be referred to as subjects of
main clauses. The use of downgrading is very low, and the temporal frame is
deictic and narrator-oriented. Causality, rather temporal shifts, drive the
progress of the narrative. French speakers opt for a similar strategy in the
sense that all animate and non-animate agents are selected for mention and
mapped as subjects of clauses. However, non-intentional agents are commonly
downgraded. The alternation from the usage of main clauses and subordinate
clauses causes disruptions in the temporal chain of bounded events, which has
the effect that the narrative advances via causal relations rather than temporal
relations. The study is supplemented with a cross-linguistic quantitative
analysis, involving studies about other languages as well, such as Dutch,
Spanish and Italian. The differences between speakers of these languages are
found to be statistically significant when it comes to references to inanimate
entities as agents/experiencers. German and Dutch use less of such references
than all of the other languages.
One concern is that the study may be analyzing how people plan and produce plot
descriptions, rather than how people plan and produce complex narratives. This
might be what is going on in the case of English, where the subjects center the
perspective on the narrator (using expressions like ''then you see'' for example).
The subjects of the experiment were asked ''what happened'', which is ambiguous
between 'describe the story' or 'tell me the story', at least in English.
Carroll et al. also sketch a formal model for the kind of macro-planning that
the speakers from English, German and French employ in the construction of the
narrations. It is assumed that all the speakers assemble a knowledge base with
the events in the movie, and that this knowledge base is essentially the same
for all individuals in the experiment. The process of narrative selection and
production, it is argued, also depends on grammatical aspects that are
language-specific. In particular, it is argued in a plausible manner that verb
order and that the presence of absence of aspectual distinctions have and effect
on the subject's narratives.
PART III. Monolingual Studies
Anke Holler (''German dependent clauses from a constraint-based perspective'')
focuses on five known kinds of non-canonically linked clauses from German
(weil-verb second clauses, continuative wh-clauses, verb second relative
clauses, free dass-clauses, and dependant verb second clauses). Holler argues
that even though all of these can be seen as subordinate, they exhibit different
flavors of dependency, which in turn indicates that subordination should be
treated as a multidimentional phenomenon. Holler provides solid arguments
against the mainstream characterization of German subordinate clauses, and shows
that the position of the finite verb is not in general a good indicator of a
given clause being subordinate or not. Rather, Holler proposes that there are
several kinds of subordination hinging on various different linguistic
properties. These have to do with phonological properties, information
structure, syntactic function, binding, scope of negation, word order freedom,
and illocutionary force. A very sharp and clear analysis is provided, in which
these levels conspire together in such a way so that the gradient dependency
patterns emerge as a general and integrated result.
Holler's proposal seems to be on the right track and to flesh out the correct
In section 4, Holler briefly mentions previous and alternative views to the
typology that is provided in the paper, but is quick to dismiss them without
much discussion. The justification is based on the author's belief that none of
the previous accounts offers a satisfactory solution for the described phenomena.
That may be the case, but any reader would definitely be interested in some more
information about the alleged shortcomings of previous research. Holler provides
a rather straightforward partition-based account in Head-Driven Phrase Structure
Grammar, in which the typology that is discussed in informal terms is now
enriched with a number of grammatical constraints that capture the different
aspects of the five kinds of German clauses under discussion.
Although I agree in general terms with Holler's approach and discussion, there
are two minor points about the formal account, which are in a way, of completely
opposite natures. First, some of the typological partitions that are proposed do
not seem to capture a meaningful generalization from the grammar's point of
view. Basically, two types of dependent clause are assumed: integrated and
non-integrated. Each of these two types branches out in finer-grained clause
distinctions, but no general grammar constraints seem to operate over the
integrated clause type, nor over the non-integrated clause type. In other words,
the distinction is linguistically intuitive, but operationally not very useful.
If this is so, then the hierarchy of dependent clauses that is adopted can be
made simpler (only containing the leaf types). Note, however, that this matter
does not create any problem for Holler's claim that subordination is
multidimentional, nor for the various kinds of phrasal dependency that are
identified. Both the pre-theoretical claims and the formal analysis would remain
mutually consistent. Second, there are identical constraints repeated across
different kinds of subordinate clause. For example, 'weakly-integrated' clauses,
'weakly-non-integrated' clauses, and 'fully-non-integrated' clauses share the
same kind of assertional force. One is therefore tempted to insert a common
supertype to these three clause types and capture their assertional force in a
general way, rather than repeating it across three different leaf-types. So, it
seems that the hierarchy that Holler presents has some structure that the
grammar does not really need on the one hand, and that there are some
generalizations that could be captured if certain additional structure was
adopted instead. If the goal is to capture meaningful generalizations and to
keep the grammar as parsimonious as possible, then it could be that there is
room for improving this grammar fragment.
Maria Averintsva-Klish (''Right dislocation vs. afterthought'') focuses on a
phenomenon occurring in German on the sentential right periphery, and argues
that - contrary to the common assumption - there are many good reasons to
distinguish between two kinds of construction involving different discourse
functions. One is dubbed 'right dislocation proper' (RD) and another is dubbed
afterthought (AT). RD is seen as a forward-looking discourse strategy which
conditions the possible discourse continuations, whereas AT is a local strategy
which does not have an impact in the following discourse segment, nor on the
global discourse structure. Averintsva-Klish provides various good arguments for
RD and AT being very different. While RD is prosodically integrated in the host
sentence, AD always yields an intonational phrase. It is also argued that this
suggests that AP is a syntactically independent (orphan) structure, not attached
at all. To motivate this, a parallel is drawn with absolute constructions in
English (e.g. ''With his x-ray vision, John located the files''). Like AD, such
PPs can occur in a number of positions in the matrix, and are always
This may be, but prosodic independence does not mean syntactic independence.
Many dependent phrases are systematically realized as prosodically independent,
one example being topicalized structures. The moved element is clearly dependent
on an embedded predicate, and yet the topical position forces it to be
prosodically independent. In sum, the distinction between AT and RD is
well-motivated, but the interpretation of the syntactic status of AD as
comparable to the one of PPs in absolute constructions is not watertight.
Other differences pointed out by Averintsva-Klish have to do with morphological
agreement. There is strict agreement between a clause-internal pro-form and the
right-peripheral RD phrase, whereas in the case of AT agreement seems is of a
more optional nature. The two constructions also differ with regard to the
insertions and optional additions they allow for. The paper also discusses the
same phenomena in French and Russian, adding further support to the different
discourse functions that underlie the AT and RD distinction.
Laurence Delort (''Exploring the role of clause subordination in discourse
structure - the case of French 'avant que''') focuses on the French expression
'avant que' (before), and its discourse relational aspects. Delort argues that
this connective may convey various discourse relations, depending on
plausibility and/or the sentence-internal/external context. The main claim of
this study is that clause subordination can affect temporal structure as well as
discourse structure. Given a clause structure [X avant que Y] where Y is
subordinate, Delort notes that it is impossible to replace Y by a temporal
adverbial, in certain contexts. Using English translations of the French data,
whereas (1a) and (1b) are both acceptable, (2b) is not.
1) a. Paul found the solution before Marie gave it to him. b. Paul found the
solution before nightfall.
2) a. Paul sought a solution for a long time before Marie gave it to him. b.
#Paul sought a solution for a long time before nightfall.
From this observation, Delort concludes that the temporal subordinate looses its
adverbial function in some discourse contexts.
Overall, I hold a favorable view of this paper, but several alternative
explanations can be raised, not considered by the author. First, the sentence
''Paul sought a solution for a long time'' is open-ended with regard to the amount
of time Paul spent seeking a solution. It is conceivable that speakers are
biased to assume that Paul's quest for a solution lasted at least a few days,
weeks or even years. If so, the subordinate clause imposes an unlikely
limitation to a typically open-ended seeking-a-solution event: by midnight.
Unexpected limitations such as these often cause surprise and degrade
acceptability ratings. For example, this is what occurs in ''Fred drove in the
highway for 2 inches''. Although the sentence is grammatical and felicitous, it
feels strange because world knowledge informs us that people can't usually do
such a thing. Conversely, ''Fred drove in the highway for 2 miles'' does not clash
with world knowledge. The oddness disappears once we embed the strange sentence
in an adequate way, which shows that there is nothing wrong semantically: ''It is
unmanly impossible that Fred drove in the highway for 2 inches''. It seems to me
that in order to rule out this and other alternatives, Delort would have to
consider various sentences that make the subordinate phrase less unexpected,
with the aid of a controlled study with native informants. For example, it could
be that ''Paul sought his wallet for a long time before nightfall'' is fine to
French speakers, or that ''Paul sought a solution for two hours before nightfall''
is fine, in which case Delort's claims are problematic. One is left wondering if
there could be a simple account for the fact that the mere presence of 'seeking'
verbs and temporal modifiers causing 'avant que' to be unable to have an
adverbial function. It could be that such interactions are neither a matter of
syntax nor of semantics, but of extra-linguistic expectations that interfere in
sentence interpretation. In sum, the central data of this paper deserve a more
exhaustive scrutiny. It would also be worthwhile to question if the subordinate
expression really does have an adverbial function. Note that ''nightfall'' is
nominal, not adverbial. One can just as easily replace it with deverbal nouns,
e.g. ''Paul found the solution before the explosion''. One can go further and
suggest that these phrases are coerced into verbal structures (roughly
paraphrasable as ''Paul found the solution before the explosion took place'' /
''Paul found the solution before nightfall took place''). The generalization would
be that the complement of the expression 'before' denotes an event, which
'before' locates in time with regard to the matrix. All in all, it would have
been useful to have a discussion about alternative interpretations of the French
In this work three different kinds of readings are identified and defined as
follows. A 'circumstance' interpretation in which the subordinate clause has a
temporal adverbial function; a 'continuation' interpretation where the
subordinate clause is a continuation of the matrix; and a 'pre-condition'
interpretation where the realization of the matrix is necessary for the
realization of the subordinate. In all of these, the temporal relation of 'avant
que' is the same: temporal precedence. Delort argues that discourse context and
extra-linguistic knowledge is responsible for allowing 'avant que' to appear in
non-circumstance discourses. The difference between the latter two are
well-motivated, and various temporal constraints are discussed. Delort concludes
that - in Segmented Discourse Representation Theory terms - 'avant que' can (at
least) trigger a Brackground relation (implying temporal overlap and
subordination in discourse) and can alternatively trigger a Narration relation
(implying temporal precedence and coordination in discourse).
Michael Franke (''Pseudo-imperatives and other cases of conditional conjunction
and conjunctive disjunction'') investigates a pragmatic discrepancy between
pseudo-imperatives realized via disjunction and conditional clauses. The puzzle
consists in the following. Although all other combinations are possible, it is
not possible to have an imperative disjoined with a positive clause like (1a)
1) a. ?Close the window or I will kiss you. [A or P+] b. Close the window or I
will kill you. [A or P-]
while it is perfectly possible to have the conditional counterparts:
2) a. If you don't close the window, I will kiss you. b. If you don't close the
window, I will kill you.
Franke assumes that (1a) is ruled out by discourse-based factors specific to
I disagree with the interpretation of the data. In my view, (1a) is as good as
(2a). If A and P+ are given a more direct causality nexus, the oddness vanishes:
3) a. Button your blouse or I will kiss you. [A or P+] b. Restrain me or I will
kiss you. [A or P+] c. Kiss me or I will kiss you. [A or P+]
There are simpler alternative explanations for these patterns, which are not
considered by the author. There are various factors that contribute to (1a)
being slightly harder to accept than (3) and (2a). First, the conditional clause
[if X then Y] makes explicit the conditional meaning, whereas [X or Y] can be
interpreted in many ways. Right off the bat, this makes (1a) harder to
understand. If a speaker attempts to interpret (1a) out of the blue and tries
the conditional meaning, then one is pushed into a situation where someone has
an uncontrollable urge to kiss whenever a window is open. Although such
pathologies exist, these are very unlike scenarios, and thus difficult to accept
without proper contextualization. Still, if such a context is explicitly
provided, the purported oddness may disappear. Imagine the following dialog
between two lovers:
Juliet - I don't want our relationship to be merely platonic anymore.
Romeo - Why? It's nice to climb to your window every night, and chat with you.
Juliet - I can't take this anymore. Close the window, or I will kiss you.
I believe the purported oddness of (1a) is a matter of plausibility and
processing, rather than a matter of the semantics/pragmatics of the connector
'or'. One can manipulate context and implied meaning to obtain felicitous
grammatical examples like the one below.
- I have a cold and I'm full of germs. Go get me some tea or I will kiss you!
Ingo Reich (''From discourse to 'odd coordinations' - On Asymmetric Coordination
and Subject Gaps in German'') addresses two German coordinate constructions known
as Subject Gaps in Coordinate Structures (SLFC) and Asymmetric Coordination
(AC), in the sense of Höle (1990). Reich argues that the two kinds of phenomena
are essentially the same, sharing both syntactic, semantic, and pragmatic
properties. The paper shows that SLFC cannot be accounted for via peripheral
ellipsis or ATB-movement, and - drawing from Höle (1990) and others - shows that
SLFC and AC actually share various properties, the most notable being verb
fronting in non-initial conjuncts, a 'one-event' interpretation (which differs
from the interpretation of coordination in symmetric structures), and the
impossibility of ATB-movement. In spite of this, Reich's analysis will involve
ellipsis. Reich proposes that the proper account of the 'one-event' semantics of
these structures involves an event subordination relation. Thus, two verbal
conjuncts S1 and S2 respectively introducing events e1 and e2 yield a complex
coordinate structure where e2 is a sub-event of e1: e2 < e1.
This claim is hard to accept because asymmetric coordinations in general, as in
''he saw us and reported us'' for example, do not seem to work that way. The event
of reporting is not a sub-event of seeing, in any intuitive way. Reich dismisses
an alternative view where e1 and e2 would be 'merged' by stating that it is
unclear how to define such a relation. However, event 'merging' (or more
commonly, 'fusion') has been discussed and formalized in mereological terms in
Bach (1986) and Link (1998:240), among others.
Reich never defines ''<'' but I suspect this would require additional relations
because ''<'' is probably transitive. Consider for example a structure with three
conjuncts. One would like to have not only e3 < e2 and e2 < e, but also to be
true that e3 < e.
Reich's syntactic account is formalized in terms of the Minimalist Program. Verb
movement in German is triggered by a (uninterpretable) strong syntactic feature
[F] that is the head of a functional projection and that must be checked by a
finite verb in overt syntax. Verb fronting in AC and SLFC is captured in terms
of a new feature [OCC] (which denotes the event subordination relations) that
also projects a functional projection and selects for [F]. Reich goes on to
describe in detail a number of other syntactic steps that lead to the two
possible structures under discussion, although none of these is defined in
formal terms, and thus, the validity of the account is difficult to access. The
fact that the subject is missing in SLFC structures is claimed to follow from a
special ellipsis operation that arises from inference-driven redundancy.
Basically, the claim is that speakers assume that given two events e2 < e, it
must be the case that the subject of event e is also the subject of event e2.
Reich notes that this entailment is not valid, but assumes that it exists
nonetheless, based on the fact that sentential entailments in natural language
are often different from the ones in mathematical logic.
The account then proceeds to specify in what conditions the subject can be
omitted. At this point, the constraints target the SLFC construction explicitly:
''in a SLFC the head [v] of the second conjunct lacks [SUBJ]'' (298).
In Minimalism terms (where usually there are no 'constructions' per se), it is
not made clear how a grammar restriction can mention a phrasal construction
during a derivation. In other words, it is not clear how Computation can
identify a structure as being or not being a SLFC construction.
PART IV. Diachronic perspectives
Rosemarie Lühr (''Old Indic clauses between subordination and coordination'')
examines the distinction between coordination and subordination in Old Indic.
This leads the study to consider the interplay between prosody and information
structure. The main verb of an Old Indic clause is usually unaccented. In the
presence of a prefix, stress will be realized on the latter. But verbs do bear
accent in subordinate clauses introduced by complementizers, relative pronouns,
and in other clauses with the same prosodic properties as subordinates. The main
goal is, then, to determine whether the latter are subordinate or asyndetic
coordinate main sentences. Lühr claims that these structures are in fact
coordinate, and that the stress pattern is there to identify the coordinate
structure, rather than being a necessary correlate of subordination.
Lühr's evidence for a coordination analysis is convincing, for example, as data
involving four conjuncts is provided. However, since there are many syntactic
traits that are specific to coordination, one would expect that more syntactic
arguments in favor of coordination could be found and used.
This work also shows that these structures are compatible with a variety of
different stress patterns that have to do with information structure import,
rather than syntactic status. Lühr also proposes that even though some of these
structures have the same verb accentuation that is observed in subordination,
they are in fact the Old Indic counterpart of German coordinate structures
showing the rise-fall contour that is observed in contrastive coordination (as
observed in Gapping, for example). Construction-specific evidence can also be
found. Bisyndetic coordination patterns (e.g. ''both ... and ...'' or ''neither ...
nor ...'') exhibit an exceptional obligatory stress pattern when the verbs in
each conjunct are identical. This is unexpected for Lühr's account, but it is
argued that bisyndetic coordination needs no prosodic cue to signal the
coordinate structure, since this is overtly marked by the double coordinators.
The paper does not provide a formulation of the mechanisms that regulate the
prosodic- and construction-dependent stress assignment patterns, but provides an
important first step for such a goal.
Svetlana Petrova & Michael Solf (''Rhetorical relations and verb placement in the
early Germanic languages - A cross-linguistic study'') present a diachronic study
about the distinction between coordination and subordination in discourse,
focusing on Old High German and on other early Germanic languages. It is known
that the placement of the inflected verb in early Germanic languages hinges on
factors pertaining to information structure and discourse organization. For
example, V2 structures with referential elements preceding the verb usually
correspond to a supportive/explanatory import, and thus can be considered
subordinate in discourse structure. Conversely, V1 (as well as cases where a
particle or an adverbial precedes the verb) correspond to sentences that are not
subordinate and link to the main story-line. In terms of discourse structure,
this is best viewed as coordination. Petrova and Solf consider other kinds of
data, mostly from declaratives, in support of the claim that verb placement
serves certain discourse functions in early Germanic languages. The data
provided are consistent with the view that the position of the finite verb in
Old High German is sensitive to the information structure, depending on the
pragmatic status of the nominal elements in the sentence. Verb fronting seems to
have a clear functional purpose, as it is used to mark episode boundaries in Old
High German. The study goes further in identifying some correlations between
verb placement and discourse-structuring phenomenon in Old English, Old Saxon,
and Old Norse, with similar discourse-structuring functions.
The paper is an important contribution, but no attempt is made to produce a
theoretical account of the phenomena under discussion. Petrova and Solf mention
the adoption of Segmented Discourse Representation Theory, but no actual
proposal for the representation of the information-structure of the
constructions and discourse linking under discussion is put forth.
Much has been said and written about Coordination and Subordination, but many of
the challenges that these topics raise for linguistics endure to this day. This
makes an important contribution to this discussion in the form of thirteen
papers. The studies in this book are generally of a high standard. Some of these
contributions are more theoretical, and others are more empirical. This volume
should be a useful addition for those who are interested in discourse, syntax,
and morphology alike, for it offers a wealth of cross-linguistic and
cross-representational information with regard to subordination and
coordination. Virtually all the contributions stress that the phenomena probably
hinge on the interplay of various factors, ranging from syntax to information
structure, which offers a valuable and rich perspective in itself. Some papers
focus mainly on syntax and semantics, and others focus on discourse/pragmatics,
and no paper actually resolves the question of whether the two levels of
structure resort to essentially the same kind of coordination/subordination
From the evidence and discussion that is provided, I find myself leaning towards
a view where there are major differences between coordination/subordination in
syntax and the clause-linking mechanisms that are observed in discourse
structure. I can recommend this book to researchers and students of pragmatics
and syntax. For the most part the contributions are accessible to readers who
are less familiar with formal theories, even though some proposals resort to
experimental methods and formal grammar accounts. If anything, one is left
wishing that the accounts were equally explicit and formalized, but scientific
progress is a slow and tentative process of data gathering and theorization.
This books certainly achieves this goal.
Bach, E. (1986). The algebra of events. _Linguistics and Philosophy_ 9, 5–16.
Goldsmith, John. (1985). A Principled Exception to the Coordinate Structure
Constraint. In William Eilfort, Paul Kroeber, and Karen Peterson (eds), _Papers
from the 21st Meeting of the Chicago Linguistics Society_.
Höle, T. (1990) ''Assumptions about asymmetric coordination in German''. In
_Grammar in Progress: Glow Essays for Henk van Riemsdijk_, J. Mascaró & M.
Nespor (eds), 329 - 340, Tuebingen: Niemeyer.
Lakoff, G. (1986). Frame Semantic Control of the Coordinate Structure
Constraint. In _Papers from the 22nd Regional Meeting of the Chicago Linguistic
Society_. Chicago: Chicago Linguistic Society.
Levin, N. and E. Prince (1986). Gapping and clausal implicature. _Papers in
Linguistics_ 19, 351–364.
Link, G. 1998. Algebraic Semantics in Language and Philosophy. hCSLI Lecture
Notes_ No. 74.
Kehler, A. (2002). _Coherence, Reference, and the Theory of Grammar_. Stanford:
McCawley, J. D. (1988). _The Syntactic Phenomena of English_ (second ed.).
Chicago: University of Chicago Press.
Huddleston R. D. and G. K. Pullum (2002), _The Cambridge Grammar of the English
Language_. Cambridge University Press.
Ross, J. (1967). Constraints on Variables in Syntax. Doctoral dissertation, MIT,
Cambridge, Massachusetts. [Published in 1986 as _Infinite Syntax!_ Norwood, NJ:
Schmerling, S. (1972). Apparent counterexamples to the coordinate structure
constraint: A canonical constraint. _Studies in the Linguistic Sciences_ 2 (1),
Schmerling, S. (1975). Asymmetric conjunction and rules of conversation. In P.
Cole and J. L. Morgan (Eds.), _Syntax and Semantics, Volume 3: Speech Acts_, pp.
211–231. Academic Press, New York.
ABOUT THE REVIEWER
Rui P. Chaves is Assistant Professor at the University at Buffalo, SUNY. His
current research activities concern syntactic theory, and the syntax-semantics