By HANS USZKOREIT. (CSLI Lecture Notes. Vol. 8.) Stanford: Center for the Study of Language and Information, 1987. Pp. ix, 190. Cloth $27.50, paper $12.95.
Reviewed by ERHARD W. HINRICHS and TSUNEKO NAKAZAWA, University of Illinois at Urbana-Champaign.
Taken in excerpts from: LANGUAGE, 65(1):141-149, 1989.
U pursues two main goals in this monograph, which is based on his 1984 University of Texas dissertation. The first goal is descriptive and is reflected in the title (henceforth abbreviated as WOCSG): how to adequately describe constraints on the order of constituents in German, a language with partially free word order. Even though we disagree with U on various aspects of his analysis, one cannot help but be impressed by the extensive list of syntactic phenomena which WOCSG addresses. Each phenomenon is illustrated by carefully chosen linguistic data and captured by an explicit set of syntactic rules which provide a benchmark for future work in the area of German syntax.
The second main goal of WOCSG is a theoretical one. U explores the question of whether the syntax of German can be accounted for in the theory of Generalized Phrase Structure Grammar (GPSG) as developed in Gazdar & Pullum 1982 and Gazdar et al. 1985. [...]
WOCSG does not simply offer an existence proof that an adequate analysis of German word order can, in fact, be given in the ID/LP format of GPSG. In order to account for the order of NPs in the so-called MITTELFELD 'middle field' of German clauses, WOCSGproposes a modification of the ID/LP format itself. Even though we remain unconvinced that the German data ultimately warrant the technical innovation that WOCSG proposes for the GPSG framework, U's careful discussion of the theoretical significance of the German word order data makes WOCSG an important work not only for the study of German syntax, but for the theory of GPSG itself.
The first six chapters of WOCSG contain a theory-neutral description of some basic facts about German word order (Ch. 1, 8-30); an extremely insightful introduction to the theory of GPSG (Ch. 2, 31-47); an account of the phrase structure of main and auxiliary verbs, the structural differences between main and subordinate clauses, and the structure of verb-second main clauses with topicalized initial constituents (Ch. 3, 48-80); an analysis of the morphosyntactic properties of separable prefixes (Ch. 4, 81-112); a treatment of the order of verbal complements and adjuncts in terms of syntactic markedness (Ch. 5, 113-27); and an informal discussion of additional syntactic constructions in German and how to incorporate them into the grammar fragment of WOCSG (Ch. 6, 128-60). Ch. 7 (161-63) offers some remarks for future research directions for the study of word order in GPSG. Finally, an Appendix (164-67) lists the set of phrase structure rules for the fragment of German syntax developed in Chs. 3-5. [...]
1. ID/LP FORMAT AND THE ECPO PROPERTY [...] At first glance, the word order paradigms in German subordinate and main clauses, which U summarizes in Ch. 1, seem to be expressible only by a set of phrase structure rules which do not exhibit the ECPO property.
(1) Fand Peter das Buch?
Found Peter the book
'Did Peter find the book?'
(2) Ich sah wie Peter das Buch fand.
I saw how Peter the book found.
'I saw how Peter found the book.'
Assuming that German has a flat structure and no VP constituent, as U argues in Ch. 3, the word order of constituents in yes/no questions, as in 1, and in subordinate clauses, as in 2, could be captured by the LP statements in 3 and 4 respectively.
(3) LP: V < X (4) LP: X < V
However, since 3 and 4 cannot be simultaneously satisfied, ECPO is violated. Unless further assumptions are made about the categories involved in the LP statements in 3 and 4, a phrase structure grammar for German subordinate and main clauses does not seem to be expressible in the ID/LP format.
In order to avoid the ECPO violation, it is imperative that the verbal categories in 3 and 4 be kept distinct through some well-chosen syntactic feature/value specification. This is, in fact, the strategy which U adopts when he postulates LP statements like the ones in 3 and 4, except that the verbal categories are distinguished through the feature specification [+MC] (for 'main clause') and [-MC], respectively. However, unless the feature [MC] is well motivated on independent syntactic grounds, the claim that all natural languages can be described by phrase structure grammars that induce ECPO is in danger of becoming vacuous, since violations of ECPO can always be explained away through the use of some arbitrary distingushing syntactic feature.
Since the [MC] feature plays such a crucial role in U's ID/LP analysis of German word order, it would have been helpful if its use had been motivated in WOCSG by other syntactic phenomena in addition to the word order facts themselves. We believe that such independent syntactic motivation can, in fact, be given. In a theory like GPSG, temporal conjunctions such as als 'when', bevor 'before' and nachdem 'after' have to be introduced through syntactic rules whose sentential categories bear the [-MC] feature in order to obtain the correct subordinate clause word order. However, even though the [MC] feature used in the WOCSG seems to be well motivated, one aspect of its use strikes us as less desirable. In order for the LP statements in 3 and 4 to work properly, it is imperative that the [MC] feature be copied onto verbal categories of bar level 0. This, in turn, has the consequence that each verb has to be marked in the lexicon as taking either + or - as a value for the [MC] feature. Since, as far as we know, no German verbs can appear only in one clause type or the other, this double specification for each verb seems highly redundant.
2.THE ORDER OF VERBAL ELEMENTS IN GERMAN. It turns out that the feature [MC] also plays a crucial role in generating the correct word order for sentences in which auxiliaries, rather than main verbs, represent the finite verb of the sentence. When sentences contain auxiliaries, the auxiliary with the widest scope takes tense rather than the main verb. Those sentences still conform to the clause types exemplified in 1 and 2, i.e. the verb-initial clause and the verb-final clause, except that it is the finite auxiliary verb that occupies the clause-initial or the clause-final position. In both types of clause, the remaining auxiliaries and the main verb are all nonfinite, or [+BSE] ('base' or bare infinitive form), and follow the argument NPs:
(5) Wird Peter das Buch finden koennen?
will[+FIN] Peter the book find[+BSE] can[+BSE]
'Will Peter be able to find the book?'
(6) Ich hoffe dass Peter das Buch finden koennen wird.
I hope that Peter the book find[+BSE] can{+BSE] will[+FIN]
'I hope that Peter will be able to find the book.'
The grammar of WOCSG will assign the constituent structure 7 to the sentence in 5.
(7)
U's analysis of auxiliaries as V makes it possible to state the general position of the finite verb. The intended effect is achieved by the Feature Co-occurrence Restriction in 8 and the LP statements in 9 and 10, which are modified versions of the LP statements 3 and 4.
(8) Feature Co-occurence Restricition: +MC -> +FIN (9) LP: V[+MC] < X (10) LP: X < V[-MC]
When the top sentence node V3 has its [MC] feature instantiated to be [+MC], as in 7, the head V (the dominating node for wird) receives the same feature by the Head Feature Convention; hence the latter node has to precede the V3 sister, due to LP statement 9. The [MC] features of all other V3 and V nodes are instantiated as [-MC], and the LP statement 10 linearizes V[-MC]s (for finden and koennen) to the right of their sisters. The analysis tree of the subordinate clause in 6, on the other hand, can be obtained by instantiating the top sentence node of 7 to be [-MC]. [...]
3. ID/LP AND CONSTITUENT STRUCTURE OF THE MITTELFELD. The range of possibilities for topicalized elements in verb-second main clauses poses a problem not only for the constituent structure given to auxiliaries in WOCSG but also for the constituent structure proposed for the so-called Mittelfeld of German clauses. The deficiency of the WOCSG grammar with respect to the Mittelfeld has, in fact, been acknowledged by U himself in work subsequent to the publication of WOCSG (cf. Uszkoreit 1987).
The Mittelfeld refers to that part of the sentence which contains adjuncts (including the subject) and complements of the main verb. German permits topicalization not only of main verbs (as in 12a), but also of any verbal adjunct or complement alone or in conjunction with the main verb. Thus, 12a/12d are all grammatical.
(12) a. Vorlesen kann Peter seinem Sohn das Buch.
Read can Peter to his son the book
'Peter can read the book to his son.'
b. Peter kann seinem Sohn das Buch vorlesen.
c. Das Buch vorlesen kann Peter seinem Sohn.
d. Seinem Sohn vorlesen kann Peter das Buch.
However, since the WOCSG grammar assigns a flat constituent structure to main verbs and their arguments, 12c and 12d turn out to be problematic for the WOCSG grammar, because they appear to be counterexamples to the widely assumed generalization that exactly one constituent can be fronted in verb-second clauses. [...]
However, even a flat structure analysis in conjunction with LP statements such as 13 will not suffice, since the syntactic features referred to by the ordering strategies are not mutually exclusive. Therefore, ordering paradoxes can arise if the relations in 13 are taken to be LP statements in the ordinary sense of GPSG.
(13) a. +FOCUS < - FOCUS
b. +PRONOUN < - PRONOUN
c. +NOMINATIVE < + DATIVE
d. +NOMINATIVE < + ACCUSATIVE
e. +DATIVE < + ACCUSATIVE
Since any two sister constituents of an ID rule have to satisfy all LP statements of the grammar, LP statements can be thought of as being conjoined by boolean AND. However, if the Mittelfeld of a given sentence contains, for example, an accusative pronoun object and a nonpronominal subject, LP statements 13b and 13d cannot be simultaneously satisfied. In order to avoid such ordering paradoxes, U suggests a modification to the LP format of GPSG itself, which makes it possible to overrule LP violations, if the violation is at the same time sanctioned by some other LP statement.
This modification of the logic of LP statements offers a satisfactory account of the word order variations in the Mittelfeld of German clause structure only if one assumes a flat structure for the Mittelfeld. However, we take this premise to be unwarranted and, instead, agree with Uszkoreit 1987, which proposes a hierarchical constituent structure of the Mittelfeld in order to provide a unified account of German fronting in verb-second clauses. Uszkoreit 1987 treats variations of order in a hierarchical constituent structure of the Mittelfeld as the result of a productive lexical process which permutes the order of subcategorized elements in the subcategorization stack of lexical entries for main verbs. Moreover, the type of LP statements which U formulates in WOCSG to account for the preference orders in the Mittelfeld are turned into constraints on such productive lexical rules. However, U's latest position has the unattractive consequence that statements about constraints on linear order, which seem to be one of the best candidates for a genuine syntactic phenomenon, have to be split up between the syntax and the lexicon, with no clear independent motivation for such a division of labor.
4. THE SYNTACTIC DISTRIBUTION OF SEPARABLE PREFIXES. While we disagree with Uszkoreit 1987 that constraints on word order should be treated as a phenomenon at the syntax-morphology interface, we fully agree with U's claim that an adequate analysis of separable prefixes in German, the topic of Ch. 4, can be found only if both syntactic productivity and lexical idiosyncrasies are taken into account. In the syntax, combinations of prefix and verb-stem such as teil-nehmen ('participate', lit.'part-take') orthographically form one word in verb/final clauses. When the finite main verb occupies the first or the second position of a clause, however, the prefix is separated from the stem and takes the clause-final position, hence demanding special attention in the study of German word order.
U cites a host of idiosyncratic properties of prefix-stem combinations which support his claim that each verb-prefix combination should be treated as a single lexical unit. For example, the separable prefix ski occurs with the verb laufen 'walk fast', but not with the semantically equally plausible verb rennen 'run'. Thus, skilaufen 'ski' is an actual word, while skirennen, with the putative meaning 'ski-race', is not. In addition, U provides numerous examples which show that the class of possible semantic relations between prefix and stem is as openended as in the case of noun-noun compounding. In fact, in extreme instances the meaning of the prefix-stem combination cannot be compositionally derived at all. U cites abnehmen 'lose weight', combining ab 'off' with nehmen 'take', as a case in point.
While each prefix-stem combination forms a single unit in the lexicon, U analyzes the prefix and the stem as separate entities in the syntax. The prefix is introduced by a lexical category SEPREF (separable prefix), whose value is the lexical string itself, while the stem is introduced by a V. In order to avoid ungrammatical combinations of the prefix and the stem, both SEPREF and V are marked by a PREFIX feature.
It is one of the strengths of the WOCSG grammar that the treatment of separable prefixes as independent syntactic units requires only minimal additions to the ID rules and LP statements which we have reviewed so far. A metarule derives a clause rule that introduces a SEPREF in addition to original expansions, i.e. the head V and NP arguments. While the existing LP statements impose initial or final position on the verb stem (depending on whether the V3 is marked as [+MC] or [-MC]), the additional LP statement in 14 distributes a lexical category SEPREF for the separable prefix to the right of any NP arguments.
(14) LP: X2 < SEPREF
U's treatment of separable prefixes makes it possible to account for the distribution of a prefix and a stem not only in verb-initial and verb-final clauses, but also in verb-second clauses. In clauses where the second position is occupied by the finite verb stem, the extracted element in the general ID rule for topicalization can be the prefix as well as any argument N2. However, U notes that there are exceptions to the well-formedness of prefix-fronted clauses, namely in cases like 15, in which the separable prefix lacks autonomous semantic content.
(15) a. Er stellt den Brief zu.
He deliver the letter ?
'He delivers the letter.'
b. *Zu stellt er den Brief.
[...]
Given the wide range of syntactic phenomena covered in WOCSG, it should come as no surprise that we fail to agree with the author on every aspect of his analysis of German word order. However, we fully agree with U's closing remark in Ch. 6 that '... it seems of no great importance whether this book has raised more questions than it has answered, or whether the opposite is true - as long as both kinds of questions are equally interesting' (163). There is no doubt that U successfully meets this criterion throughout the book and that WOCSG provides an invaluable source for any linguist interested in German constituent structure and word order.