CSLI MONTHLY ----------------------------------------------------------------------- January 1987 Vol. 2, No. 4 ----------------------------------------------------------------------- A monthly publication of The Center for the Study of Language and Information ------------------ Contents Functional Uncertainty 1 by Ronald Kaplan, John Maxwell, and Annie Zaenen CSLI Visiting Scholars 3 CSLI Publications 3 ------------------ FUNCTIONAL UNCERTAINTY Ronald M. Kaplan, John Maxwell, and Annie Zaenen Utterances in natural language obviously are made up of words and morphemes of different types arranged in a temporal or left-to-right sequence. By virtue of the particular words and morphemes a given utterance contains and their particular ordering, the utterance carries meaning---the speaker's claims about the world, himself, the discourse, etc. The problem of grammatical description is to determine just how a language uses the surface properties and relations of an utterance to encode the information that it carries. In simple sentences the connection between surface form and underlying meaning relations, for example, predicate-argument relations, is often easy to see. For the English sentence `The teacher telephoned the student,' the teacher is understood as the agentive argument of the telephoning and the student is the goal or target argument of the action. Here local relations (immediate adjacency and precedence) in the surface string directly encode the local relations (between the predicate and its arguments) of a more abstract representation. An equivalent Japanese sentence would use other properties of the utterance to encode the same abstract predicate-argument relations, attaching the particle `ga' to indicate the agentive argument and `ni' to mark the goal. But again, local surface properties map in a simple way onto local predicate-argument relations. Of course, it is quite unusual for the connection between surface properties and predicate-argument relations to be as clear as it is in this example. The local properties of the first words in the garden-path sentence `The cherry blossoms in the spring' are compatible with two different predicate-argument configurations, but only one of the local possibilities is consistent with the few remaining words in this string. There are familiar and uncontroversial grammatical devices for characterizing finite degrees of syntactic `uncertainty'. Every linguistic theory permits finite disjunctions of rules and senses of lexical entries, and the readings of an utterance are derived by independently considering all possible combinations of the alternatives that are relevant to the words it contains. With a lexicon and grammar that provides for finite disjunction, substantial computational resources may be required in the worst case to assign the proper predicate-argument relations to a given string, but the grammatical logic of `finite uncertainty' for simple sentences is quite straightforward. Unbounded Uncertainty For some constructions in natural language, however, there is no finite bound on the number of possible predicate, argument relations that a given word or phrase may enter into, given the information that is locally available in the surface context of the phrase. In English topicalizations, questions, and relative clauses, a phrase at the beginning of the string is understood as being an argument or modifier of a predicate that appears later on. For example, `The girl who John saw ___ yesterday was Mary' is understood with the `girl' as an argument to the `saw' predicate, as it would be if it appeared in the string at the "gap" position indicated by the underline, but there is nothing in the information locally available at the front of the string that determines this particular argument relation. `The girl who John saw Bill talking to ___ yesterday was Mary' begins with the same sequence of words but in this case the `girl' is understood as an argument of the `talk' predicate. In some languages case or other morphological markings will limit the kind of relations such an initial constituent can enter into, but that does not solve the main problem, namely, that there is no obvious bound on the number of predicates that the initial noun phrase can bear an argument relation to. Thus no finite enumeration of disjunctive specifications (e.g., first argument of second predicate, second argument of third predicate...) can be sure to define the predicate, argument configuration required by the words found at arbitrary distances to the right. This `unbounded uncertainty,' considering nothing but the information at the beginning of the string, is resolved, of course, when the string as a whole is taken into account: the basic constraint on these so-called long-distance dependencies is that the initial phrase has to be interpreted as an argument or adjunct of a subsequent predicate which is otherwise missing that argument or which can be modified by the adjunct. Long-distance dependencies thus exemplify a class of phenomena wherein unbounded local uncertainties are resolved by information available only from arbitrarily far away; some seemingly unrelated infinitival constructions of German, Dutch, Japanese, Finnish, and other languages also have this general characterization. Unbounded uncertainties raise a number of important conceptual, linguistic, and technical issues. The traditional transformational account of these dependencies was formulated entirely in terms of phrase structure representations and operations. This is not surprising, since phrase structures were the only representational structures those theories admitted. Phrase structures were used to characterize the surface organization of the utterance, for which part-of-speech categories, phrasal hierarchies, and temporal order were quite appropriate. Phrase structure representations, so-called deep structures, were also used to encode the underlying predicate-argument relations of a sentence, although in this case the appropriateness of part-of-speech labeling, the single-mother constraint, and temporal order is much less clear. Strings with long-distance dependencies were derived from corresponding deep structures by transformations that deleted argument material from positions near their predicates (the gap positions above) and inserted them in clause initial positions. Linguistic investigations based on this transformational mechanism were very successful in bringing to light a large number of further restrictions on what may intervene between a fronted phrase and the predicate that it is related to. But it has proven difficult to give simple and natural statements of these restrictions in terms of elementary phrasal properties. Phrasal or Functional? Lexical-functional grammar, as presented by Kaplan and Bresnan (1982), offers a quite different set of representations and operations for grammatical description. Phrase structure representations, called constituent structures (c-structures), are used in LFG only to characterize the superficial arrangement of words and phrases in a sentence. Predicate argument relations are grammatically encoded in a structure that has different primitives and very different mathematical properties. The functional structure, or f-structure, is a hierarchy of finite functions that assigns values to the subject, predicate, object, and other primitive grammatical relations of a sentence. F-structures are described in algebraic terms, through Boolean combinations of equations that define the values that make them up. The equations (f1 SUBJ)= f2 and (f2 NUM)=SG, for example, together assert that f1 is a function that takes the symbol SUBJ into f2, a function that takes NUM into SG, indicating that the subject's number is singular. F-structure does not incorporate any notion of part-of-speech or syntactic category labelling, it does not reflect temporal precedence in an inherent relation among its elements, but it does naturally allow the same information to be seen at the end of two different access paths, an analog of a multi-mothered tree. There are no transformational operations in LFG. LFG thus provides a collection of conceptual and descriptive primitives different from those of traditional transformational theories and their present-day structuralist descendants. Phrase structure is formally quite distinct from the functional structure that encodes underlying predicate-argument relations, and grammatical relations (subject, object, adjunct) play a much more central role in grammatical descriptions than do the phrasal notions of category, precedence, and dominance. However, LFG's basic representations and operations do not by themselves offer an obvious way of characterizing unbounded uncertainties. An f-structure representing the predicate-argument relations in long-distance dependency sentences cannot be described with finite disjunctions of simple function-argument-value equations based solely on immediate dominance in a context-free constituent structure. Some augmentation of the formal system is necessary, but exactly what kind of additional mechanisms should be defined? Kaplan and Bresnan (1982) chose to extend the phrasal machinery of LFG to account for long-distance dependencies. They made such a minimal departure from traditional approaches mainly for two reasons. First, the large body of known constraints, the result of transformational investigations, were framed entirely in phrasal terms, so this seemed to be the natural domain for describing these phenomena. Second, no clear outline of a possible functional extension was at hand, while there did exist an earlier computational model for a sufficient c-structure-based device, the HOLD cell mechanism of one of LFG's immediate precursors, the Augmented Transition Network formalism (Woods, 1970; Kaplan, 1974; Wanner & Maratsos, 1978). Thus, Kaplan and Bresnan proposed a simple notation that forced an identity between the f-structures corresponding to nodes that were arbitrarily distant in the c-structure. The various restrictions on long-distance dependencies could then be stated in terms of properties of the categorial paths between the two nodes whose f-structures were so identified. But does this allow us to express the correct generalizations? Kaplan and Bresnan (1982) and many others have observed that the assimilation of long-distance dependency restrictions to properties of phrasal organization is tenuous at best. For instance, a topicalized phrase does not always have the same syntactic category that is required of arguments in predicate-local position. In the following strings a topicalized sentential complement (S') corresponds to an argument that can be filled only by a noun phrase when it appears local to its predicate: That he might be wrong he didn't think of. *He didn't think of that he might be wrong. He didn't think of that problem. We and other LFG researchers at CSLI, Xerox, and elsewhere have been reexamining the traditional statements of the restrictions on long-distance dependencies, attempting to determine whether they cluster more closely with properties of a sentence's surface form or with properties of its underlying predicate-argument relations. This question is particularly significant for an LFG account because of LFG's clear formal distinctions between c-structure and f-structure; transformational approaches have to a certain extent side-stepped this question because phrasal representations are used at all levels of description. Surface and underlying structures are, of course, highly correlated, but there are a variety of situations where they are slightly out of alignment and which therefore can provide evidence relevant to this question. Consider, for example, the restrictions on the number of intervening NP or S nodes on a dependency path, wherein some NP or S nodes act to block the path while others do not. It has long been known that the blocking cannot be ascribed to the identity of the categories without taking a larger context into account. Traditionally, the relevance of a larger phrasal context has been assumed, as illustrated with the following English contrast: Who did Mary say that Bill saw __? *Who did Mary go away before Bill saw __? It is generally presumed that this difference in grammaticality is due to the fact that the `that'-clause in the first example is a part of the verb phrase whereas the `before'-clause is a sister of the verb phrase. There is, however, another way to analyze this contrast: The S node that does not block the dependency functions as a complement to the verb while the blocking one functions as an adjunct. The idea that "island constraints" depend on grammatical functions does not originate with us; the earliest generative proposal in that vein that we are aware of is Cattell (1976). Most of the earlier accounts, however, try to define the functional relations in terms of tree configurations. As discussed in Kaplan and Zaenen (forthcoming), Icelandic provides a striking example of the problems that these attempts can lead to. Long-distance dependencies are blocked in some Icelandic sentences whose surface structures have exactly the same phrasal configurations as sentences where the dependencies are perfectly acceptable. What differs is the functional organization: just as in English, the dependency is unacceptable when the initial phrase must be interpreted as an adjunct but permissible when it is an argument of the subsequent predicate. A direct appeal to functional notions permits a straightforward, coherent statement of the constraint, whereas the encoding in phrase structure requires a succession of transformationally-related abstract trees with no independent justification. It might seem that a direct functional account would miss other important generalizations. Kaplan and Bresnan cite the following examples to show that syntactic categories, which do not appear in f-structure, can sometimes be required to match: She'll grow that tall/*height. She'll reach that *tall/height. The girl wondered how tall she would grow/*reach. The girl wondered what height she would *grow/reach. But what seems to be a requirement of matching categories (`grow' taking an adjective phrase and `reach' taking an NP) can instead be seen as an interaction between the argument subcategorization requirements of the predicates involved, which are formalized in LFG in terms of the Completeness and Coherence conditions. The predicate `grow' imposes a grammatical predication relationship between its subject and the head of the following adjective phrase `tall', while no such relationship obtains between `She' and `height'. `Tall' expects a grammatical argument to come from an external source, the subject of the matrix predicate, while `height' is grammatically complete. In the starred examples above, when the initial `tall' phrase is linked to the `reach' predicate it does not receive a subject in violation of Completeness, and when `height' is linked to `grow' and does receive a subject, the Coherence condition is violated. The categorial information in c-structure is not needed to explain this contrast; it follows from independently motivated functional considerations. Kaplan and Zaenen (forthcoming) show that the apparent categorial mismatching requirement in the `think of' paradigm given above can be accounted for in a similar way, but it would lead us too far afield to demonstrate that here. A Functional Formulation Our reexamination of these and many other long-distance dependency restrictions strongly suggests that the relevant generalizations are indeed best stated in functional or predicate-argument terms. And this has forced us to define and investigate a new formal device that permits a functional statement of constraints on unbounded uncertainties. We started from an idea that Kaplan and Bresnan (1982) briefly considered but quickly rejected on mathematical and (we have just suggested, mistaken) linguistic grounds. They observed that each of the possible underlying positions of an initial phrase could be specified in a simple equation locally associated with that phrase. In the topicalized sentence `Mary John telephoned yesterday,' the equation (in LFG notation) (^ TOPIC)=(^ OBJ) specifies that `Mary' is to be interpreted as the object of the predicate `telephoned'. In `Mary John claimed that Bill telephoned yesterday,' the appropriate equation is (^ TOPIC)=(^ COMP OBJ), indicating that `Mary' is still the object of `telephoned', which because of subsequent words in the string is itself the complement (indicated by the function name COMP) of the top-level predicate `claim'. The sentence can obviously be extended by introducing additional complement predicates (`Mary John claimed that Bill said that .... that Henry telephoned yesterday'), for each of which some equation of the general form (^ TOPIC)=(^ COMP COMP .... OBJ) would be appropriate. The problem, of course, is that this is an infinite family of equations, and hence impossible to enumerate in a finite disjunction appearing on a particular rule of grammar. For this technical reason, Kaplan and Bresnan abandoned the possibility of specifying unbounded uncertainty directly in functional terms. In our recent work we have reconsidered the general strategy that Kaplan and Bresnan began to explore. Instead of formulating uncertainty by an explicit disjunctive enumeration, however, we provide a formal specification that characterizes the family of equations as a whole. A characterization of a family of equations may be finitely represented in a grammar even though the family itself has an infinite number of members. We develop this notion from the elementary descriptive device in LFG, the functional-application expression. This has the following interpretation: (f s)=v holds if and only if f is an f-structure, s is a symbol, and the pair in f. This notation is straightforwardly extended to allow for strings of symbols, as illustrated in expressions such as (^ COMP OBJ) above: (f sx) is equivalent to ((f s) x), for s a symbol and x a (possibly empty) string of symbols. (f E) is equivalent f, where E is the empty string. (Note: A, B, and E represent the greek letters alpha, beta, and epsilon) Finally, we allow the argument position in these expressions to denote a (possibly infinite) set of strings. This is our functional representation of unbounded uncertainty: (f A)=v holds if and only if for some x in the set of strings A, (f x)=v. In the case in which A is a finite set this formulation is equivalent to a finite disjunction of equations over the strings in A. Passing from finite disjunction to existential quantification enables us to capture the intuition of unbounded uncertainty as an underspecification of exactly which choice of strings in A will be compatible with the functional information carried by the surrounding surface environment. We, of course, impose the further requirement that the membership of A be characterized in finite specifications. More particularly, it seems linguistically, mathematically, and computationally advantageous to require that A in fact be drawn from the class of `regular languages'. The characterization of uncertainty in a particular grammatical equation can then be stated as a regular expression over the vocabulary of grammatical function names. The infinite uncertainty for the topicalization example above can now be specified by the equation (^ TOPIC)=(^ COMP* OBJ), involving the Kleene closure operator; a specification for a broader class of topicalization sentences might be (^ TOPIC)=(^ COMP* GF), where GF denotes the set of primitive grammatical functions. The restriction discussed above for English and Icelandic that adjunct clauses are islands might be expressed with the equation (^ TOPIC)=(^ (GF-ADJ)* GF). With this notation it becomes a very simple matter to express the varying constraints for different kinds of uncertainty constructions, and we are now investigating the general nature of these constraints across different constructions and across different languages. One remarkable consequence of our functional approach is that appropriate predicate-argument relations can be defined without relying on empty nodes or traces in c-structure. After further exploration, we expect to formally restrict the phrasal component of LFG to prohibit such bookkeeping devices and thus make the c-structure much more faithful to the sentence's superficial organization. We have also studied the mathematical and computational properties of our functional uncertainty conception. It is clearly decidable whether a given f-structure satisfies a functional description that includes uncertainty specifications. Since a given f-structure contains only a finite number of function-application sequences, it contains only a finite number of strings that might satisfy an uncertainty equation. The membership problem for the regular sets is decidable and each of those strings can therefore be tested to see if it makes the equation hold. It is less obvious that the satisfiability problem is decidable. Given a set of equations describing a functional structure for a sentence, can it be determined that a structure satisfying all the equations does in fact exist? For a trivial description with a single equation, the question is easy to answer. If the equation has an empty uncertainty language, containing no strings whatsoever, the description is unsatisfiable. Otherwise, it is satisfied by the f-structure that meets the requirements of any string in the language, say the shortest one. The difficult case arises when the functional description has two uncertainty equations, say (f A)=vA and (f B)=vB. If A contains (perhaps infinitely many) strings that are initial prefixes of strings in B, then the strings that will be mutually satisfiable cannot be chosen independently from the two languages. The choice of x from A and xy from B implies a further constraint on the values vA and vB: for this particular choice we have (f x)=vA and (f xy)=((f x) y)=vB, which can hold only if (vA y)=vB. Kaplan and Maxwell (forthcoming) show, based on a state-decomposition of the finite-state machines that represent the regular languages, that there are only a finite number of ways in which the choice of strings from two uncertainty expressions can interact. The original equations can therefore be transformed into an equivalent finite disjunction of derived equations whose uncertainty expressions are guaranteed to be independent. The original functional description is thus reducible to a description without uncertainty, where each of the remaining regular languages is replaced by a freely chosen member string. The satisfiability of descriptions of this sort is well established. A similar proof of satisfiability has been developed by Mark Johnson. Our notion of regular functional uncertainty thus has very nice mathematical properties. The state-decomposition involved in the satisfiability proof translates into a very attractive algorithm for resolving functional uncertainties as other phrasal and functional constraints are computed during the parse of a sentence. This algorithm expands the uncertainties incrementally, introducing at each point only as much disjunction as is necessary to avoid interactions with other functional information that has already been taken into account. We have recently added this algorithm and the functional uncertainty notation to our LFG Grammar Writer's Workbench, and we can now rigorously but easily test a wide range of linguistic hypotheses. We have also begun to investigate a number of computational heuristics for the efficient, controlled expansion of uncertainty. We have obtained, for example, significant reductions in computational resources by not expanding uncertainty expressions involving grammatical functions to which the Coherence condition applies until after a local predicate has been incorporated into the f-structure. In a left-to-right parsing model, the processing load therefore increases in relative clauses just after the predicate is seen, and this might have a variety of interesting psycholinguistic implications. We first proposed the idea of functional uncertainty as sketched in this paper to account for the properties of long-distance dependencies within the LFG framework. In this framework, it has already shed new light on long-standing problems like island constraints (see, e.g., Saiki (1986) for an application to Japanese). But the notion is potentially of much wider use: first, it can be adapted to other 'Bay Area' grammar formalisms to handle facts of a similar nature; and second, it can be used to handle phenomena that are traditionally not thought of as falling into the same class as long-distance dependencies but that nevertheless seem to involve nonlocal uncertainty. A discussion of its application in the LFG framework to infinitival complements can be found in Johnson (1986) for Dutch and Netter (1986) for German; Karttunen (1986) discusses how similar extensions to Categorial Unification Grammar (CUG) can account in a simple way for related facts in Finnish that would otherwise require type-raising. Halvorsen has suggested that scope ambiguities in semantic structures might also be characterized by this device. Our new, clearly formalized conception of uncertainty offers a novel perspective for solving a wide variety of linguistic problems. We expect research over the next year or so to uncover further natural applications for this mechanism and also to suggest refinements and restrictions to improve its explanatory significance. References Cattell, R. 1976. Constraints on Movement Rules. Language 52, 1. Johnson, M. 1986. The LFG Treatment of Discontinuity and the Double Infinitive Construction in Dutch. CSLI report CSLI-86-65. Kaplan, R. M. 1974. Transient Processing Load in Relative Clauses. Doctoral dissertation, Harvard University. Kaplan, R. M., and J. Bresnan. 1982. Lexical-functional Grammar: A Formal System for Grammatical Representation. In J. Bresnan (ed.), The Mental Representation of Grammatical Relations. Cambridge: MIT Press. Kaplan, R. M., and J. Maxwell. Forthcoming. Functional Uncertainty. Kaplan, R. M., and A. Zaenen. Forthcoming. Wh-constructions and Constituent Structure. In M. Baltin and A. Kroch (eds.), Alternative Conceptions of Phrase Structure. Karttunen, L. 1986. Radical Lexicalism. In M. Baltin and A. Kroch (eds.), Alternative Conceptions of Phrase Structure. Netter, K. 1986. Getting Things out of Order. COLING 11. Saiki, M. 1986. A New Look at Japanese Relative Clauses: A Lexical Functional Grammar Approach. Descriptive and Applied Linguistics 19. Wanner, E., and M. Maratsos. 1978. An ATN Approach to Comprehension. In M. Halle, J. Bresnan, and G. A. Miller (eds.) Linguistic Theory and Psychological Reality. Cambridge: MIT Press. Woods, W. 1970. Transition Network Grammars for Natural Languages. Communications of the ACM 13. ------------------ CSLI VISITING SCHOLARS Robin Cooper Department of Linguistics and Phonetics Lund University and Department of AI and Cognitive Sciences University of Edinburgh Dates of visit: January 1987--February 1987 Cooper worked with Jon Barwise on a book called "Linguistic and Logical Aspects of Situation Semantics." Elisabet Engdahl Department of Linguistics and Phonetics Lund University and Department of AI and Cognitive Sciences University of Edinburgh Dates of visit: January 1987--February 1987 Engdahl worked with Annie Zaenen on the outline of their LSA summer course "Topics in Germanic Syntax" and advised students on their ongoing work in that area. She also gave a lecture about bound anaphora and wh-constructions to the natural language group (NLTT) at Xerox PARC, discussed her work on questions and on parasitic gaps with various researchers, and participated in STASS activities. Marilyn Ford Department of Linguistics Griffith University Dates of visit: November 1986--January 1987 Ford returned to CSLI for her "summer vacation" and worked with Joan Bresnan and David Swinney on language processing. Norbert Gstrein Innsbruck University Dates of visit: November 1986--June 1987 Gstrein is here on a graduate scholarship from the Austrian-American Educational Commission. He has recently been working on the logic and semantics of questions, and will be studying situation semantics while at CSLI. William Hanson Department of Philosophy University of Minnesota Dates of visit: January 1987--June 1987 Hanson is interested in modal and philosophical logic, philosophical questions about alternative logics (what Quine has called the question of deviant logics), and semantics of programming languages. While at CSLI, he plans to pursue these interests and study situation theory. Boaz Lazinger Director Division of Computer and Technology National Council for Research and Development Ministry of Science, Jerusalem Dates of visit: July 1986--July 1987 Lazinger is studying a systems approach to natural language understanding, deductive reasoning in document retrieval systems, and NLP interfaces to existing software. Kiyong Lee Department of English Korea University Dates of visit: December 1986--December 1987 Lee is visiting CSLI on a senior research grant from the Korean-American Educational Commission and the Council for International Exchange of Scholars. He hopes to acquaint himself with new developments in situation theory and semantics, and to write an introductory book for Korean readers. While working on some foundational aspects of situation theory, he is very much interested in testing its adequacy in treating some concrete problems, especially those related to negation, quantification, and tense/aspect in Korean. He will be participating in the STASS project while he is here and also continue developing a computationally tractable, functor-driven, phrase structure grammar of natural language by amalgamating a categorial grammar with HPSG. Jan Tore Lonning Mathematics Institute Oslo University Dates of visit: January 1987--August 1987 Lonning is interested in the semantics of natural languages, and has been involved in the study of mass terms and of collective readings of plural noun phrases. He will continue his work on how noun phrases interact with other phenomena such as tense and "intensional contexts" in relation to some of the STASS project's work, and hopes to look at some of the formal properties of LFG and related formalisms in relation to work of the FOG project. Peter Ludlow Dates of visit: September 1986--September 1987 Ludlow's work is primarily centered on developing computationally tractable semantic theories for natural language. In particular, he is interested in developing a tractable semantics for intensional contexts and for quantification. Ronald Nash Dates of visit: January 1987--July 1988 Nash is at CSLI on a postdoctoral fellowship from the Social Sciences and Humanities Research Council of Canada. He is interested in the philosophy of mind and normative psychology, and is particularly interested in the work of CSLI's RATAG and DIA projects with respect to the cognitive theory of emotion on which he has recently worked. He hopes to construct a more formal model while he is here, and will be looking at the various formal models being considered at CSLI. Kasper Osterbye Institute of Electronical Systems, Aalborg University of Aarhus Dates of visit: September 1986--September 1987 Osterbye's recent work has been on programming languages, especially dealing with interactive higher-level debugging. At CSLI, he is participating in the SDL project. Gordon Plotkin Department of Computer Science University of Edinburgh Dates of visit: December 1986--January 1987 Plotkin returned to CSLI to continue his work on building models of situation theory using techniques from domain theory. Ken-ichiro Shirai College of General Education Chukyo University Dates of visit: January 1987--March 1987 Shirai is interested in empirical and theoretical issues in formal semantics with special reference to those concerning Japanese and other oriental languages. Syun Tutiya Associate Professor Department of Philosophy Faculty of Letters Chiba University Dates of visit: November 1986--September 1987 Tutiya is interested in the development of speech act theory within the framework of situation theory and situation semantics. He is also interested in quantification in Japanese, in Frege and the history of logic after him, and has been translating "Situations and Attitudes" into Japanese. While here, he will be joining the STASS project. Arnold M. Zwicky Professor of Linguistics, Ohio State University Visiting Professor of Linguistics, Stanford University Dates of visit: January--March annually, and June 1987--August 1987 Zwicky is conducting research in general linguistic theory (in syntax, morphology, and phonology) aimed at developing explicit proposals for a set of autonomous but interfacing components of grammar. He has a special interest in phenomena---in particular, clitics---that cut across several of the traditional domains. At CSLI he is participating in the GPSG workshop and the phonology workshop. ------------------ NEW CSLI PUBLICATIONS The following reports have recently been published. They may be obtained by writing to Trudy Vizmanos, CSLI, Ventura Hall, Stanford, CA 94305-4115 or publications@csli.stanford.edu. 68. Radical Lexicalism Lauri Karttunen 69. What is Intention? Michael E. Bratman 70. Understanding Computers and Cognition: Four Reviews and a Response Mark Stefik, Editor 71. The Corresponding Continuum Brian Cantwell Smith 72. The Role of Propositional Objects of Belief in Action David J. Israel 73. From Worlds to Situations John Perry 74. Two Replies Jon Barwise ------------------ POSTDOCTORAL FELLOWSHIPS The Center for the Study of Language and Information (CSLI) at Stanford University is currently accepting applications for a small number of one-year postdoctoral fellowships commencing 1 September 1987. The awards are intended for people who have received their Ph.D. degrees since June 1984. Postdoctoral fellows will participate in an integrated program of basic research on situated language---language as used by agents situated in the world to exchange, store, and process information, including both natural and computer languages. Awards are intended for scholars interested in at least one of the following areas of research: discourse semantics and discourse parsing, axiomatizing situation theory, developing a formal theory of representation. For more information about CSLI's research programs and details of postdoctoral fellowship appointments, write to: Dr. Elizabeth Macken, Assistant Director Center for the Study of Language and Information Ventura Hall Stanford University Stanford, California 94305-4115 APPLICATION DEADLINE: 15 FEBRUARY 1987 -------------------------------------------------------------------- Editor's note Selected commentary about Monthly articles or other matters will be published in future issues. Please send correspondence to the Editor of the Monthly at CSLI or by electronic mail to Monthly-Editor@csli.stanford.edu. --------------------------------------------------------------------- - Elizabeth Macken Editor