LINGUIST List 15.1982

Fri Jul 2 2004

Review: Ling Theories/Pragmatics: Du Bois, et al (2003)

Editor for this issue: Tomoko Okuno <>

What follows is a review or discussion note contributed to our Book Discussion Forum. We expect discussions to be informal and interactive; and the author of the book discussed is cordially invited to join in.

If you are interested in leading a book discussion, look for books announced on LINGUIST as "available for review." Then contact Sheila Dooley Collberg at


  • Olesya Khanina, Preferred Argument Structure

    Message 1: Preferred Argument Structure

    Date: Mon, 28 Jun 2004 02:25:18 +0400
    From: Olesya Khanina <>
    Subject: Preferred Argument Structure

    EDITORS: Du Bois, John W.; Kumpf, Lorraine E.; Ashby, William J. TITLE: Preferred Argument Structure SUBTITLE: Grammar as architecture for function SERIES: Studies in Discourse and Grammar 14 PUBLISHER: John Benjamins YEAR: 2003 Announced at

    Olesya Khanina, Moscow State University, Philological Faculty, Department of Theoretical and Applied linguistics

    INTRODUCTION This is an interesting book on some questions of the relationship between grammar and pragmatics studied within a corpus-based approach. It includes 16 contributions (including a 10-page Introduction), a Preferred Argument Structure bibliography and name, language, subject indexes. All the contributions develop the Preferred Argument Structure framework, first introduced in Du Bois (1987), and are based (with two exceptions) on papers presented during the 'Preferred Argument Structure: The Next Generation' conference held at the University of California, Santa Barbara in the spring of 1995.

    As the title of the conference -- unfortunately, not of the book -- shows, the contributors aimed to confirm previous findings by Du Bois about grammar and pragmatics interaction, as well as to improve and enlarge the theory. Before presenting the overall evaluation followed by the overview of individual papers, it's worth noting that the collection is restricted to quite specific points of argument structure realization, leaving aside all the others. The object of all the studies are (a) overt realization (lexical vs. pronoun vs. zero) and (b) information status (new vs. given) of core arguments (S, A and O, as in Dixon (1979, 1987, 1994)) and, in a few cases, the choice of diathesis. Unfortunately, the avoidance of other points important for argument structure study discussion isn't clearly stated anywhere in the book (desired especially in the Introduction or program paper by Du Bois). Why other questions are not interesting for the present framework, will they be studied later or never within the framework? That's especially important if one thinks of potential reader -- not a fan of the framework but the general argument structure researcher.

    That's the reason for you, probably, to disregard the review and the book itself if you are not interested in discourse information flow regulations, even though you are working hard on some other sides of argument structure (e. g. case frame choice). However, if you are interested, you'll find the collection extremely useful.

    OVERALL DESCRIPTION AND EVALUATION I would consider it useful start with four main Preferred Argument Structure (PAS) constraints, which serve as the basis for all the studies, as not all LINGUIST subscribers may be aware of them. (I personally wasn't before reading the volume.) The constraints are the following: (1) No more than one core lexical argument per clause. (2) No more than one core new argument per clause. (3) No lexical A(ctor). (4) No new A.

    As all discourse constraints, they are assumed to be violable, but in real speech the violations are found quite seldom. These constraints describe the starting point of grammaticization of discourse trends into ergative-absolutive or nominative-accusative types of argument structure tokens. More precisely, ''an ergative system could be seen as a way to mark new information in a consistent way'' (new information can be introduced anywhere, except A), ''whereas a nominative-accusative system could be seen as a way to code topics consistently'' (topics tend to be expressed in A and S positions) (p. 257).

    In the present review, I'll try to avoid evaluation of the PAS approach in general (cf. for example, Goldberg (2003) for a critical discussion) in order to concentrate on presentation and evaluation of the collection itself. Its starting point was indeed the PAS framework, but almost in all the cases the papers' output is larger.

    The structure of the book is clearly outlined in the Introduction (by John W. Du Bois, Lorraine E. Kumpf and William J. Ashby) concluded with a short summary of every paper. One can learn here that only one contribution (apparently, by Du Bois himself) deals with theoretical matters of PAS, all the rest being devoted to practical questions of the hypothesis's applicability to different languages, genres, types of speakers (grown-ups, children, aphasics) etc., and further enlargement.

    It's good that the papers can be read separately: there is a brief outline of the PAS constraints, listed above, and some comments on them in the beginning of each contribution. At the same time, the methods of text analysis and the logic of results' presentation, i.e. the principles of tables and figures, is pretty much the same, which allows the reader not to waste a lot of time on understanding the matters peripheral to the linguistic output. Thus, for every research, the texts in questions were divided into clauses, which were coded for their syntactic type (transitive, intransitive, nominal etc.), and all NPs were coded for their morphosyntactic and pragmatic features. Then the authors edited requests for their data (for example, 'what is the percentage of intransitive clauses with lexical S?') which were really easy to answer after the codification procedure. At last, these requests served for them as a base for generalizations and further inquiries for PAS constraints.

    This uniformity is very useful for the reader, but quite pressing for the editors, as now one can easily note all the controversies between the contributions. It seems that the results of different studies become incomparable to a great extent, if the same notions mean different things for the contributors. For example, NPs coded as 'new' are not of the same nature in all the papers. Sometimes, it can be simply all NPs first mentioning a referent (cf. Clancy, p. 84) or NPs whose referents were not mentioned and not activated otherwise before (cf. Kumpf, p. 111). Or the authors can use different terms and different number of activation status types: new vs. given (Clancy), new vs. non-new (Ashby & Bentivoglio, Allen & Schroeder), new vs. accessible vs. given (Kumpf, Genetti & Crain), previous subject vs. active vs. old vs. new (Arnold), evoked vs. mentioned vs. inferable vs. accessible vs. brand-new (Durie), etc. It's evident that the percentage of new As calculated for the same text with the help of each of this techniques might appear to be different. Even though in some cases it's just a matter of terminological synonymy, such a situation within a collection of papers whose results aim to be comparable seems to be rather strange.

    All my other criticism, valid to the whole volume, addresses to the editors, too. For example, studying more or less the same object, the contributors don't always manage to avoid repetition in the results. Taking as it is, it's not really a drawback, but sometimes their ignorance about each other's results look strange: normally, you don't expect one of the findings of previous paper to be presented as something completely new and fascinating for its originality in the next one (cf. among others, Arnold, p. 237 and Helasvuo, p. 260). I understand that this point makes the life worse for the contributors whose papers go further in the collection, but probably if the editors had cared about it a little bit more, the authors of individual papers wouldn't have suffer from the problem.

    At last, the order of the contributions is not always clear. It's mentioned in the introduction that ''the sequence of chapters was chosen rather with an eye to facilitating a gradual progression into the theory...'' (p. 2), but I'm not sure the current order, indeed, make the reader's life easier. For example, three papers on Mayan languages (by England & Martin, Hofling and Martin), which cite each other or sometimes argue with each other, are situated not very iconically: on pp. 131-158, one can get a general idea of Mayan languages in order to apply it to individual languages studies on pp. 385-436. Two papers devoted to language acquisition (by Clancy and Allen & Schroeder), which are also referring to each other, can be found on pp. 81-108 and pp. 301-338, respectively. At last, Du Bois' theoretical paper is quite difficult to understand in the beginning of the collection, as he praises the PAS framework, or discourse approach to grammar in general, almost without any examples. As a result, the reader, first, doesn't understand what exactly this approach suggests and, then, why it's better that the others. I would suppose that it could be read easier and with greater impact, if it were found at the end of all the contributions, summarizing them and underlining the value of the framework.

    The impression from the volume seems to be slightly worse because of some temporal oddities. Unfortunately, rather often in publishing practice books go out considerably later they were written, so the delay in several years, that can be easily noticed by author's reference to not always the latest studies in the domain, is quite ordinary. But here the situation is more dramatic: in 1987 Du Bois suggested the PAS idea, in 1995 the conference took place, in 2003 the volume was published. Therefore, in the most of contributions, the reviews of contemporary literature look a bit odd, rarely citing theoretical papers after mid-90s or even early 90s. I'm speaking here only about theoretical matters, as it's often easier to add a couple of references to the latest papers on your language or narrow specialization, but theoretical papers may force you to change the conception of the whole contribution. In this volume, the latter was the most remarkable for me in research on anaphora.

    At last, there are also some remarks to be made to the technical editing. Thus, references in the body text lack uniformity among papers: I have met the following variants of citing (Clancy, 1993), (Clancy 1993), Clancy (1993) or Clancy 1993. I noticed also a number of missed references:

    p. 62: ''Ashby & Bentivoglio 1997'' p. 112: ''Brock (1986)'', ''Pritzos (1992)'' pp. 167, 169, 172: ''Thompson MS'' p. 247: ''Siewerska 1994'' p. 259: ''Chafe 1979, 1994'' p. 261: ''Chafe 1994'' p. 277: ''Corston-Oliver (2002)'' (in references: ''Corston 2002'').

    Then, there are some misprints: p. 40: ''see. . . fn. 4'' should be ''. . fn. 3'' p. 112: ''Wallet and Piazza 1988'', but p. 130: ''Wallat, Cynthia'' p. 115: ''he samples'' should be ''the samples'' p. 125: empty part of a line p. 185: ''requently'' should be ''frequently''; empty part of the page p. 211: ''is justified. . '' should be ''is justified. '' p. 366: empty part of a line p. 196: ''lexical and mentions'' should be ''lexical mentions''.

    Two papers (by England & Martin and by Weber) lack abbreviations, even though the former cites glossed Mayan examples and the latter cites commented examples of English conversation.

    In sum, some papers are more fascinating and some are less, but the former are more numerous, so the general impression is really good -- one can read almost any contribution not only because of a professional imperative, but by pure interest. In addition, the most of contributions are written in an easy-to-read language not abusing with difficult terminology. Not being a discourse researcher, I had a nice time reading it and finding a lot of amazing things about different languages.

    INDIVIDUAL PAPER DESCRIPTIONS AND EVALUATIONS John W. Du Bois, ''Argument structure. Grammar in use'', pp. 11-60. From the Du Bois's paper a reader can get a general knowledge of PAS and to understand, more or less, the framework's place in current argument structure research.

    At first the author outlines his views on grammatical structure in general, pointing out that grammar is performed and modified by its use, i.e. by real speakers' demands. Therefore he considers ''illusory'' the belief that ''event semantics by itself could supply the entire basis for understanding argument structure'' (p. 12). That's how he comes to the PAS theory whose aim is announced to answer the question of ''how argument structures are used to do the full range of things that speakers need to do''. (It's a matter apart, whether the PAS really provides the answer promised here.) Having said this, Du Bois opens the subject of discourse pragmatics participation in argument structure formation (is it indeed only pragmatics that ''do the full range of things the speakers need to do''?). Studying discourse factors inside the clause, even in its heart -- argument structure, -- turns out to be quite challenging and rather new for current linguistics, and I suppose it to be the great invention of PAS framework.

    As the author explains himself, ''I do not attempt to give a full explication of the theory here. . . but I do discuss the significance of the model with respect to a number of the current issues in argument structure research'' (p. 15). Unfortunately, he tries to illustrate its significance almost without any general overview of other current frameworks and achievements in the field, at least a short one. As a result the reader is supposed to explore the framework's battle (and victory, of course) with invisible opponents: for a non-specialist, it turns out to be rather hard to understand Preferred Argument Structure place in the modern linguistics of argument structure without knowing the latter.

    My real reservation here is a number of claims which rest unproved: sometimes Du Bois limits in his explanations with 'it's like this because it's like this'-style (cf. for example, the claim about insufficiency of semantic argument approaches). So the main disadvantage of the PAS framework, as it appears in the present contribution, is that it claims to be more significant that it really is. But fortunately, it's visible only in Du Bois' paper, as all other contributions just show how useful the PAS theory is in their work and one can see that for their goals it performs really well.

    Patricia M. Clancy, ''The lexicon in interaction. Developmental origins of PAS in Korean'', pp. 81-108. In this contribution the PAS hypothesis is checked on a corpus of Korean child-caregiver discourse (2 Korean girls audio-taped during a year from the age of 1,8/1,10). Here two types of PAS research can be found -- quantitative and qualitative. The first one is represented by confirmation of all the PAS constraints and by revealing the strong interference that exists between animacy/person and grammatical role (animate A, inanimate O; first/second person A, third person O), a special characteristic of young children's speech, cf. also Du Bois (1987). The second one is questioning the lexical foundations of PAS: Clancy searches for the most frequent verbs in the children discourse and works out the reasons why they are used to govern information flow (to introduce a new participant or to track an old one).

    The author suggests that the most prominent function of young children's speech is the attention management which is firmly connected with 'here-and-now' dimension of their discourse implicating quite particular participants. Looking at individual verbs with their individual argument structure, Clancy can convincingly track PAS emergence at the early stages of language acquisition. Interestingly, further it's PAS itself that becomes ''an important potential source of raw material for the acquisition of grammar'' [p. 105] (word order, case endings, etc.). The latter is proved to be possible because of the place that PAS takes in children's language ability: it links lexical level (individual properties of individual verbs) and discourse level (ways of new participant introduction), being a temporary substitute for grammar.

    Very easy to read, this paper provides the PAS hypothesis with substantial explanation of its origins. Even though it's based only on one language data, it looks much more plausible then Du Bois' a little bit too general explanation that makes us only to accept PAS existence, but not to understand the reasons of its overwhelming popularity within different languages.

    William J. Ashby and Paola Bentivoglio, ''Preferred Argument Structure across time and space. A comparative analysis of French and Spanish'', pp. 61-80. Having noted all PAS studies restriction to modern languages and to geographical, but not historical variations, the authors decided to have a look at two medieval texts, in French and Spanish, and compare their PAS to modern French and Spanish, respectively (cf. their previous research (Ashby 1995, Ashby & Bentivoglio 1993)). As Du Bois' original paper was based on narratives, primarily oral narratives, they tried to find the text that would be the most 'oral' and thus chose 'Chanson de Roland' and 'Cantar de Mio �id'.

    In general, these texts demonstrated the same grammar-pragmatics alignment of features as their modern counterparts, i.e. provided evidence for PAS constraints preservation across time. Few deviations, together with some grammatical changes observed were carefully described and explained, which made the analysis even more plausible.

    Lorraine E. Kumpf, ''Genre and Preferred Argument Structure: Sources of argument structure in classroom discourse'', pp. 109-130.

    This contribution is devoted to a study of PAS variations across genres. The author takes one genre -- American high school science teacher's classroom discourse -- and defines its specificity. She finds out that the latter is the best seen through peculiar treatment of PAS constraints, even though they are by no means violated. All the research is based on four 50-minute videotaped samples.

    Teachers' goals to ''establish joint attention and create optimally available information'' (p. 113) presuppose quite low 'information pressure' (cf. Du Bois (1987)). At the same time one can observe rather high quantity of lexical mentions, even for non-new entities. As Du Bois and some other researches have shown for different languages these two characteristics to be not compatible with each other, Kumpf comes to considering it a 'distinguishing trait' of teachers' discourse. Thus, PAS first serves to provide a linguistic definition of the genre and then to ''supply supporting architecture for the discourse'' (p. 127). And if the former is obviously valuable for linguistics in general, the latter might be considered a bit framework-internal, i.e. it's not that evident where the reason is and where the consequence is -- is it a number of single syntactic choices, based on clause-internal principles, that presuppose the discourse structure or is it the overall discourse structure that dictates syntactic choices within a single clause? And if the question is accurately answered in Clancy's paper on Korean children, here it seems to be left beyond the scope of the study.

    Nora C. England and Laura Martin, ''Issues in the comparative argument structure analysis in Mayan narratives'', pp. 131-157. As the 'seminal' Du Bois' article was based on one Mayan language (Sakapulteko) narratives, the authors became interested in other Mayan languages and other genres, especially because first look at them made visible some 'oddities' whose incompatibility with PAS constraints had to be explained. Their corpus consisted of 18 Sakapulteko pear stories (used originally by Du Bois), 3 Mam folktales, 4 Tektiteko folktales, 3 Mocho folktales, 5 Mocho personal narratives, 4 Q'anjob'al narratives.

    The first part of the contribution concerns the decisions linguists need to make in order to get data for PAS research. Taking more diverse narratives than the Du Bois', England & Martin meet a lot of difficulties in defining a lexical mention, a new mention, a clause, a clause type, basic arguments themselves. They show how many compromise decisions one has to make willing to compare languages with different grammatical structure. These decisions seem to authors to be important, as, first, they provide a valuable insight for the language structure analysis in general, i.e. not only concerning PAS, and then their description allows different PAS studies to be comparable with each other, as labels like 'transitive subject' or 'zero mention', among others, are far from being used uniformly in different languages. The second part of the paper presents revised analysis of preliminary oddities in some languages and a summary of PAS variations among the languages. If general PAS constraints are quite well confirmed by the data, there still rest some puzzles, like low incidence of transitive clauses in Mam and some others.

    Being an interesting comparative paper, my only reservation here is not really friendly treatment of Mayan evidence for those who have no previous acquaintance with it. A list of glosses and some preliminary remarks on language structure could make the contribution more open for non-Mayan linguists (it becomes even more evident, if compared to all other papers of the book (except that by M. Durie, see further), that do have facilitating remarks of that kind).

    Mark Durie, ''New light on information pressure. Information conduits, ''escape valves'', and role alignment stretching'', pp. 159-196. The paper begins with quite extensive theoretical preliminary named 'Motivations for grammatical form' whose aims go far beyond the scope of the paper itself, but are of a substantial value as a number of PAS generalizations relevant for the whole collection. Then the author returns to the domains of his present research which consists of two parts: ''information pressure'' issues, notion first introduced in (Du Bois 1987), and a case study of Acehnese language, whose genetic and geographic affiliation is unfortunately missing.

    The author suggests to make difference between 'information pressure' - ''intuitive, psychological notion'' (p. 163), and 'referential density' -- a metric defined as a number of text referents divided by the number of their mentions in the text. The study of Acehnese texts (an oral narrative, a written narrative and a conversation) allows to define the correlates of greater information pressure: predominance of intransitive predications (noted before in (Du Bois 1980, 1987)), increased use of lexical mentions and more frequent coding of core referents by non-core coding sites. As Acehnese is an active language (unifying A and Sa vs. O and So), it provides interesting data for PAS research: contrary to the Du Bois' hypothesis that greater information pressure would ''correlate with realignment towards ergative-like coding pattern'', the Acehnese doesn't do it, but shows ''stretching of the lexical density hierarchy, involving differentiation within the Actor and Undergoer roles'' (p. 189). That's how Durie comes to the following hierarchy: A <y"a << So <y" <y"e unusual for other languages studied within PAS framework. Thus he argues that the PAS is found out to be partly language-specific, i.e. there is always a certain 'negotiation' between discourse patterns and grammar, ''in other words, pragmatic features of texts are in part conventional and thus language specific'' (p. 191). And even if general PAS constraints are true to all the languages, quite often they reside from much finer language-internal distinctions.

    Despite a slight lack of clarity in language and structuring of the contribution, it seems to be quite valuable to the whole collection, as it refines the original Du Bois' analysis, making it also more diverse in language material.

    Carol Genetti and Laura D. Crain, ''Beyond Preferred Argument Structure. Sentences, pronouns, and given referents in Nepali'', pp. 197-223. The following contribution is based on 10 narratives of Nepali, a dependent-marking language, where argument roles are coded not in verb, as in majority of PAS research languages, but in arguments themselves which entrains a greater use of pronouns. In addition, so called 'clause-chaining' is widespread in the language. The authors argue that these morphosyntactic characteristics of Nepali influence a lot its PAS that follows the general patterns suggested by Du Bois with some deviations. The paper shows these deviations do not violate the PAS constraints, but rather modify them predictably.

    Genetti & Crain find out the significant syntactic unit to be a sentence (as a chain of clauses), not a clause itself: that's how Du Bois' 'Not more than one lexical mention per clause' constraint becomes 'One and only one overt mention per sentence' constraint. The new constraint presupposes a use of pronominal NPs for main participants of the narrative (they can't any more be tracked by verbal affixes). This fact, together with a greater use of pronouns in general (see above), leads to a quite high number of overt mentions, unusual for previous PAS research. Continuing the study of argument coding choice in discourse, the authors define several factors which influence the use of pronouns. Thus, they produce a model of competing motivations -- syntactic, pragmatic, discourse, -- that interact in creation of ''preferred patterns of distribution in the production of natural discourse'' (p. 219).

    As a result, this contribution could be named among the best ones not only because of its clarity, but also due to its input to PAS research. Unlike the most of other papers, it's aim has been not only to prove Du Bois' hypothesis or to give its revised variant, but to suggest, as it title indicates, further types of constraints defining preferred argument structure of a language.

    Jennifer E. Arnold, ''Multiple constraints on reference form: Null, pronominal, and full reference in Mapudungun'', pp. 225-145. Based on data from a quite different language, the paper is inspired by the same idea as the previous one: to state the parameters defining the real usage of possible reference choice (null, pronominal or full NP), i.e. ''not only when it is acceptable to use, e. g. a pronoun, but also when people are more likely to use one'' (p. 225). Even though the question is by no means new for modern linguistics, Arnold suggests some promising paths pointing specially on multifactoral analysis.

    Along with confirming general PAS constraints, her research argues that the form of reference depends on both syntactic and discourse status of the anaphor and the antecedent, as well as on comparative structures of both clauses, the anaphor's and the antecedent's ones (so called 'parallelism effects'). It's worth noting that Mapudungun (y" Mapuche) is one of the appropriate languages for this kind of study, as it permits null subjects and null objects and has an inverse system. These characteristics allow the researcher to look on an individual parameter's impact to the whole, leaving all the others constant (cf. for example, impossibility to divide grammatical functions and semantic roles in languages like English).

    My slight reservation for this contribution would be its weaker theoretical argumentation. Being perfect in interpretation of her corpus data, the author doesn't really enter into polemics with other approaches to reference form research. Quick mentioning of some other studies in the introduction and a couple of remarks on Optimality Theory argumentation seem to be insufficient for the research of the kind. Together with some rather vague statements like ''for transitive subjects, lexical arguments may be suppressed, whereas for intransitive subjects, lexical arguments may be allowed more frequently'' (p. 237) it makes the paper less impressing it could be.

    Marja-Liisa Helasvuo, ''Argument splits in Finnish grammar and discourse'', pp. 247-272. The contributor starts from a number of misleading common places linguistic community, by her idea, is still supposed to share more or less, and aims to disprove them by her research of Finnish conversation (a corpus of 6 multiparty excerpts, about 5 minutes each). Among them, she lists (a) classification into straight accusative, ergative and active languages (i.e. the classification where 'argument splits' are something marginal); (b) word order patterns are not an object for argument splits studies; (c) pronouns exclusion from typological studies of clause structure.

    In her paper, Helasvuo focuses on personal pronouns impact into Finnish argument structure profile in general, and into the process of subject grammaticization. Summarizing the former, she argues that agreement patterns divide all NPs into 1st and 2nd person pronouns, from one side, and 3rd person pronouns and full NPs form the other; at the same time, word order patterns draw this border between all personal pronouns, on the one hand, and full NPs, on the other. The latter is connected with a special role 1st and 2nd person pronouns have in discourse: ''they are always given, identifiable, and tracking'' (p. 268). As other PAS research and the author's personal findings have already shown, these characteristics are typical for subjects, so pronouns can be, thus, regarded as ''the best or most prototypical subjects''. This consideration means that all other NPs filling subject position are trying to follow the pronouns in their discourse parameters. That's how we come to Helasvuo's conclusion that it's extremely important to look at pronouns if one wants to understand PAS of the language.

    Even though not always accurate in dealing with definitions and summarizing (cf. for example, p. 262, 2nd paragraph), the paper is quite interesting and substantial. It seems to be specially valuable, first of all, due to its study of cases where the status of subject is not really evident. Using discourse analysis, the author solves some syntactic problems of different Finnish constructions, making an important non-grammatical contribution to grammatical puzzles.

    Simon H. Corston-Oliver, ''Core arguments and the inversion of the nominal hierarchy in Roviana'', pp. 273-300. This paper represents a convincing example of explanation of typologically unusual grammar features by real language use, i.e. discourse motivation. An Austronesian language Roviana, studied by the author in his fieldwork, has several features of the kind: (a) split-ergativity consists of ergative-absolutive vs. neutral case marking (ergative-absolutive vs. nominative-accusative being normal); (b) absolutive case is more marked than ergative case; (c) split-ergativity is conditioned by clause type, main clauses having ergative-absolutive distinction and subordinate clauses being neutral (normally, in this kind of split, main clauses have nominative- accusative distinction and subordinate ones -- ergative-absolutive distinction); (d) ergative-absolutive distinctions are applied to NPs on the left of Silverstein's scale (cf. (Silverstein 1976)), unlike all previous predictions that ergativity expands form its right side (cf. (Dixon 1994)) ~ so called 'inversion of the nominal hierarchy', as in the title; (e) verbal pronominal affixes refer only to objects; (f) only 3PL pronominal affix has no overt expression.

    The main answer to this puzzle was found in Roviana si/se particle: diachronically it was used to mark new information, and as new arguments tend to appear only in S/O positions -- both in Roviana and universally (= one of the main PAS claims), it was reanalyzed as absolutive marker. The small number of new arguments in subordinate clauses -- both in Roviana and universally, led to the absence of this particle, i.e. the absence of case distinction out if the main clause. Having cited the paper's explanation to (a-c) phenomena, I'll just say that for (d-f) the author uses the same discourse-based approach, that seems to work quite well for Roviana.

    While enjoying the fruitful marriage of discourse and grammatical parameters, a reader might ask why all the discourse processes mentioned by Corston-Oliver don't interact with grammar in other languages so that the issues (a-f) became if not wide-spread, at least normal, in them. And if the relative rarity of new information markers could explain (a-c), for (d-f) the question will be still open.

    Shanley E. M. Allen and Heike Schroeder, ''Preferred Argument Structure in early Inuktitut spontaneous speech data'', pp. 301-338. The goals of the following contribution have a lot in common with Clancy's in her paper on Korean acquisition. Having studied videotaped excerpts of four Inuktitut-speaking children (aged from 2 to 3,6), the authors, first, demonstrate the existence of all the four PAS constraints, and second, look more precisely on the slight deviations from PAS predictions. Allen & Schroeder launch their research with the thoughts that it would be interesting to compare Inuktitut data to Du Bois's Sakapulteko, as both are ergative languages with very rich and complex verbal morphology. In addition, there is not so much research on PAS acquisition, as well as on PAS in Eskimo-Aleut languages.

    If compared to other languages studied within PAS framework (direct comparison is done for Sakapulteko (Du Bois 1987), early Korean (Clancy, present volume) and another Eskimo-Aleut language Yup'ik (Rubino 1996)), the unusual feature of Inuktitut is reported to be a relatively low percentage of lexical arguments, on the one hand, and of transitive clauses, on the other. For structurally close Sakapulteko and Yup'ik, the first difference is explained by peculiarities of children discourse -- low information pressure and underdeveloped language system leading to decreased use of lexical mentions. For early Korean, it turns out to be more structural: rich verbal cross- referencing allows not to use overt NPs more often. As for the second difference, the authors argue that it's connected with a wide range of fully productive detransitivizing processes (passive, antipassive, noun incorporation). Indeed, they allow to have 1st and 2nd person pronouns -- the most relevant referents in children discourse -- as subjects independently from these referents' roles in the situation, and thus, are used extremely often, so intransitive predications prevail.

    As it was already said about Clancy's contribution on language acquisition, this type of research is very valuable for PAS framework, as it supplies the theory with its developmental origins. And again this is done convincingly and accurately, which sometimes lacks for other papers of the volume.

    Susan E. Kohn and Ana Cragnolino, ''The role of Preferred Argument Structure for understanding aphasic sentence planning'', pp. 339-351. To understand the basic nature of the PAS constraints, Kohn & Cragnolino decided to look at its role in sentence planning. For the current study, they have restricted the domain by taking into consideration only decontextualized isolated sentences with transitive verbs. The experiment compared normal speakers of English and those having agrammatic aphasia (neurologically-induced language deficits): the differences in two groups' results led to definition of the stage at which the PAS constraints operate. The corpus of 1558 sentences for 30 control subjects and 620 sentences for 13 aphasics was created by asking each person to generate sentences based on 'target verbs' that were shown to them by researches.

    The results of the study are quite promising for the theory, as, first, even isolated sentences follow the Du Bois' constraints -- the claim was proved to be true not only for control subjects, but to most of aphasics; second, PAS constraints violation was strongly associated with syntactic ungrammaticality. Thus, the PAS principles were demonstrated to operate at the early stages of sentence planning. This fact, together with the similar findings in language acquisition research (see Clancy's and Allen & Schr�der's contributions), gives even more weight to the principles, as it shows how basic they are. It seems that psycholinguistics of the kind could support a lot the PAS framework by supplying it with explanational research in comparison to other more descriptive typological findings.

    Elizabeth G. Weber, ''Nominal information flow in the talk of two boys with autism'', pp. 353-383. The second contribution on speech features of people with language disorder is, in its turn, a real corpus study. Based on 30-minutes spontaneous play interaction excerpt, it illustrates that two boys with autism do follow all the PAS constraints, even though there are some other slight deviations from ordinary discourse patterns (like overuse of non-arguments, etc.). It would be just the confirmation on broader discourse material of Kohn & Cragnolino's findings, if not autism were completely different in its language-related characteristics from aphasia.

    Citing a rich bibliography of behavioral and communicative research of autism, the author reminds that it's defined by all types of interaction problems. Among others, there can be mentioned these persons' inability to initiate and maintain a conversation, to infer mental states of others (and thus, to differentiate Given-New, etc.) and all sorts of deficit connected with social use of the language. One can easily predict these peculiarities to be reflected in their discourse: indeed, autistic people are quite often diagnosed by discourse mismatches.

    That's why it's really surprising to observe the PAS constraints, that are claimed to be discourse ones, functioning normally in autistic speech. And if Weber comments them only as ''an interesting addition to the literature on the language of persons with autism because they reveal a previously unidentified discourse competence'' (p. 379), I would then wonder whether it's really a discourse competence. Adopting a different point of view, the PAS constraints could be regarded as intrinsic part of verbal semantics or clause syntax, i.e. something residing within a clause and thus not being a discourse-level characteristic. Actually, this possible interpretation has been slightly irritating me from the very first paper of the collection, as the contributors never discuss this option, but Weber's research provides real arguments for the non-discourse nature of the constraints. Unfortunately, nor she neither the editors mention this point.

    Charles Andrew Hofling, ''Tracking the deer. Nominal reference, parallelism and Preferred Argument Structure in Itzaj Maya narrative genres'', pp. 385-410. The contribution reports an attempt to follow all preferred argument structure constraints and trends, without limiting to the four suggested by Du Bois and explored by many others (see other papers of the volume). Such an extensive work becomes possible as soon as the author doesn't deal with a corpus, but just two texts -- a personal narrative about recent events and a mythic tale, both recounted by the same speaker of Itzaj Maya and devoted to the same subject (deer- hunting).

    Hofling looks in the smallest details at how various kinds of information are introduced for the first time and tracked throughout the text afterwards, at different impacts made by case, animacy, topicality, morphosyntactic form and their interaction. As a result, he, first, shows logical and predictable nature of the discourse: cf. for example, a fundamental difference in marking topical humans, on the one side, and inanimate environment serving for the context where they act, on the other side. More generally, nominal reference, compared to a skeleton of any text, is demonstrated to play a crucial role in discourse organization. Second, the author defines the difference between two types of narratives; he highlights the relativity of the latter and claims that ''genre distinctions are marked by different frequencies. . . rather than absolutely'' (p. 407).

    Being very accurate and sometimes really insightful, the analysis still seems to be somewhat hermetic: It's based on such a scrupulous study of individual sentences that it becomes really hard to imagine its theoretical impact. Indeed, it could be a great demonstration of possibility of individual discourse explanation, as well as of the authentic reality of such features as topicality, animacy, etc., but, unfortunately, not a component for cross-linguistics comparison which the PAS framework aims to build.

    Laura Martin, ''Narrator virtuosity and the strategic exploitation of Preferred Argument Structure in Mocho: Repetition and constructed speech in Mocho narrative'', pp. 411-435. The last paper follows exactly the same model as the previous one: basing on a single Mocho folk narrative, the author tries to define, in terms of PAS, techniques the narrator uses to make the story interesting and amazing, as well as to hold listeners' attention. She presents two basic mechanisms: repetition and direct quotation. The former consists mainly of referring to a prominent participant by the same lexical NP, without using any synonyms or pronouns and with very restricted zero mentions. The latter is said to be underestimated in Mayan studies, as its discourse role is significantly bigger in comparison to direct speech of European languages: one can hardly find any narrative lacking it.

    Analyzing individual discourse units and individual stylistic impacts of the techniques, Martin raises also a number of questions about speaker's consciousness of them, as well as of PAS constraints in general. Suggesting further paths of investigation, the paper gains some theoretical value, as all the criticism addressed to the previous contribution can be equally mentioned here: the results of the study are still too individual to serve for a base for cross-linguistic comparison.

    CONCLUDING REMARKS For all the results of individual studies, the authors suggests a perfect discourse explanation, showing how significant the discourse regulations are in language use. But sometimes the same results make a reader think of another explanation: for example, it could be argued that in discourse flow the speaker chooses not an argument structure itself, but rather a verb that has necessary argument structure, so probably, PAS hypothesis results not in differentiating core arguments, but in differentiating types of verbs on their argument structure basis. As M. Durie notes in footnote (7), p. 192, ''a necessary task, which Du Bois didn't attempt in his 1987 paper, and which goes beyond the scope of this paper, will be ''to evaluate to what extent Du Bois's account of the full range of Silverstein [Silverstein 1976] hierarchy effects is more satisfactory than many of the competing explanations.'' And that's really what this collections lacks, a detailed comparison of the PAS approach to the others (cf. among others, (Hawkins 2002, Hawkins 2004) for typological asymmetries in argument structure), if it claims to be not only a descriptive tool for 'pragmatic linking', but also a functional explanation theory of basic clause principles.

    REFERENCES Ashby, William J. 1995. French presentation structure. In Amastae, John et al. (eds.). Contemporary research in Romance linguistics. Amsterdam, Philadelphia: Benjamins, 91-104.

    Ashby, William J. and Bentivoglio, Paola. 1993. Preferred Argument Structure in spoken French and Spanish. Language Variation and Change 5, 61-76.

    Dixon, Robert M. W. 1979. Ergativity. Language 55, 59-128.

    Dixon, Robert M. W. 1987. (ed.). Studies in Ergativity. Amsterdam: North-Holland.

    Dixon, Robert M. W. 1994. Ergativity. Cambridge: Cambridge University Press.

    Du Bois, John W. 1987. The discourse basis of ergativity. Language 63, 805-855.

    Du Bois, John W. 1980. Beyond definiteness: The trace of identity in discourse. In Chafe W. L. (ed.). The Pear Stories: Cognitive, Cultural, and Linguistic Aspects of Narrative Production. Norwood, NJ: Ablex, 203-274.

    Goldberg, Adele E. 2003. Pragmatics and Argument Structure. In Horn, Larry and Ward, Gregory (eds.). Handbook of Pragmatics. Malden, MA: Blackwell, 84-112.

    Hawkins, John A. 2002. Symmetries and asymmetries: Their grammar, typology and parsing. Theoretical Linguistics 28, 211-227.

    Hawkins, John. A. 2004. Efficiency and Complexity in Grammars. Oxford: Oxford University Press.

    Rubino, Carl. 1996. The introduction of new information and Preferred Argument Structure in Central Alaskan Yup'ik narratives. In Mithun M. (ed.). Santa Barbara Papers in Linguistics Volume 7: Prosody, Grammar, and Discourse in Central Alaskan Yup'ik. Santa Barbara: Linguistic Department, University of California at Santa Barbara, 139-153.

    Silverstein, Michael. 1976. Hierarchy of features and ergativity. In Dixon, Robert M. W. (ed.). Grammatical Categories in Australian Languages. Humanities Press, 112-171.

    ABOUT THE REVIEWER Olesya Khanina is a PhD student of Moscow State University, Philological Faculty, Department of Theoretical and Applied linguistics. Her research interests includes typology of argument structure (possible ways of aspect and actancy interaction), as well as typology of desiderative meaning & expression. She is especially interested in field data for these and other typological studies (fieldwork in Tatar, Chuvash, Balkar (Turkic), Nenets (Uralic)).