Anagnostopoulou, Elena, Henk van Riemsdijk, and Frans Zwarts (eds). 1997:

Materials on Left Dislocation

(Linguistik Aktuell / Linguistics Today 14).
Amsterdam: John Benjamins. Pp. viii + 349.
ISBN: 1-55619-233-9. Hfl 145.- or $ 86 (hb)

reviewed by

Nils Langer

School of Modern Languages, University of Newcastle upon Tyne, NE1 7RU
E-mail: Download this review ( 27K, Rich Text Format)]
Copyright Notice:

First published in Web Journal of Modern Language Linguistics in association with the publishers (to be announced). © 1997 Nils Langer.
The moral rights of the author(s) to be identified as author(s) of this work are asserted in accordance with §§.77 and 78 of the Copyright, Designs and Patents Act 1988. This work may be reproduced without the consent of the author, in part or in whole in any manner and in any medium subject only to the two following conditions:
(a) no charge shall be made for the copy containing the work or the excerpt,
(b) a copy of this notice shall precede the work or the excerpt.

The Book

"The facts, to state it bluntly, are highly subtle and often murky" (Vat, [:70]). This is certainly a good description of what one feels when exposed for the first time to data of Left Dislocation and analyses that attempt to explain this data. However, the present book, in presenting this collection of eleven papers and a short introduction, provides not only a substantial starting point for those who have not worked on LD before but it also leads the reader to the insights of the most recent research.

Because no book had previously been published that dedicated itself entirely to the subject of Left Dislocation [: v], Materials on Left Dislocation has been designed to be more than just a conference proceedings. Although five of the eleven collected papers are revised versions of papers presented at the Tilburg Workshop on Left Dislocation in 1994, five[footnote 1] more "historical" articles, written between 1974 and 1983 have also been included in order to give the reader access to some of "often quite inaccessible texts" [: v] on the subject, only one (Hirschbühler, 1975) of which had previously been published in a major journal.

The early papers are:

Riemsdijk, H. van: Left Dislocation (introduction), pp 1-12; Riemsdijk, H. van & Frans Zwarts (1974, unpubl.): Left Dislocation in Dutch and Status of Copying Rules, pp 13-30; Rodman, R. (1974, prev. publ. in Papers in Linguistics 7): On Left Dislocation, pp 31-54; Hirschbühler, P. (1975, prev. publ. in Linguistic Inquiry 6): On the Source of Lefthand NPs in French, pp 55-66; Vat, J. [footnote 2] (1981, prev. publ. in GAGL 20): Left Dislocation, Connectedness and Reconstruction, pp 67-92; Cinque, G. (1983, prev. publ. in Ehlich & Riemsdijk: Connectedness in Sentence Text and Discourse): 'Topic' Constructions in some European Languages and 'Connectedness', pp 93-118; Zaenen, A. ("early '80s" [: 146], unpubl.): Contrastive Dislocation in Dutch and Icelandic, pp 119-148;

and the recent papers are:

Anagnostopoulou, E.: Clitic Left Dislocation and Contrastive Left Dislocation, pp 151-192; Demirdache, H.: Dislocation, Resumption and Weakest Crossover, pp 193-232; Escobar, L.: Clitic Left Dislocation and other Relatives, pp 233-274; Hoof, H. van: Left Dislocation and Split Topicalization in Brabant Dutch, pp 275-306; Wiltschko, M.: Parasitic Operators in German Left-Dislocation, pp 307-339.

All the papers deal with the phenomenon of Left Dislocation within the mainstream generative framework, thus, depending on the time of publication, EST, GB and Minimalism. The exceptions to this are Zaenen's paper, couched in the framework of LFG and Vat's paper including an analysis using Riemsdijk & Williams (1980)'s L-model. The languages discussed circle around the Germanic and Romance family, though a major discussion of Modern Greek is provided by Anagnostopoulou while Demirdache discusses mainly Hebrew and Egyptian data.

The Phenomenon

The term Left Dislocation (LD), first properly described in Ross (1967/1986), is generally used to refer to a special type of constructions where the left most constituent of a sentence is co-indexed with a phrase inside the clause as in

(1) The king, I haven't seen him for some time. (HT)LD
In (1), the king can, pre-theoretically, be seen as a copy of the pronoun him; both phrases refer to the same entity, and, grammatically, both appear to function as the direct object of seen. This contrasts with the construction of Topicalization where no pronoun (or else) appears inside the root clause as in

(2) The king, I haven't seen ____ for some time. Topicalization
An initial contrast is that the gap in (2) suggests that the topicalized NP has moved to the sentence-initial position. The story looks more complicated with regard to (1). Although Ross (1967/1986) suggested that both in LD and Topicalization, the left-most NP has moved to its landing site, most researchers (including those in the book) assume that in LD-constructions, the left-dislocated NP is base-generated in its "surface" position (Riemsdijk & Zwarts, Rodman, Hirschbühler, etc). The problem that immediately arises with this line of argumentation is, of course, to what extent structures like (1) are part of sentence grammar, or whether it would not make more intuitive and technical sense to assume LDs of the type in (1) to be pragmatic phenomena. Just to point to one problem: If the king is part of the root clause in (1), then it requires case and a theta role, which, presumably, it will receive from the lexical verb seen. [footnote 3] However, the case- and theta-assigning properties of seen are fully satisfied by the presence of the object pronoun him. Even though him and the king are coreferential and receive the same case, an intial observation of the example in (1) would make one suspect at least a violation of the theta-criterion. One could perhaps conceive of a technical circumvention around the problem by assuming that the king and him are base-generated under the object-NP node simultaneously but it would then be a construction specific exception (if proven to be technically possible) and hence theoretically undesirable. Cinque [: 99] does indeed conclude that LD of the type illustrated in (1) is a "discourse grammar construction."

The issue is further complicated by the fact that there are at least three types of LD, all of which are extensively discussed in the present book, both intra- and crosslinguistically.

The type found in English, i.e. Hanging Topic Left Dislocation (HTLD) has already been exemplified in (1). Here, there is a 'normal' pronoun inside the root clause, co-indexed with the LD-ed phrase.

In German and Dutch, Contrastive Left Dislocation (CLD) is fairly common, where the there is a d-pronoun[footnote 4] right-adjacent to the left dislocated phrase as in

(2) Den Kuchen, den habe ich schon lange nicht mehr gesehen
the-ACC cake, the-ACC have I already long not anymore seen
"The cake, I haven't seen for a long time"
The third type of LD is Clitic Left Dislocation (CLLD) which is found, e.g. in Italian, Spanish and Greek

(3) A tuo fratello, non gli hanno ancora dato il visto (Cinque, [: 94])
to your brother, not him-CL have-3s yet given the visa
"Your brother, they haven't yet given him the visa"
Unlike in the two other constructions in (1,2), the syntactic position of the remnant inside the root clause is not fixed but depends on the particular syntax of clitics in the respective languages. The question, obviously, emerges as whether the three types of constructions shown above can and should be unified under the same term (i.e. left dislocation). The similarities and differences can be narrowed down by answering more particular questions such as

The Articles

The book presents answers and analyses to all of these questions. I will now provide very short summaries of the main theses of the articles in the book.

The early papers in Part I of the book introduce a lot of data, most of which remains crucial to the discussions throughout the book. R&Z, Rodman and Hirschbühler use their examples primarily to defend the position that the HTLD is not derived from movement, contra Ross (1967/86) who had claimed that LD is the manifestation of a "copying rule" and thereby on a par with Topicalization. R&Z go as far as to desire the elimination of copying rules from the theory of grammar in principle, showing in particular that LD (HTLD) violates a number of movement rules and constraints (Topicalization, adverb preposing, A over A principle, etc). Instead they claim that LD (HTLD) is a sentence satellite [: 24] like vocatives, exclamations and OM-sentences.[footnote 6]

A similar line is pursued by Hirschbühler who provides a substantial amount of French data showing that LD in French would violate several constraints on movement, e.g. epithets, Complex Noun Phrase Constraint (CNPC), anaphors, embedded sentences etc, yielding the conclusion that the LD-ed NP has to be base-generated as the sister of S(entence) and cannot have moved there.

Rodman concentrates on showing that "LD is not a pronoun-leaving version of Topicalization" [:32]. Data such as

(4a) The beans, George ate three / most / the rest of them. (HT)LD
(4b) * The beans, George ate three/ most / the rest of . Topicalization
shows that one cannot analyse HTLD as simply a Topicalization structure where the object is overtly spelled out. In addition, Rodman shows that (HT)LD and Topicalization differ semantically in that topicalized XPs can never introduce new information into the discourse [:33] whereas LD-ed phrases can.

Vat uses data from idioms, reflexives, crossover effects, scope phenomena and case connectedness, a.o. to show that Contrastive Left Dislocation[footnote 7] (cf. 2) (and CLLD) contrasts with HTLD, in that the latter involves base-generation of the LD-ed phrase as opposed to CLD where, following Vergnaud's analysis of relative clauses, both

"the element in dislocated position and the d-pronoun are generated as sisters in an argument position..., together they are moved into COMP by WH-movement. Next,...[the LD-ed element] is moved to the position under S" [i.e Spec, CP2], whereas the d-pronoun remains in COMP." [: 83]

Base-generating two elements in the same argument position will of course cause problems with conforming to the theta-criterion and case-assignment properties (cf Wiltschko, addressing that problem). Vat assumes that case is uniquely assigned to the complex trace "since neither of them can be marked as the trace's head." [:87]. Using evidence mainly from Italian, Cinque compares HTLD, Clitic Left Dislocation (CLLD) and Topicalization, concluding that whereas the former is a discourse grammar phenomenon, the latter two constructions are part of sentence grammar in that they involve movement and chain formation. CLLD distinguishes itself from HTLD in a number of point some of which being that the LD-ed phrase is not categorially restricted to NPs, that multiple LD is possible, that CLLD is possible in both root and embedded clauses, that case connectedness is obligatory [: 96].

Whereas the tonic pronoun in HTLD structures is simply a "normal" object pronoun filling an argument position, the clitic(s) in CLLD land(s) in an A'-position, forming part of a chain. The argument properties are thus instantiated on the syntactic chain, not on any one position. This proposal resolves familiar reconstruction problems though it remains awkward to suggest a chain with two overt members. Presumably this is simply a language particular property of Italian (and all other languages licensing CLLD).

The similarity between Topicalization and non-HTLD left dislocations is also alluded to in Zaenen's article who argues that Contrastive Dislocation (CLD)[footnote 8] is a type of topicalization (cf (2)), not (HT)LD, in that it is to be analysed as the "syntactic[..] linking [of] the initial an argument position within the sentence and not... as a left dislocated constituent followed by a sentence in which a pronoun has been topicalized." [: 119]. Using data from Dutch and Icelandic, Zaenen develops an analysis within the LFG framework in which the contrast between (HT)LD and Topicalization derives from the fact that in (HT)LD "the initial constituent is only anaphorically bound to a constituent within the sentence" [:120] whereas in Topicalization (and thus also in CLD), "the initial constituent is linked to a function in the lower clause and hence all connectivity phenomena that could hold between this constituent and the rest of the sentence will also hold between the sentence-initial constituent and the rest of the sentence" [:120][footnote 9]. Whereas the difference between Topicalization / CLD and (HT)LD is thus of a syntactic nature, the difference between Topicalization and CLD "seems to be that the latter can only be used felicitously when the speaker has a 'recoverable' referent in mind for the initial constituent...[(]specificity requirement [)]" [:142]

A serious claim for the uniformity of CLD and CLLD is presented in Anagnostopoulou's article where she postulates that the differences between CLD and CLLD fall out naturally from (a) the difference in landing sites of the left-dislocated phrase (adjunction to CP (CLD) and IP (CLLD) resp.) and from the independently motivated syntactic differences between the respective remnants: XPs (XMAX) (the status of the remnant in CLD) have different syntactic properties from clitics (XMAX and XMINsimultaneously) (as found in CLLD). The uniform format is given in

(5) [FP XPi(LD-ed) [FP....Opi....ti...] [: 158, after Demirdache 1991]
Both CLD and CLLD have a gap inside the root clause, both are unbounded, sensitive to islands and connectivity and both have an overt mediation between the fronted XP and the gap, namely a d-pronoun or clitic respectively. This overt mediation functions as a resumptive operator which is licensed in different syntactic positions crosslinguistically. Cinque's paradox, namely that CLLD is insensitive to wh-islands, is resolved by arguing that the clitic undergoes long XP-movement at LF.

Covert LF-movement also features in Demirdache's analysis of Hebrew. It is proposed that dislocation structures with gaps created by movement and dislocation structures with resumptive pronouns have "the same LF: an A'-trace occupies the variable site." [: 193]. However, in order to capture the empirical differences, namely the sensitivity (or lack thereof) to island and Weak Crossover (WCO) effects, Demirdache suggests that the variation exists as to the level at which the A'-trace is created: at S-structure in gap-constructions and at LF when a resumptive pronoun (which is the in-situ counterpart of a null operator) is present.

Escobar compares Spanish CLLD with HTLD, F(ocus)-movement and two types of relative clauses and concludes that "CLLD is an a par with [appositive] Relative Clauses." [: 253]. Having established that CLLD is different from HTLD (sensitive to islands, cannot co-occur with F-movement, etc) and F-movement (lack of subject inversion effect, sentence topic and clitic both must be specific, etc), Escobar moves on to discuss the properties of restrictive and appositive relative clauses with regard to resumptive clitics, doubly filled COMP, strong quantifiers, Weak Crossover, a.o. Establishing that "CLLD shares most properties with Appositives" [: 266], Escobar suggests the following structures:

(6a) [CP A Juan [TP ei [AgrSP yo [AgrS [AgrOP ei [AgrO lo [VP conozco ti ]]]]]]] CLLD
(6b) [CP A Juan [TP ei [AgrSP yo [AgrS [AgrOP ei [AgrO lo [VP conozco ti ]]]]]]] AppRel
Juan / whom I him know-1s
Unifying Appositives and CLLD, Escobar concludes that a clitic is the "minimal copy in the chain headed by the sentence topic, a "specific" indefinite, or a strong quantifier [or a relative pronoun]" [: 270].

Split Topicalization (ST) and Split Left Dislocation (SCLD) in Brabant Dutch, the subject of van Hoof's paper, are illustrated in

(7a) Koeien heeft-ie een helehoop in de achterste wei (ST)
cows has-he a lot in the furthest meadow

(7b) Koeien die heeft-ie een helehoop in de achterste wei (SCLD)
cows those has he a lot in the furthest meadow [: 280]
"Cows, he has a lot of them in the furthest field"

The two constructions are split in the sense that only the noun koeien of the NP een helehoop (koeien) has moved (if a movement analysis is assumed) to the sentence initial position. The theoretical problem with ST lies with the fact that it displays properties of both movement (island sensitivity, V2 effect, connectedness etc) and non-movement (strong adjective inflections, determiner overlap, welche-pronoun) constructions. Van Hoof suggests that the left-dislocated element is base-generated in its surface position and that the empty position in the split NP is filled by a pro and that the d-pronoun has moved to its landing site from an independently generated position inside VP. In the second part of her paper, van Hoof argues that in order to unify (S)Topicalization and (S)CLD one has to distinguish between two prosodic variants of CLD, one of which is identical to (S)T whereas the other one is a "special case of HTLD with an additional contrastive interpretation effected by the d-pronoun in [Spec,CP]." [:299].

Wiltschko argues that at least in German, (C)LD is not in contrast to topicalization but rather a subset thereof. In particular, the LD-ed constituent is a syntactic operator P(ARASITIC)Op that is parasitic on, i.e. shares the same trace as the topicalised XP. Crucially, the LD-ed XP is base-generated in its surface position, so movement constraints apply only to the topicalised d-pronoun. Syntactic A'-chains hold between the topic and the trace and the LD-ed XP and the trace respectively, but not between the LD-ed XP and the topic, as illustrated in

(8) DPi [CP Proni.......ti POperator TOperator vbl
Feature sharing is guaranteed by the chain POpi - ti, making the feature identity between the topic and the LD-ed a mere, though unavoidable, by-product. Immediate problems that arise with regard to the theta-criterion and case-assignment are assumed not to arise because "there is just one over case- and theta-marked element per chain." [:319]; this does not explain, however, how the verb can assign two identical theta-roles as opposed to one in non-topicalised structures.

The Editing

In general, the editing quality of this book is quite satisfactory and the high quality of the content of the papers make it easy for the reader to overlook the existing spelling errors. However, a few comments might be allowed about the general format and lay-out of the book. Although the existence of an index is very welcome, the inclusion of abstracts introducing the individual papers could have still improved the book.

Furthermore, though this might be just a matter of the reviewer's personal taste, the use of end- rather than footnotes should have been avoided. A point, however, which is quite important is that often, especially in the early papers, the general format for examples, i.e. "foreign original - word for word gloss - English translation", has often not been adhered to, to the effect that some of the articles can only be truly understood, the reviewer feels, if the reader understands most of the examples in the original language, e.g. in cases where the reader is supposed to compare left-dislocated idiomatic Dutch and their topicalized counterparts, which can be very difficult when only a word for word gloss is provided. There is quite a editorial discrepancy in general between the papers in part I and II. The reviewer welcomes very much the editors decision to make available the older papers but is somewhat disappointed by the reluctance to update the papers so that all examples are provided with a translation and that the references are changed so that quoted manuscripts, which have subsequently been published, now appear with the publishing information, not just as Ms. These criticisms seem a little pedantic but one would expect that re-publishing old papers would involve a little more than the mere re-printing thereof.

The Conclusion

As I said at the beginning of this review, this appropriately titled book provides an excellent introduction to the topic of Left Dislocation in Generative Syntax. Including the early papers enables the reader to understand the problems involved with this construction from very early on. Understanding the initial problems, in turn, makes the later papers all the more accessible and after reading the book one feels that one has had a substantial insight into the intricate problems that are part of a more than superficial analysis. The terminology is mostly uniform, there is plenty of data and the papers seem well selected, covering all major issues, most of which are controversial amongst the contributors. I can thoroughly recommend this book, although I am aware that the substantial price, as so often, may prevent some potential buyers from stocking it in their private library.


Demirdache, H. 1991. Resumptive Chains in Restrictive Relatives, Appositives and Dislocation Structures. PhD thesis. MIT.

Riemsdijk, H. van & E. Williams. 1980. "NP-structures". ms (cf Vat, [: 92])

Ross, J. R. 1967/86. Constraints on Variables in Syntax. PhD thesis. MIT. publ. as: Infinite Syntax ! . Norwood: ABLEX Publ.

Vergnaud, J.R. 1974. French Relative Clauses. PhD thesis. MIT.


[1]Demirdache's paper is from the 1990s but had not been presented at the workshop.[Return to text]

[2]A pseudonym for a collective of seven authors from Amsterdam University.[Return to text]

[3]Assuming for ease of exposition that verbs still assign Case (contra Minimalism).[Return to text]

[4]d-pronoun stands for demonstrative pronoun. However, a demonstrative in the traditional sense is rarely used in German CLD. Here, as can also be seen in example (2), a normal definite determiner, acting as a pronoun is used.[Return to text]

[5]Obviously, the differences cannot be merely crosslinguistic given that at least for German, two types of LD are grammatical (HTLD and CLD), and e.g. in Italian there is HTLD and CLLD (Cinque).[Return to text]

[6]It is nowhere explained in the book what OM-sentences are, though the following Dutch example is given

Nog een pilsje en ik ga er vandoor
'One more beer and I'm leaving' (R&Z, [: 24])
I mention this to illustrate that the reprinting of older articles should encourage the editors to explain or define unfamiliar terminology.[Return to text]

[7]Vat also discusses Dutch CLD where the d-pronoun remains in situ as in

(i) Die man, die ken ik niet (CLD)
That man, that-one know I not
"That man, I don't know"

(ii) ? Die man, ik ken die niet (HTLD with stranded d-pronoun)
That man, I know that-one not
(iii) ? Die man, ik ken hem niet (HTLD)
That man, I know him not
(Vat, [: 70])

though he prefers to analyze it as a type of HTLD. In more recent, i.e. Minimalist, terms one would perhaps be more inclined to compare (i) and (ii) given that they share identical numerations (at least of the overt elements).[Return to text]

[8]Zaenen refers to Contrastive Dislocation as CD [:119]; I will use CLD for uniformity of terminology.[Return to text]

[9]In this perhaps not all too clear formulation, it is crucial to note that initial constituent and sentence-initial constituent refer to different elements, namely the LD-ed phrase and the d-pronoun respectively.[Return to text]

Return to The Issue 2 Contents Page| The WJMLL Home Page