L. Steels and P. Vogt, "Grounding adaptive language games in robotic agents," in Proc. 4th Eur. Conf. Artificial Life, 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....two way mappings between meanings, signals, and meanings again. That is, speakers must evolve a mapping between private meanings and public signals, and hearers must evolve a complementary mapping such that private meanings can be recovered, more or less reliably, from public signals (see, e.g. [31, 43]) The fact that under the right circumstances such mappings can arise has now been safely established. Have such findings in ALife been picked up on by linguistics The short answer is no. An examination of recent Introduction to Linguistics courses and textbooks shows that ALife does not ....

Steels, L., and Vogt, P. (1997). Grounding adaptive language games in robotic agents. In P. Husbands and I. Harvey (eds.), Proceedings of the Fourth 34 European Conference on Artificial Life (ECAL '97). Cambridge, MA: MIT Press/Bradford Books, pp. 474-482.

....to interact with an actual instance of a service. The grounding provides the concrete specification of these details. This use of the term grounding , particularly within the context of a semantic Web is somewhat unfortunate. DAML S grounding has little to do with semantic grounding [27, 35, 54] 3.1 DAML S Processes The DAML S process model is intended to provide a basis for specifying the behavior of a wide range of services, and draws on work in several fields. This includes work in AI on standardization of planning languages [25] work in programming languages and distributed ....

Luc Steels and Paul Vogt. Grounding adaptive language games in robotic agents. In C. Husbands and I. Harvey, editors, Proceedings of the Fourth European Conference on Artificial Life (ECAL97), London, 1997. MIT Press.

....Method The main idea is to consider other agents as personal entities which you can rely on or not. Reliability is expressed through a trust value with which each agent labels its neighbors. The trust value is initially computed through interaction, following a proactive playing agents procedure [15]. Each agent ask the other agents about a list of known items and gathers their opinion on such items. The agents ask the queried agents about their opinion on the item that the user either loves or hates . According to similarity between the opinion provided and their own, agents are able to ....

....runs. Such an assumption is reasonable taking into account that most of the multi agent system platforms currently developed and FIPA [4] compliant provide such a service. Then we elaborate the initial trust of agents in the world using a procedure that we have called playing agents following [15]. The querying agent asks other agents in the world (enquired agents) one by one about, an item of the training set. We can apply this procedure because each agent has been generated from the same training set, so they are able to provide answers about items belonging to such set. Then, the agent ....

L. Steels and P. Vogt. Grounding adaptive language games in robotic agents. In Proceedings of the Fourth European Conference on Artificial Life, pages 473--484, 1997.

....a i j is an agent identifier and t i,i j is a number between [0,1] that represents the truth value that a i has on a i j . Initially the contact list is empty. Thus, agents contact other agents in the world and elaborate the initial trust using aprocedurethatwehavecalledplaying agents following [7]. The agent asks for the opinion about the items that the user loves or hates to the enquired agent.The answer consist of a quantitative value v i,j , between 0 and 1, that represents the degree of interest the agent has on the product (0 hates, 1 loves) and is computed as follows: v i,j = # ....

L. Steels and P. Vogt. Grounding adaptive language games in robotic agents. In Proceedings of ECAL'97,pages 473--484, 1997.

.... learning of acoustic vocabularies and higher level language structures from speech recordings tran scribed at only the semantic level [17] Physical grounding of concepts has been explored in the context of robotics as an alternative to the symbol processing view of artificial intelligence [18] [19]. The model presented in this paper departs from previous work in language learning in that both words and their semantics are acquired from sensor input without any human assisted transcription or labeling of data. IV. CELL: CROsS CHANNEL EARLY LEXICAL LEARNING To explore issues of grounded ....

L. Steels and P. Vogt, "Grounding adaptive language games in robotic agents," in Proceedings of the Jth European Conference on Artificial Life, 1997.

....are given, while drinks are Figure la represents a communication system based on grounded signals. Communication relies on simple indexical associations between objects and signals. This situation refers to the models of the evolution of language that only focus on lexicon emergence (e.g. 4] [27]) Communication signals, that are directly grounded in the organisms environment, do not have any symbolic or syntactic properties. Figure lb shows a system based on grounded symbols (e.g. 26] 2] In the top layer there are links between symbols, and references between symbols and objects. ....

....computation has recently been applied to studying the emergence and auto organisation of communication lexicons. Some models have been used for the simulation of the emergence of simple lexicons in populations of simulated organisms (e.g. 23] 4] 19] in small communities of robots [27], or in on line Internet agents [26] In these studies, organisms evolve shared lexicons for describing entities and relations of the environment. These models, that focus on lexicon emergence, do not make any explicit and direct reference to the role of syntax in language origin. Their aim is to ....

[Article contains additional citation context not shown here]

Steels L. & Vogt P. (1997). Grounding adaptive language games in robotic agents. In P. Husband & I. Harvey (eds). Proceedings of the Fourth European Conference on Artificial Life, London: MIT Press, 474-482

....Although none of the agents has a global view of the language, and none of the agents does an explicit optimization, a coherent system of vowels emerges that happens to be optimized for acoustic distinct iveness. The results presented here fit in and confirm the theory of Luc Steels [Steels 1995, 1997, 1998] that views languages as a complex dynamic system and the origins of language as the result of self organization and cultural evolu tion. 1 Introduction Language is considered to be important for the understanding of intelligence. Although animals are often quite capable of behavior ....

....that is characteristic of humans. This more abstract kind of intelligence is of a symbolic nature, and therefore associated with language. Understanding the nature and the origin of language is therefore of crucial importance to the understanding of the nature and origin of human intelligence [Steels 1995, 1997, 1998] 1.1 The origins of language Some scholars have assumed that the human faculty for language is innate and genetically determined in a very specific way [Chomsky 1980; Pinker Bloom 1990] It is obviously true that humans have a unique capability for learning and using language. If a ....

[Article contains additional citation context not shown here]

Luc Steels and Paul Vogt (1997) Grounding adaptive language games in robotic agents. In: Husbands, Phil & Harvey, Inman (eds.) Proceedings of the Fourth European Conference on Artificial Life, Cambridge (MS): MIT Press, pp. 474-482, 1997

....robot s lexicon agrees with the others. The naming games require that each robot have a similar and useful method of categorising the environment; this is provided by having each robot play a discrimination game with itself where sets of feature trees compete to best categorise the environment [4, 9]. Using these two games, both robotic and simulated agents can successfully create coordinated lexicons using language games [8, 9] Rather than self organising a language from scratch, it is also possible for one robot to learn the language from another which already posses the full vocabulary. ....

.... the environment; this is provided by having each robot play a discrimination game with itself where sets of feature trees compete to best categorise the environment [4, 9] Using these two games, both robotic and simulated agents can successfully create coordinated lexicons using language games [8, 9]. Rather than self organising a language from scratch, it is also possible for one robot to learn the language from another which already posses the full vocabulary. This imitative learning method has been successfully used to coordinate a vocabulary between a human teacher and a single robot ....

Luc Steels and Paul Vogt. Grounding adaptive language games in robotic agents. In I. Harvey and P. Husbands, editors, Proceedings of the 4th European Conference on Articial Life, 1997.

....more detail their consequences. In fact, interest in studying language s origin and evolution using computer simulations has increased considerably in the last few years. Some researchers have explored the evolution of language using models that describe language as a set of signal meaning pairs (Steels, 1996; 1997; Oliphant and Batali, 1996; Di Paolo, 1997) These authors use populations of agents that play a communication game via the exchange of signals. At each time step a signal is selected according to a matrix that assigns a probability value to each signal in correspondence to each meaning. With ....

....of signals. At each time step a signal is selected according to a matrix that assigns a probability value to each signal in correspondence to each meaning. With this simulation approach it is possible to study the different conditions that allow the evolutionary emergence of shared vocabularies. Steels and Vogt (1997) have experimented with adaptive language games in pairs of physically embodied robot agents. The language game includes six steps: establishing contact with the other robot, identifying the communication topic, categorizing the surrounding world, speaker s encoding of the communicative signal, ....

Steels, L., & Vogt, P. (1997). Grounding adaptive language games in robotic agents. In P. Husbands & I. Harvey (eds.), Proceedings of the Fourth European Conference on Artificial Life, Cambridge, MA: MIT Press.

....in press) has contributed to the rebirth of interest in the origin and evolution of language. Computational models can directly simulate the evolution of communication and the emergence of language in populations of interacting organisms (Cangelosi Parisi, in press; Dessalles Ghadakpour, 2000; Steels, 1997). Various simulation approaches are used such as communication between rule based agents (Kirby, 1999) recurrent neural networks (Batali, 1994; Ellefson Christiansen, 2000) robotics (Kaplan, 2000; Steels Vogt, 1997) and internet agents (Steels Kaplan, 1999) Artificial Life Neural ....

.... interacting organisms (Cangelosi Parisi, in press; Dessalles Ghadakpour, 2000; Steels, 1997) Various simulation approaches are used such as communication between rule based agents (Kirby, 1999) recurrent neural networks (Batali, 1994; Ellefson Christiansen, 2000) robotics (Kaplan, 2000; Steels Vogt, 1997), and internet agents (Steels Kaplan, 1999) Artificial Life Neural Networks (ALNN) are neural networks controlling the behavior of organisms that live in an environment and are members of evolving populations of organisms. ALNN models have been used to simulate the evolution of language ....

Steels, L., & Vogt, P. (1997). Grounding adaptive language games in robotic agents. In P. Husband & I. Harvey (Eds). Proceedings of the Fourth European Conference on Artificial Life (pp. 474-482), London: MIT Press.

....In artificial intelligence, the problem arises in many di#erent settings. A number of studies have emerged where linguistic agents interact with each other in simulated worlds and one studies whether coherent or coordinated communication ultimately emerges (see, for example, Steels 1996, Steels Vogt 1997, Steels Kaplan 1998, Oliphant Batali 1996, Oliphant 1997, Briscoe 2000, Kirby 1999] Much of this kind of research employs the simulation methodology of Artificial Life. In this paper, we create a mathematical framework for these kinds of problems and derive a number of analytic results. We ....

Steels, L. & Vogt, P. (1997) Grounding adaptive language games in robotic agents. In: Proceedings of the Fourth European Conference on Artificial Life, (Husbands, P. & Harvey, I. eds). Cambridge, MA: MIT Press.

.... rasa learning of acoustic vocabularies and higher level language structures from speech recordings transcribed at only the semantic level [11] Physical grounding of concepts has been explored in the context of robotics as an alternative to the symbol processing view of artificial intelligence [3, 26]. The model presented this paper departs from previous work in that both words and their semantics are acquired from sensor input, and the goal of the system is to acquire spoken human language. The idea of learning from evidence across multiple modalities used in CELL has been explored in the ....

L. Steels and P. Vogt. Grounding adaptive language games in robotic agents. In Proceedings of the 4th European Conference on Artificial Life, 1997.

....Contribution to robotics The work presented in this paper does not follow directly from any previous studies of robotics. Indeed, little work has yet been done in teaching a physical robot a synthetic form of communication. Closest works are those of Yanco and Stein [58] and Vogt and Steels [59], in which a robot learns a lexicon to describe sets of actions and perceptions respectively. The present study differs from those works in two main aspects: 1. The learning and behavioral capacities of the robots result from a single connectionist architecture, which has general ability for ....

....occurs in a highly noisy and sensory rich environment. The experiments simulated only very partially some of this complexity, but as such it was already an important step compared to previous work in the domain. In the robotics works on grounding of a lexicon, which were cited earlier ( 58] and [59]) the spatial and temporal variability of the word meaning pattern is almost inexistent. Learning occurs between only the relevant (to the learning) sensory channels, which are the motor and radio channels [58] and the infra red, light and radio channels in [59] This reduces strongly the ....

[Article contains additional citation context not shown here]

L. Steels and P. Vogt. Grounding adaptive language games in robotic agents. In P. Husbands and I. Harvey, editors, Proceedings of the Fourth European Conference on Artificial Life, ECAL97, pages 473--484. MIT Press, 1997.

....in the game change their rules (i.e. their word object associations) to be more successful in future games. The naming game has been investigated through computer simulations as well as in robotic experiments in which the meanings are grounded in the sensori motor behavior of the robots [9]. More formally, there is a set of of size where each agent has contact with a set of of size , In the present S 0 1 2 1 1 1 2 2 2 1 2 1 1 2 2 1 1 2 2 1 2 1 1 1 1 1 a a a a a s s s s h t a;t a;t s;t s;t h;t s;t h;t h;t o h;t h;t s;t L 2 ae O 2 2 N 2 N 2 A L O 2 ....

Steels, L. and P. Vogt (1997) Grounding adaptive language games in robotic agents. Submitted to ECAL-97, Brighton.

....between radio signals and object features were learned in less than 30 teachings, which correspond to about 15 to 30 minutes of physical experiment. Similar experiments on grounding radio signals into robots sensor capabilities were carried out previously by Yanco Stein [44] and Steels Vogt [38], who used respectively reinforcement learning and evolutionary techniques. Their experiments showed that a vocabulary of 5 and 3 words was learned after 900 and 60 training examples respectively. Our method then seems faster at learning a larger or similar vocabulary. In addition, it is more ....

....general than the above mentioned methods, as we were not restricted in the sensor stimuli the robots could talk about. In Yanco Stein s work [44] the vocabulary consisted only of the robot s actions because the learning algorithm was based on an action selection mechanisms. In Steels Vogt [38], the vocabulary concerned only the robots external perceptions as these were the only perceptions they could share. By contrast, the mutual following strategy we use in our work allows the two agents to share a common context of both external (face the same direction) and internal perceptions ....

L. Steels & P. Vogt, (1997) `Grounding adaptive language games in robotic agents', Proceedings of the Fourth European Conference on Artificial Life, pp.473-484, Brighton, U.K., MIT Press/Bradford Books.

....1996; Bullock, 1997; de Bourcier and Wheeler, 1997; Di Paolo, 1997; Werner and Todd, 1997; Noble, 1998) 2. Those which suggest that cultural transmission between generations alone is capable of developing and refining entirely learned communication systems (e.g. Hutchins and Hazelhurst, 1995; Steels and Vogt, 1997; Batali, 1998; Livingstone and Fyfe, 1999; Batali, in press; Hurford, in press; Kirby, in prep; Kirby, in pressa; Kirby, in pressb; Oliphant, in press) 3. Those which suggest that positive interactions between genetic and cultural transmission are capable of developing and refining ....

Steels, L. and P. Vogt (1997). Grounding adaptive language games in robotic agents. In P. Husbands and I. Harvey (Eds.), Fourth European Conference on Artificial Life, pp. 474--482. Cambridge, MA: MIT Press.

..... establishing a joint attention [between speaker and listener] that creates a meaningful social setting necessary for the development of language [21] Other works implemented such attentional mechanisms as processes distinct from the learning mechanisms, e.g. Steels Vogt s pointing strategy [50] and Yanco Stein s action selection mechanism [57] By contrast, we develop a single cognitive architecture which enables both associative learning, selective attention from parsing of continuous sensory information, and the creation of a mutual binding between the two agents by means of mutual ....

....for achieving this basic social interaction, this proved to be a powerful external (behavioural instead of cognitive) attentional mechanism. Similar experiments on grounding radio signals in robots sensor capabilities were carried out previously by Yanco Stein [57] and Steels Vogt [50] who used respectively reinforcement learning and evolutionary techniques. We showed in [7] that our model is more efficient and faster than the above models to learn a vocabulary of the same size and type. Moreover, our model 35 (the associative memory and the following scenario) has the ....

[Article contains additional citation context not shown here]

Steels L. & Vogt P. (1997), `Grounding adaptive language games in robotic agents', Proceedings of the Fourth European Conference on Artificial Life, pp. 473-484, Brighton, U.K., July 97.

....by the robots themselves and not designed and programmed in by an external observer. They must also be grounded in the sensori motor experiences of the robot as opposed to being disembodied, with the input given by a human experimenter and the output again interpreted by the human observer. (Steels and Vogt 1997: 474) 45 Cangelosi and Parisi (1998) for example, have in computer simulations studied the evolution of a language in a population of ALife agents that live in a simulated world containing edible and poisonous mushrooms, of which they have to find the former but avoid the latter in order ....

....allowed them to communicate about the world they lived in, i.e. the approach and avoidance of the mushrooms they encountered. Experiments on the development of language and meaning in groups of robotic agents through adaptive language games have been carried out by Steels (1998; see also Steels and Vogt 1997, Steels and Kaplan 1999) In the experimental setup used by Steels and Vogt (1997) a number of mobile robots moved around in a physical environment of limited size, containing some additional objects. The robots acquired a common vocabulary of word meaning pairs (where the meaning of a word is ....

[Article contains additional citation context not shown here]

Steels, Luc and Vogt, Paul (1997). Grounding adaptive language games in robotic agents. In Husbands, P. and Harvey, I., editors, Fourth European Conference on Artificial Life, pages 474-482, Cambridge, MA. MIT Press.

....learning of irregular languages, such as human languages, which requires complex syntax, the understanding of ambiguity and the integration of exceptions to grammatical rules. Synthesis. Little work has yet been done in teaching a physical robot a synthetic form of communication (e.g. YS93] SV97] Our work differs from those studies in several aspects: 1) The learning and behavioural capacities of the robots result from a single cognitive architecture; it is a connectionist model which has general ability for extracting spatio temporal regularities in a dynamic environment [BH99] 2) ....

L. Steels and P. Vogt. Grounding adaptive language games in robotic agents. In P. Husbands and I. Harvey, editors, Proceedings of the Fourth European Conference on Artificial Life, ECAL97, pages 473--484. MIT Press, 1997.

....of elementary signals and a general learning mechanism can result in the development of coordination. 1 Introduction The work reported in this paper is part of the Origins of Language programme, in which principles are investigated that may explain how languages develop, see e.g. Steels, 1996, Steels and Vogt, 1997, De Jong and Vogt, 1998] From an evolutionary point of view, the development of communication may be explained by the increased potential for coordinated action that comes with it. Since these two phenomena are interrelated, we study them in combination. Most scientific disciplines start out by ....

....situations, an outside observer can observe indications that communication or coordination takes place. We use the term measures for these observable indications. Measures are defined quantitatively and can thus be determined exactly. For examples of measures used in this research programme, see [Steels, 1997], Kaplan, 1998] Measures need to be defined specific to each experiment. The difficulties with defining communication result from the wish to find general operational criteria that determine the presence of communication in all its appearances. I suggest that these problems can be circumvented ....

Steels, L. and Vogt, P. (1997). Grounding adaptive language games in robotic agents. In Husbands, C. and Harvey, I., editors, Proceedings of the Fourth European Conference on Artificial Life, Cambridge MA and London. MIT Press.

....a result the obstacle avoidance module returns the direction furthest away from all obstacles, the behavior system for the control of robot can use this to steer clear from all obstacles. This method has also been used by Horswill in Polly [10] 4. 5 Seeing other robots In some experiments (e.g. [25]) a robot needs to see other robots. Since robots are the only thing moving in the ecosystem, the Find other robots module only needs to look for motion in the image not caused by the observer itself (this implies that robots have to move in order to be seen) One could use optical flow analysis ....

Luc Steels and Paul Vogt, Grounding adaptive language games in robotic agents. Submitted to the 4th European Conference on Artificial Life, Brighton, UK, 1997.

....choose their actions and at the same time use the feedback on their behavior to adapt their lexicon. Several authors describe the development of communication where agents are either teacher or learner, possibly alternating. Some examples are [Yanco and Stein, 1993] Billard and Hayes, 1997] and [Steels and Vogt, 1997], all of which are interesting in that real robots are used to investigate the development of communication. An example of work where agents are not appointed different roles is [MacLennan, 1991] An investigation of the issue of altruism, i.e. how can language evolve if only receiving, and not ....

Steels, L. and Vogt, P. (1997). Grounding adaptive language games in robotic agents. In Husbands, C. and Harvey, I., editors, Proceedings of the Fourth European Conference on Artificial Life, Cambridge MA and London. MIT Press.

....review of it) In our approach, we use a neural network in order to address the problem of how to attach the meaning to a word, e.g. how can one agent s perception and description of the world make sense for another agent that is physically different. In opposition to related robotics studies ([9], 16] 12] the learning of the language is not implicit in the mechanism and the learning architecture which we use. A simple following strategy allows implicitly the agents to share the same physical context. Learning to agree on a common vocabulary results from a general capability to form ....

L. Steels & P. Vogt, (1997) `Grounding adaptive language games in robotic agents', Proceedings of the fourth european conference on Artificial Life, Brighton, U.K., July 97.

....mechanism, coordination signals may be sufficient for the development of coordination. 1 Introduction The work reported in this paper is part of Luc Steels programme of research into the origins of language, in which principles are investigated that may explain how languages develop, see e.g. [16, 18, 5]. From an evolutionary point of view, the development of communication may be explained by the increased potential for coordinated action that comes with it. Since these two phenomena are interrelated, we study them in combination. Ultimately, our aim is to let agents autonomously develop ....

L. Steels and P. Vogt. Grounding adaptive language games in robotic agents. In C. Husbands and I. Harvey, editors, Proceedings of the Fourth European Conference on Artificial Life, Cambridge MA and London, 1997. The MIT Press.

....concepts are diverse, as may be judged from the examples in the paper. Situation concepts are especially suited to serve as a basis for communication. In the model of language evolution investigated here, concept formation interacts with a process linking concepts to words or signals, see [ Steels, 1997; Steels and Vogt, 1997 ] In literature on the evolution of language, communication often corresponds directly to actions, which limits communication to the instruction of other agents. Examples include [ Werner and Dyer, 1991 ] where sounds tell agents in a simulation how to move to the ....

....are diverse, as may be judged from the examples in the paper. Situation concepts are especially suited to serve as a basis for communication. In the model of language evolution investigated here, concept formation interacts with a process linking concepts to words or signals, see [ Steels, 1997; Steels and Vogt, 1997 ] In literature on the evolution of language, communication often corresponds directly to actions, which limits communication to the instruction of other agents. Examples include [ Werner and Dyer, 1991 ] where sounds tell agents in a simulation how to move to the emitter of the sound, Yanco ....

Luc Steels and Paul Vogt. Grounding adaptive language games in robotic agents. In C. Husbands and I. Harvey, editors, Proceedings of the Fourth European Conference on Artificial Life, Cambridge MA and London, 1997. The MIT Press.

....section. 3 Adaptive language games To illustrate how a set of anchored symbols can be developed from the bottomup, an experiment is presented in which two mobile LEGO robots bootstrapped a symbolic communication system. To achieve this, the robots engaged in a series of adaptive language games [14,17] in which they tried to communicate the form that stands for an object and adapt their internal structures in order to improve their performance on later occasions. Various types of language games have been implemented such as observational games, guessing games and sel sh games, which di er from ....

....the amount of sensory data as if it were a sieve. 3.2.1 Sensing A guessing game started when both robots were standing close to each other with their backs facing each other. During the sensing phase, the robots In the original implementation, the robots aligned themselves autonomously [17], but to speed up the experiments, the robots were placed by hand for this experiment. 50 100 150 200 250 0 10 20 30 40 50 60 Intensity time (PDL cycles) s0 s1 s2 s3 0 50 100 150 200 250 0 10 20 30 40 50 s0 s1 s2 s3 Fig. 3. The sensing of robot A (left) and robot B (right) ....

L. Steels and P. Vogt. Grounding adaptive language games in robotic agents. In C. Husbands and I. Harvey, editors, Proceedings of the Fourth European Conference on Arti cial Life, Cambridge Ma. and London, 1997. MIT Press.

....given the enormous complexity of the cognitive processing required for grounded language, even for handling single words. So it is at least necessary to do computer simulations. Here we go one step further and do experiments with autonomous mobile robots, in line of similar work reported in ([Steels and Vogt, 1997], Billard et al. 1998] Vogt, 2000] Robotic experiments are motivated as follows: 1. They force us to make every claim or hypothesis about assumed internal structures and processes very concrete and so it is clear how the theoretical assumptions have been operationalised. 2. We can use ....

Steels, L. and Vogt, P. (1997). Grounding adaptive language games in robotic agents. In Harvey, I. and Husbands, P., editors, Proceedings of the 4th European Conference on Artificial Life, Cambridge, MA. The MIT Press. 32

....approach see e.g. 50] The main question that still needs to be answered is how these robots may increase their complexity, so that they can be classified as cognitive agents. We think that building up a language based on grounded perception is necessary to increase cognitive intelligence [42][51]. Recently more scientists investigated the origination of language on robotic agents. One example of such a research was the one done by Holly Yanco at the AI Lab at MIT [57] She used written commands for human robot communication with one robot. This robot first learned to associate these ....

.... of language games [56] The purpose of the experiment is twofold: 1) To show how a group of distributed agents can build a coherent vocabulary by means of adaptive language games, and (2) to show how autonomous agents may generate distinctions to discriminate between objects in their environment [51]. The basic assumption we make is that the agents already have the mechanisms to communicate, i.e. they know how to communicate, although initially they have no language. The task was only to implement the language games. Language, however, is based on meaning. In the theory described by Steels ....

[Article contains additional citation context not shown here]

Steels, L. and P. Vogt (1997) Grounding adaptive language games in robotic agents. Submitted to ECAL97.

....that can help the robots in performing this task. The paper reports preliminary results of this experiment. In the past few years, research at the Artificial Intelligence Laboratory of the Free University in Brussels focuses on the origins and evolution of language and meaning, for an overview see (Steels, 1997). The approach is that both language and meaning are viewed as complex dynamical adaptive systems as proposed by Luc Steels (1996a, 1996b) Language, for instance, is complex because there is a huge expressibility in the system. Furthermore, there is no language user that has complete competence ....

.... are similar to those used by (Werner and Dyer, 1991, McLennan, 1991, Kirby and Hurford, 1997) and (Oliphant, 1996) An overview of research in this field can be found in (Hurford et al. 1998) In previous experiments a lexicon has been grounded by mobile robots on perceptual sensory information (Steels and Vogt, 1997, Vogt, 1998a) In these experiments it has been observed that the feedback systems used were not sufficient for learning a completely coherent lexicon, so that the performance in the communication stayed poor. Furthermore, other work (De Jong, 1997) suggested that it might be useful to ground a ....

[Article contains additional citation context not shown here]

Steels, L. and Vogt, P. (1997). Grounding adaptive language games in robotic agents. In Husbands, C.

....Vrije Universiteit Brussel, Artificial Intelligence Laboratory paul arti.vub.ac.be 1 Introduction In the past few years, research at the Artificial Intelligence Laboratory of the Free University in Brussels is focused on the origins and evolution of language and meaning. In previous experiments [2] a lexicon has been grounded by mobile robots on perceptual sensory information. In this paper work is presented on an experiment where two robots follow each other. While following each other, the robots try to develop a lexicon about the movements they perform through the motor commands they ....

....backward. They do so by playing a series of what we could call follow me games. The robots categorize segments of the time series of motor commands using the so called method of delays as proposed by [1] The creation and learning part of the categorization and the lexicon formation is based on [2]. In a follow me game, one of the robots (the speaker) leads a second robot (the hearer) Every time the speaker changes its direction, a language game is played. In a language game, the speaker tries to communicate its movement. It does so by first segmenting the time series of the motor ....

L. Steels and P. Vogt. Grounding adaptive language games in robotic agents. In C. Husbands and I. Harvey, editors, Proceedings of the Fourth European Conference on Artificial Life, Cambridge Ma. and London, 1997. MIT Press.

....point of view it would be more interesting if the robots could ground the language system themselves. In the past few years, research at the Artificial Intelligence Laboratory of the Free University in Brussels is focused on the origins and evolution of language and meaning, for an overview see (Steels, 1997). The approach is that both language and meaning are viewed as complex dynamical adaptive systems as proposed by Luc Steels (1996a, 1996b) Language, for instance, is complex because there is a huge expressability in the system. Furthermore, there is no language user that has complete competence ....

.... see e.g. Prigogine and Strengers, 1984) The models that are used are similar to those used by (Werner and Dyer, 1991) McLennan, 1991) Kirby and Hurford, 1997) and (Oliphant, 1996) An overview of research in this field can be found in (Hurford et al. 1998) In previous experiments (Steels and Vogt, 1997), Vogt, 1998a) a lexicon has been grounded by mobile robots on perceptual sensory information. In these experiments it was found that the feedback systems used were not sufficient for learning a completely coherent lexicon, so that the performance in the communication stayed poor. Furthermore, ....

[Article contains additional citation context not shown here]

Steels, L. and Vogt, P. (1997). Grounding adaptive language games in robotic agents. In Husbands, C. and Harvey, I., editors, Proceedings of the Fourth European Conference on Artificial Life, Cambridge Ma. and London. MIT Press.

....model of language interaction in a group of distributed agents, called the naming game. Keywords: origins of language, evolution of language, self organization. 1 Introduction Exciting recent research in the origins and evolution of language (see overviews in [Hurford et al. 1998] and [Steels, 1997c] is showing that when language is viewed as a complex adaptive system, it becomes possible to understand how a set of distributed agents is capable to reach a shared set of conventions, even if there is no global controlling agency or prior design. The main mechanism responsible for the ....

.... through computational simulations and is related to systems proposed and investigated by [MacLennan, 1991] Werner and Dyer, 1991] and [Oliphant, 1996] We have developed more complex variations of the game where the meaning consist of symbolic descriptions derived from discrimination games [Steels, 1997a] The game has also been implemented on physically grounded mobile robotic agents [Steels and Vogt, 1997] and on visionbased robotic talking heads , watching dynamically evolving scenes [Steels, 1997b] Of course in natural languages both the form and the meaning are vastly more complex than ....

[Article contains additional citation context not shown here]

L. Steels and P. Vogt. Grounding adaptive language games in robotic agents. In I. Harvey and P. Husbands, editors, Proceedings of the 4th European Conference on Artificial Life, Cambridge, MA, 1997. The MIT Press.

.... In addition, meaning emerges from interactions with the environment, individual adaptation and self organization (Steels, 1996b) Our research group investigates various aspects of the emergence of language ranging from lexicons, meaning, phonetics, syntax and language change, for an overview see (Steels, 1997). Also research is underway on the evolution of communication (De Jong, 1997) Our language system is comparable to the ones for instance described in (Oliphant, 1996) McLennan, 1991) and (Werner and Dyer, 1991) Most of our experiments were carried out in computer simulations, whereas some ....

.... 1991) and (Werner and Dyer, 1991) Most of our experiments were carried out in computer simulations, whereas some other were grounded in physical robotic experiments like in the emergence of phonetics (De Boer, 1998) the talking heads experiment (Belpaeme et al. 1998) and in mobile robots (Steels and Vogt, 1997). This paper discusses experiments of the latter kind. Other research on grounding communication in mobile robots can be found in (Billard and Dautenhahn, 1997) where a student robot learns a lexicon by imitating a teaching robot, using a connectionist model, and in (Yanco and Stein, 1993) where ....

[Article contains additional citation context not shown here]

Steels, L. and Vogt, P. (1997). Grounding adaptive language games in robotic agents. In Husbands, C. and Harvey, I., editors, Proceedings of the Fourth European Conference on Artificial Life, Cambridge Ma. and London. MIT Press.

.... may autonomously develop distinctions [19] and how they may develop autonomously a lexicon for expressing these distinctions [20] A first experiment in physical grounding, in which these capabilities were instantiated on robotic agents playing adaptive language games, has been presented in [23]. The present paper goes beyond this earlier work by using vision as source of sensory data and by showing the very beginnings of syntax. The research reported here is related to a lot of work currently being done in machine learning as well as recent work on the origins of language, as discussed ....

....before them. Figure 2: Typical example of dynamical scenes used in the talking heads experiment. They consist of autonomous robots roaming in an ecosystem with a charging station, competitors, and obstacles. These robots have been used in other language grounding experiments, as reported in [23]. They are then decoded by the hearer. A language game suceeds if the meaning decoded by the hearer fits with his observations and conceptualisations of the same scene. Otherwise the game fails and various repair actions are undertaken by each agent. The observation time is initially short so ....

Steels, L. and P. Vogt (1997) Grounding adaptive language games in robotic agents. In Harvey, I. et.al. (eds.) Proceedings of ECAL 97, Brighton UK, July 1997. The MIT Press, Cambridge Ma., 1997.

....and hence a spiral of increased complexity. Due to space limitations, this paper does not elaborate on this second feature. 4. 1 Category Profiles It is well known that linguistic categories (like natural kinds in general) cannot be defined in terms of strict necessary and sufficient conditions [12], 6] For example, many adverbs in English end in ly (slowly, quikly, steadily, etc. but there is no general rule that says that they all have to do so. Many nouns are based on classifying objects (table, man, mouse) but they need not be. In addition, there is enormous flexibility to ....

Steels, L. and P. Vogt (1997) Grounding adaptive language games in robotic agents. In Harvey, I. et.al. (eds.) Proceedings of ECAL 97 The MIT Press, Cambridge Ma., 1997.

No context found.

Steels, L. and P. Vogt (1997) Grounding adaptive language games in robotic agents. Submitted to ECAL-97.

....explored through computational simulations and is related to systems proposed and investigated by (Oliphant, 1996) MacLennan, 1991) Werner and Dyer, 1991) a.o. It has even been implemented on robotic agents who develop autonomously a shared lexicon grounded in their sensori motor experiences (Steels and Vogt, 1997), Steels, 1997) The naming game focuses on associating form and meaning. Obviously in human natural languages both form and meaning are non atomic entities with complex internal structure, but the results reported here do not depend on this internal complexity. We assume a set of agents A where ....

....simulations and is related to systems proposed and investigated by (Oliphant, 1996) MacLennan, 1991) Werner and Dyer, 1991) a.o. It has even been implemented on robotic agents who develop autonomously a shared lexicon grounded in their sensori motor experiences (Steels and Vogt, 1997) (Steels, 1997). The naming game focuses on associating form and meaning. Obviously in human natural languages both form and meaning are non atomic entities with complex internal structure, but the results reported here do not depend on this internal complexity. We assume a set of agents A where each agent a 2 A ....

L. Steels and P. Vogt. 1997. Grounding adaptive language games in robotic agents. In I. Harvey and P. Husbands, editors, Proceedings of the 4th European Conference on Artificial Life, Cambridge, MA. The MIT Press.

No context found.

L. Steels and P. Vogt, "Grounding adaptive language games in robotic agents," in Proc. 4th Eur. Conf. Artificial Life, 1997.

No context found.

Steels, L., & Vogt, P. (1997). Grounding adaptive language games in robotic agents. In Husbands, P., & Harvey, I. (Eds.), Proceedings of the Fourth European Conference on Artificial Life (ECAL'97), pp. 474--482. MIT Press / Bradford Books, Cambridge, MA.

No context found.

L. Steels and P. Vogt. Grounding adaptive language games in robotic agents. In Proc. of the 4 European Conf. on Artificial Life, 1997.

No context found.

Ma, C., Steels, L., Vogt, P., Amyot, R., Press, T.M.: Grounding adaptive language games in robotic agents. (2001)

No context found.

L. Steels and P. Vogt. Grounding adaptive language games in robotic agents. In C. Husbands and I. Harvey, editors, Proceedings of the Fourth European Conference on Artificial Life, Cambridge Ma. and London, 1997. MIT Press.

No context found.

L. Steels and P. Vogt. Grounding adaptive language games in robotic agents. In Proc. of the 4 European Conf. on Artificial Life, 1997.

No context found.

L. Steels and P. Vogt. Grounding adaptive language games in robotic agents. In C. Husbands and I. Harvey, editors, Proceedings of the Fourth European Conference on Artificial Life, Cambridge Ma. and London, 1997. MIT Press.

No context found.

L. Steels and P. Vogt. Grounding adaptive language games in robotic agents. In Proceedings of the Fourth European Conference on Artificial Life, pages 473--484, 1997.

No context found.

Steels, L. and Vogt, P. (1997). Grounding adaptive language games in robotic agents. In Proceedings of the 4th European Conference on Artificial Life. The MIT Press, Cambridge, Ma.

No context found.

Steels, L., & Vogt, P. (1997). Grounding adaptive language games in robotic agents. In P. Husbands & I. Harvey (Eds.), Proceedings of the Fourth European Conference on Arti#cial Life (ECAL '97 ) (pp. 474--482). Cambridge, MA: MIT Press/Bradford Books.

No context found.

Steels, L and P. Vogt. "Grounding adaptive language games in robotic agents." In Harvey, I. et.al. (Eds.) Proceedings of ECAL 97, Brighton UK, July

No context found.

Steels, L and P. Vogt. "Grounding adaptive language games in robotic agents." In Harvey, I. et.al. (Eds.) Proceedings of ECAL 97, Brighton UK, July

No context found.

L. Steels and P. Vogt, `Grounding adaptive language games in robotic agents', in Proceedings of the 4 th European Conference on Artificial Life. Cambridge, MA. The MIT Press., (1997). Planning and Scheduling 513 J. Sierra-Santib a nez

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC