From Signals to Structured Communication

Please register - it's free!

This symposium brings together researchers from neurobiology, ethology, philosophy, linguistics, economics, game theory, and computer science. To facilitate the emergence of a diverse, unconventional and exciting set of perspectives on communication, it will focus on certain common threads - notably, issues having to do with structure, compositional or other - that run through all these research themes.

Schedule:

How do functional neuronal circuits develop, and once developed how do they retain stable activity throughout the life-time of the animal despite the turn-over of the receptors and channels necessary for signaling? I will address these questions using experiments from developing and adult crustacean stomatogastric ganglion preparations. I will describe the anatomical and physiological development of the stomatogastric ganglion of the lobster, Homarus americanus as well as changes in adult activity patterns that result from long term removal of modulatory inputs. I will discuss the results from a series of modeling studies that suggest that simple activity-dependent rules can allow neurons and networks to develop and respond to perturbations. Because the activity of single neurons depends not on the number of any single ion channel, but on the number and kinds of all its channels, stable electrical excitability requires the coordinate regulation of the conductance densities of all channels. Likewise, network activity requires the coordinate regulation of both synaptic strength and intrinsic cellular excitability. These computational studies address the problem of how cellular and circuit homeostasis occurs in the face of all of the mechanisms for cellular plasticity. The crustacean stomatogastric nervous system produces rhythmic motor patterns that depend both on the presence of bursting neurons and a large number of inhibitory connections. Therefore, the self-assembly and maintenance of stable circuit behavior requires the coordinate tuning of both intrinsic membrane properties and inhibitory synapses.

[back to the schedule]

Song imitation in birds provides good material for studying the basic biology of vocal learning. I will present new experimental and analytic approaches, which now allow us to induce the rapid onset of song imitation in young zebra finches and to track trajectories of vocal change over a 7-week period until a match to a model song is achieved. The bird is trained to peck at a key to trigger song playbacks from a tiny speaker placed inside a plastic models of a 'tutor birds'. This training induced rapid song imitation under tight experimental control, but in an environment that can nevertheless simulate the natural social and acoustic arena. Our system records the entire vocal ontogeny of the bird, extract acoustic features and track vocal changes in real-time during vocal learning. Using these techniques we found that exposure to a model song induces the prompt generation of repeated structured sounds (prototypes) followed by a slow transition from repetitive to serial delivery of syllables. Those stages are reminiscent of early reduplicated (canonical) babbling in human infants, and the latter transition to variegated babbling. In the birds, we could track this transition to great details, and revealed two surprising phenomena: (i) Imitations of dissimilar sounds can emerge from successive renditions of the same prototype, and (ii) developmental trajectories for some sounds followed paths of increasing acoustic mismatch until an abrupt correction occurred by period doubling. These dynamics are likely to reflect underlying neural and articulatory constraints on the production and imitation of sounds. I will present a revised view of the vocal imitation process, which attempt to capture the array of operations that the bird can perform to alter sounds. I will argue that the scope of such operations can be rather wide, affecting more than a single sound, and therefore stabilizing the global dynamics of song imitation is a non-trivial computational task. This revised view of vocal imitation will be examined against the traditional view of vocal imitation as an approximation process, guided by selection and/or by instructions.

[back to the schedule]

Let's assume that for each (meaningful) complex expression E of a natural language, E has a "core" meaning that is determined by the (semantic properties of the) parts of E and how those parts are arranged in E. This raises specific questions about how core meanings are determined for particular constructions, but also a more general question: is there some common explanation, perhaps in terms of basic semantic properties, for why the meanings of complex expressions are determined by the meanings of constituent expressions? One traditional way of representing meanings treats each complex expression as divisible into (i) a component associated with some function, and (ii) a component associated with some argument in the domain of that function; the meaning of a complex expression is then taken to be the value of the relevant function given the relevant argument. From this perspective, associated with Frege and Montague, natural language meanings are compositional because functions determine values given arguments; and the semantic correlate of combining expressions, to form a more complex meaningful expression, is simply function-application.

An alternative perspective is presupposed, often without explicit recognition, by a growing body of work inspired by Davidson's treatment of sentences involving various adverbial phrases-as in 'Shem poked Shaun softly with a pencil'. On these views, the semantic correlate of combining expressions is much more like predicate-conjunction or perhaps intersection. The idea is that 'poked', 'softly' and 'with a pencil' represent properties of events; the subject 'Shem' and object 'Shaun' represent, respectively, certain individuals as the agent and patient of some event; the whole sentence means that something was (a) done by Shem, and (b) an event of poking, and (c) done to Shem, and (d) done softly, and (e) done with a pencil. I think there is much to be said for this view: it provides a relatively simple way of capturing a range of otherwise puzzling facts about some specific natural language constructions; and it provides an attractively simple conception of semantic compositionality, at least for certain core aspects of meaning. In my talk, I'll illustrate these virtues of the "conjunctivist" approach, focussing mainly on a cluster of causal constructions-like 'Pat boiled the soup', in English, and more interesting related constructions found in other languages.

• A paper, Function and Concatenation, is forthcoming in a volume of essays, "Logical Form" (edited by G. Preyer, Oxford: OUP).

Human beings communicate using linguistic systems with a complexity of syntactic structure that is apparently without parallel elsewhere in the natural world. Over the past few decades, linguists have put forward many hypothesized linguistic universals, i.e., principles of design that, despite the diversity of human languages, hold for every language. These include several intriguing and apparently quite deep principles governing things like rule interaction, phrase structure, constituent order, command relations, and phonological representation.

Chomsky has been suggesting for over thirty years that such universals are contingent facts due to genetically transmitted quirks of human brain structure. This conjecture can hardly be said to have picked up any significant support from adjacent fields such as psychology, neurophysiology, or genetics. My own view (admittedly, equally speculative) is that the universals linguists have proposed divide into three categories: first, those that are just false (the most common fate, about which I will say little here); second, those that are trivial (this happens in more cases than one might think, and I will cite a few examples); and third, those that, though valid and not exactly trivial, might best be compared with principles in fields such as economics (as construed by economists like von Mises, who took it as obvious that economics was a discipline of a priori analysis) and praxiology (the philosophical study of `efficient action' due to the Polish philosopher Kotarbinski).

I will review (informally, without a great deal of technical analysis) several supposed `linguistic universals' that do not seem to be specific to natural languages, or to language at all. Rather, they embody principles that are simply more likely to be present than absent any `good' cognitive, computational, or other complex system (where `good' is used here in the sense whose analysis is praxiology's central task). I will try to plant the seeds of a suspicion that the etiology of human linguistic structure lies not in species-specific neurobiology, but in principles that would characterize the grammatical systems of any intelligent entities -- features that could not be very much different in a system as complex as the communicative needs of human beings seem to demand.

[back to the schedule]

In this talk, I introduce and then criticize a recently popular approach to the semantics of interrogatives. According to what I will call the "set-of-answers" strategy, an interrogative denotes a question, where a question just is a set of answers. This is the approach I will argue against.

My preference is for a more traditional story — the "force and radical" approach — according to which interrogatives are composed of a sentence radical and a force indicator. The force indicator contributes interrogatival force to the sentence type, while the radical, given a context, denotes either a proposition, or a propositional function — depending upon whether the sentence is a yes-no or wh- interrogative.

I argue that the latter view is preferable because it goes some way towards explaining how interrogatives are used in communication, in "askings".

Background Reading:

• Hamblin, C. 1973: Questions in Montague English. Foundations of Language, 10, 41-53.
• Higginbotham, J. 1993. Interrogatives. In K. Hale and J. Keyser (eds), The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger. Cambridge: The MIT Press.
• Higginbotham, J. and May, R. 1981: Questions, quantifiers and crossing. The Linguistic Review, 1, 41-80.
• Karttunen, L. 1977: The syntax and semantics of questions. Linguistics and Philosophy, 1, 3-44.

I will try to demonstrate what we, economists, can say about linguistics by presenting two short investigations in which we use economic reasoning to address linguistic issues. The first discussion will be an attempt to derive properties of binary relations from considerations of functionality. The second discussion will introduce strategic considerations to explain pragmatic phenomena in debates.

Reading:

• Ariel Rubinstein, Economics and Language, Cambridge University Press, 2000.

Language is the most interesting biological trait that evolved in the last several hundred million years. It is not only an evolutionary invention that radically changed the performance of one species and the appearance of the planet, but also the mode of evolution. Since evolution is a mathematical theory, I will discuss mathematical models for the evolutionary dynamics of animal communication and human language. I will analyze how natural selection can lead to grammatical communication systems. I will study the population dynamics of language acquisition and calculate the restrictions that must be imposed by universal grammar for the emergence and evolutionary stability of linguistic coherence.

Further reading:

• Nowak M A, Krakauer D (1999) The evolution of language, Proc Natl Acad Sci USA, 96, 8028-8033
• Nowak M A, Plotkin J B, Jansen V A A (2000) The evolution of syntactic communication, Nature, 404, 495-498
• Nowak M A, Komarova N L, Niyogi P (2001) Evolution of universal grammar, Science, 291, 114-118

Say that a group of people face a "coordination problem" in that each person wants to participate only if many other people do (for example, group hunting, political demonstrations, buying a Macintosh computer). People most often "solve" coordination problems through communication. However, game theory shows that to solve a coordination problem, it is not enough for a message simply to be received by each person; each person must know that each other person has received it, each person must know that each other person knows that each other person has received it, and so forth. In other words, it is not enough for the message to simply be "known"; it must be "common knowledge" (each person knows that other people know it, and so forth). This argument is well known (the philosopher David Lewis made it in 1969) but its empirical relevance has not been fully realized. I argue that the concept of common knowledge is extremely helpful in looking at "public rituals," such as public ceremonies, rallies, and media events. I consider (among other things) the Super Bowl, political techniques during the French Revoution, data from US network television advertising, the question of how a group's social network affects its ability to coordinate, and the "instability" of Bentham's panopticon prison design. Most work on the "theory of mind" (a person's ability to understand the mental states of others) has focused on how meaning is created and understood in two-person interactions. My argument suggests that "theory of mind" is an essential aspect of how large-scale social institutions such as public rituals "work."

The talk is based on Rational Ritual: Culture, Coordination, and Common Knowledge , forthcoming Princeton UP, May 2001.

[back to the schedule]

The talk will present a program for inducing structured lexical entries for words, similar to those postulated in linguistics, from large samples of language. Some principal ideas are:

1. The a lexical entry is a complex term which is built up from relations symbols and variables using operations such as conjunction and lambda binding.
2. Part of the problem in learning a lexicon is to construct relation symbols which are used in many lexical entries.
3. Given a constrained family of possible lexicons, learning can be achieved by numerical optimization using the EM algorithm.

• Inducing a Semantically Annotated Lexicon via EM-Based Clustering. With Stefan Riezler, Detlef Prescher, Glenn Carroll, and Franz Beil. In 37th Annual Meeting of the ACL, 1999, Maryland. (.ps/ .ps.gz)
• Inside-Outside Estimation of a Lexicalized PCFG for German. With Franz Beil, Glenn Carroll, Detlef Prescher, and Stefan Riezler. In 37th Annual Meeting of the ACL, 1999, Maryland. (.ps/ .ps.gz)
• Two-dimensional clusters in grammatical relations. Paper presented at IJCAI Lexicon workshop, Stanford, 1995.

This talk focuses on grounded communication: verbal interactions about a situation experienced through a sensori-motor apparatus. This requires an integration of many competences: speech, vision, behavior, language, and learning. I will focus on open-ended communication, in which neither the meaning nor the forms used to express meaning are fixed in advance but arise out of the interactions between the agents and the environment and are progressively negotiated. Evolutionary language games are presented as a useful framework to organise the required learning and language negotiation.

I will then focus specifically on some experiments. First some language games for physical robots (including the AIBO) that focus on the acquisition of single words, for objects, properties of objects, body parts, actions, movements, etc. These experiments show how a strong structural coupling between processes generating meaning and processing associating form with meaning can explain the co-evolution of language and meaning. Finally I will sketch how we seek to study the origins of grammar through robotic experiments.

Some publications (all downloadable from http://www.csl.sony.fr/General/Publications/index.php3):

• Steels, Luc (2000) Language as a Complex Adaptive System. In Schoenauer, M., editor, Proceedings of PPSN VI, Lecture Notes in Computer Science, Berlin, Germany, September 2000. Springer-Verlag.
• Steels, Luc (2000) The Emergence of Grammar in Communicating Autonomous Robotic Agents. In: Horn, W., editor, Proceedings of ECAI 2000, pages 764-769, Amsterdam, August 2000. IOS Publishing.
• Steels, L. and Kaplan, F. (1999) Situated grounded word semantics. In Dean, T., editor, Proceedings of the Sixteenth International Joint Conference on Artificial Intelligence IJCAI'99, pages 862-867, San Francisco, CA., 1999. Morgan Kaufmann Publishers.
• Steels, L. (1998) The origins of syntax in visually grounded robotic agents. Artificial Intelligence, 103:1-24 1998.