The Foundations to Build a Virtual Human Society
Daniel Thalmann
Computer Graphics Lab
EPFL
CH 1015 Lausanne
Daniel.Thalmann@epfl.ch
http://ligwww.epfl.ch
Abstract: The simulation of the dynamics of human societies is not an easy
task, and we propose a bottom-up approach, which consists in defining the
human entity as a single, autonomous and independent entity (called Agent),
and then immerse many of them in a common environment to let them interact
and co-operate as the environment evolves. To achieve our goal and create
Virtual Human societies, we try to extend the concepts we have developed for
crowd simulations to encompass more autonomy at the level of individuals.
1. Introduction
The goal of our research effort is to develop a behavioral model for animation of believable virtual humans (actors) in virtual environments. In this context, believability means that the behavior of a synthetic actor has to be indistinguishable from that of a human. For instance, an actor inhabiting a typical virtual environment composed of objects and other actors, only concerned to perform a given task without reacting to objects or actors it encounters or to what these actors are doing would not look believable. The behavior of these actors would be more that of a robot than that of a human. To be believable, an actor has to be affected by what takes place around it and needs to engage in social behaviors with other actors. Therefore, the behavioral model of an actor needs to be versatile enough to allow a change of behavior, the emotional state of the actor must be reflected and must affect its behavior, the interpersonal relationships with the other actors must be taken into account and possibly bring the actor to engage in social interactions.
Our ultimate research objective is the simulation of Virtual Worlds inhabited by a Virtual Human Society (e.g. Fig.1), where agents will co-operate, negotiate, make friends, communicate, group and ungroup, depending on their likes, moods, emotions, goals, fears, etc. But such interaction and corresponding groups should not be programmed. Behaviour should emerge as a result of a multi-agent system sharing a common environment, in our case, sharing a virtual environment. For example in a panic situation we do not model the group behaviour, because each human reacts differently depending for example on its level of fear. If we model the individual entity, there will be groups of different behaviours (not programmed explicitly) as result of the interaction of common individual behaviours.
Fig.1. A Virtual Human Society
The simulation of the dynamics of human societies is not an easy task, and we propose a bottom-up approach, which consists in defining the human entity as a single, autonomous and independent entity (called Agent), and then immerse many of them in a common environment to let them interact and co-operate as the environment evolves.
To achieve our goal and create Virtual Human societies, we try to extend the concepts we have developed for crowd simulations to encompass more autonomy at the level of individuals. For interactive training simulations such as modelling of crowd responding to emergency situations we extend our current crowd model with inclusion of multiple dynamic physical and emotional attributes for individuals in crowds. Such attributes are used to represent various states of agents such as levels of fear, aggressivity, tiredness or injuries on different body parts affecting both agents behaviours and rendering of agents appearances. Agents are able to access states of other agents and have possibly influence on it, e.g. for modelling of people calming down other people in panic decreasing thus their fear level.
We also extend and modify our generic behavioural control system to be usable in simulation of emergency situations, where combination of various levels of agents' state variables and perceived state of the environment produce both corresponding overt behaviours and change of internal states (e.g. when agents are inside toxic gas area their breathing capacity decreases of 10% each 10 seconds). For transitions between normal state background behaviours of crowd and behaviours corresponding to particular state of the emergency situation, system should also allow to specify different levels of priorities of behaviours. Real-time requirements of training simulations put constraints on complexity of individual behaviour, so the system using various levels of behaviour complexity can be useful for decreasing load of computational resources. For example rules for background behaviour associated with places in virtual urban environment common for all agents can be stored in informed
environment database [1], then when an emergency situation occurs the state of
agents change and priority system would accordingly switch to individual behavioural rules.
We also simulate the main steps of the groups' so-called "life-cycle": the initial formation, maintenance and closure stages. This includes the requirement of defining what are the necessary conditions for the transition from one stage to another, as well as a proper handling of members input and output through a mechanism of collective decision making. An interesting area for further research is modelling of the process of crowd formation. In previous works crowds have been always considered as already formed units with more or less uniform behavior placed in very particular environment corresponding only to narrow purpose of simulation e.g. pedestrian just fleeing from burning building or marching crowd during political demonstration. However in the real world there is no global controller of crowds assigning people to particular groups and determining their behaviors. Collective behavior emerges from interaction of individuals, crowds are dynamically assembled and disassembled and over time change their behavior. For such modelling notion of contagion from sociology can be useful. Contagion is process where crowd members become increasingly preoccupied with each other's behavior, and outlooks and events that would normally concern them become insignificant. Eventually each crowd member becomes so focused on other's activities that he or she loses self-consciousness, which usually is a "means of barricading oneself against influence of others".
Several group structures should emerge, each with a different configuration of the members' roles and status. Such issues as leadership, power distribution and roles conflict can be tackled. The internal evolution of the group, through dynamic members' roles distribution,
There are mainly three factors allowing to judge of such a Virtual Reality system: the autonomy, the interaction and the degree of presence. The autonomy expresses the ability of autonomous virtual humans to react to simulated stimuli. The interaction measures the number of adjustable parameters of the system and the presence measures the degree of immersion of the user in a virtual world and at which extent, he can interact with it using specialized devices.
2. Crowds
2.1 Characteristics of a crowd
Crowds are ubiquitous feature of everyday life. People have long assembled collectively to observe, to celebrate, and to protest various happenings in their everyday lives. The collective assemblages or gatherings called crowds are ongoing features of the social word and, as a consequence, have long been the object of theorizing and inquiring, ranging from psychologistic renderings of LeBon [2] and Freud [3] to the more sociological accounts of Smelser [4] to the highly systematic and empirically grounded observations of McPhail [5].
With computers it become possible not only to observe human crowds in the real world, but also to simulate various phenomena from the domain of collective behavior
in virtual environment. Collective behaviors have been studied and modeled in computers with very different purposes:
? To animate large number of characters for entertainment industry in movies and computer games [6].
? To populate simulated virtual worlds for training of military personnel [7] or policemen dealing with crowds [8].
? To model movement of a large number of people to support architectural design both for everyday use of a buildings and for emergency evacuation
conditions [9][10].
? And finally, for applications concerning sociological and behavioral simulations [11].
There have been different approaches to modeling of the crowds, because of different purposes of the models required by various application domains. The modeling techniques range from those that do not distinguish individuals such as flow and network models, to those that represent each individual as being controlled by rules based on physical laws or behavioral models.
Advantage of macroscopic models is their low requirements on computational capacity making simulation of even very large crowds possible. However as noted by Sime [12], pedestrian movement models that represent people as "nonthinking objects" whose movement is modelled by analogy to laws of physical science neglect the psychological aspects that are important for effective modelling.
2.2 A crowd modelling
Animating crowds [13] is challenging both in character animation and a virtual city modeling. Though different textures and colors may be used, the similarity of the virtual people would be soon detected by even non-experts, say, “everybody walks the same in this virtual city!” . It is, hence, useful to have a fast and intuitive way of generating motions with different personalities depending on gender, age, emotions, etc., from an example motion, say, a genuine walking motion. The problem is basically to be able to generate variety among a finite set of motion requests and then to apply it to either an individual or a member of a crowd. It also needs very good tools to tune the motion [14].
The proposed solution addresses two main issues: i) crowd structure and ii) crowd behavior. Considering crowd structure, our approach deals with a hierarchy composed of crowd, groups and agents, where the groups are the most complex structure containing the information to be distributed among the individuals. Concerning crowd behavior, our virtual agents are endowed with different levels of autonomy. They can either act according to an innate and scripted crowd behavior (programmed behavio r), react as a function of triggered events (reactive or autonomous behavio r) or be guided by an interactive process during simulation (guided behavio r). We introduced the term
Figure 2. Group guided by a leader
In our case, the intelligence, memory, intention and perception are focalized in the group structure. Also, each group can obtain one leader. This leader can be chosen randomly by the crowd system, defined by the user or can emerge from the sociological rules. Concerning the crowd control features, The crowd aims at providing autonomous, guided and programmed crowds. Varying degrees of autonomy can be applied depending on the complexity of the problem. Externally controlled groups,
The train station simulation (Figure 3) includes many different actions and places, where several people are present and doing different things. Possible actions include “buying a ticket”, “going to shop“, ”meeting someone”, “waiting for someone”, “making a telephone call”, “checking the timetable”, etc. This simulation uses external control (RBBS [16]) to guide some crowd behaviors in real time.
Figure 3.Train station simulation.
3. Agents
3.1 Why agents ?
The creation of groups of virtual humans capable of behaving and interacting realistically with each other in 3D environments requires the development of a specific agent architecture. Most current agent architectures are dedicated to the fulfilment of precise tasks/problem solving, e.g. Ingrand [17], or to the simulation of purely emotional behaviour, e.g. Botelho [18]. However, we think that pure goal oriented and emotional behaviour are not sufficient in order to fully model human interaction in groups.
In real life, we daily engage in many interactions and social activities with different people, adapting our behaviour to the situation, dealing with complex motivations and different cultures, and also simply following routines. Our social status, gender, age or cultural origin are, among others, very important criteria to explain why we enjoy interacting with some people and feel uneasy with others.
Sociologists have identified and described specific social mechanisms used by the individuals which are necessary to the creation and maintenance of groups, and have underlined the importance of the participants' social identity in the group's structure and behaviour, e.g. Forsyth [19].
These facts need to be taken into account in the development of an agent architecture allowing the simulation of group behaviour. The participating agents must be modeled at a higher level, adding social identities, social motivations and social reasoning to them, on top of their specific competency.
3.2 Behaviour and mental states
One important point in the simulation is the believability of the individual virtual humans, they should behave as real humans, including capabilities such as: perception, language understanding and generation, emotions, goal-driven behaviour, reactivity to the environment, memory, inference, appearance of thought and personalities, interpersonal interactions, social skills and possibly others. A very important factor for the believability are emotions. Emotions must affect everything about the entity, the way it moves, the way it talks, the expression on its face (see Figure 4) , “The expression must be captured throughout the whole body as well as in the face” [20].
Fig.4 Emotional state
3.3 Perception and sensors
In order to implement perception, virtual humans should be equipped with synthetic or virtual sensors [21] like visual, tactile and auditory sensors. These sensors should be used as a basis for implementing everyday human behaviour such as visually directed locomotion, handling objects, and responding to sounds and utterances. The actor-environment interface, or the synthetic sensors, constitute an important part of a behavioral animation system. As sensorial information drastically influences behavior, the synthetic sensors should simulate the functionality of their organic counterparts. Due to real-time constraints, we did not think that it is necessary to model biological models of sensors. Therefore, synthetic vision [22] only makes efficient visibility tests using powerful graphics engine that produce a Z-buffered color image representing an agent's vision. A tactile point-like sensor may be represented by a simple function evaluating the global force field at its position. The synthetic "ear" of an agent will be represented by a function returning the on-going
sound events. What is important for an actor’s behavior is the functionality of a sensor and how it filters the information flow from the environment, and not the specific model of the sensor.
More precisely, each agent will be running an inference engine in a separated process, and this engine will be responsible of the high-level behaviours and decisions. That is, based on plans and input from the environment (visual perception, communications, etc.) or from internal states (tiredness, obligations...), the behavioural engine will select the proper actions to perform.
3.4 Emotional and social relationships
Let us now consider a place where a lot of human relationships take place like a work place office or factory for example. These relationships could be organized in two sets: social relations and hierarchical relations. While the hierchical relations change rarely, the social relationship, that is, all informal and inter-personal relations between the staff would evolve during the simulation. These relations are normally developed outside working hours, like coffe breaks or lunch time, and they are normally based on people likes, activities, fears etc.
In order to realistically simulate how humans interact in a specific social context, it is necessary to precisely model the type of relationship they have and specify how it affects their interpersonal behaviour. Sociologists have identified several dimensions that are important to provide for the simulation of any group behaviour: 1) power (dominance and submissiveness of the agents), 2) attraction (friendliness and unfriendliness), 3) instrumental control (hierarchical rank) and 4) emotional expressiveness. For the development of the agents’ behaviour in an organizational context, taking into account the hierarchical relationship is of course crucial, but the importance of the more informal dimensions should not be underestimated.
3.5 Physical parameters
Physical parameters such as hunger, tiredness, types and levels of injuries should have influence on both behaviours and visualisation of virtual humans. For example, according to their level of tiredness, people would be able to perform different types of activities and this would have different influence on person’s fatigue; people burned by fire at various body parts would have their appearance changed appropriately; mobility of persons affected by toxic gas would decrease and eventually they could become unconscious.
These attributes should be perceivable and modifiable by other agents. For example in order to be able to simulate emergency situations in a factory where one worker is injured and other worker are helping him, agent should perceive other as injured, identify injured part of body so that it could apply appropriate first aid and also change its physical state to reflect effect of the aid (e.g. increasing mobility of the injured agent). Dynamic evolution of such attributes according to agent’s interaction with environment and other agents should be possible to specify by a system of rules.
3.6 Social parameters and behavior
The agents’ behaviour within a group is highly dependant on their location in the sociometric structures but also on their own social identity. This social characterization should be done using social-statistical variables, 1) culture, 2) gender and 3) age, and taking into account the agent’s roles within the group. A distinction between 4) task roles (e.g. function), and 5) socioemotional roles (e.g. confident) is often used. An additional 6) status rating (prestige) can also be introduced.
A typical behavior of an actor for a party scenario may be the following. The actor arrives at the party and looks what happens around it. Some food and drinks may stand on a table. Probably, it will be surrounded by friends and unknown persons. The actor may decide to take a drink and some food or to have social interactions with other actors. These interactions may be friendly conversations or the courtship of another actor from the opposite gender. In a friendly conversation, two actors may also share a drink or an actor may be rejected by the other actor when their friendship is not reciprocal. Depending on the issue of a courtship, an actor may also leave the party with another one. The behavior for a supermarket scenario is mainly a sequence of tasks to achieve. The actor comes in the supermarket, takes all the needed articles and leaves the supermarket. Nevertheless, the actor may also be attracted by unforeseen articles or may interact with other actors. Both scenarios are the source of emotions. The objects like drink and food at the party or article in the supermarket are the source of liking and disliking emotions. Behaviors generated by these emotions may be to take a liked object or to ignore or avoid a disliked one. The actions of the actor itself or from other actors may generate pride and shame or admiration and reproach emotions. For instance, an actor knocking down an object will be ashamed and another actor will reproach it. An actor helping another actor to take an unreachable object will be pride and the other actor will admire it. Events like the ownership of an object or the acceptance of a communication by another actor may generate joy emotions or the lost of an object or the refusal of a communication may generate distress emotions. The emotions happy for, sorry for, resentment and gloating are generated when an actor sees another actor having joy or distress emotions. The emotions hope and fear, relief and disappointment, satisfaction and fears confirmed depends on prospective events like the possible ownership of an object or a special expectation from a communication.
3.7 Verbal and non-verbal communication
Social interactions requires to reproduce verbal communication. In such a graphical environment, inter-agents communication would be more than just an exchange of messages. We need to create models for sound propagation, which should be suitable for verbal communication. These models do not need to be as accurate as for accoustic simulations, but should carry enough information in order to avoid situations such as too people conversing while they are in two different rooms, for instance. We should also take into account some human interaction elements, such as, turn taking (ability to interrupt others while they are talking, or wait for pauses indicating that it the other agent is waiting for an answer), conversation tracking (that
is, mainly the ability to suspend and continue a conversation, to handle multiple conversations and queries at the same time), interpretation and reactions to silences during the speech, gazing while conversing, etc.
Interesting implication of introducing communication abilities for agents could be in modelling of emergent collective behaviour. Emergent crowds can be formed based on transfer of both emotional states and behaviours for example by verbal and non-verbal communication between physically close agents. Such transfer can depend on personalities of involved agents, where charismatic persons will be more easily able to affect others, or where some agents with high level of dominance will be resisting to adopt crowd behaviour. Further possibility of such emotional and behavioural influences can be used to indirectly control the crowd, where one of the agents would be externally guided by real human user.
To increase the believability of behaviors resulting from emotions we included nonverbal communication elements. A nonverbal communication is concerned with postures and their indications on what people are feeling. Postures are the means to communicate and are defined by a specific position of the arms and legs and angles of the body. As stated in [23], 65 percent of the substance of a face to face communication is conveyed by nonverbal elements and many psychologists suggest that a posture is an important means of conveying interpersonal attitudes. The effects of nonverbal communication, though unconscious, are nevertheless important. Let us have an example: two persons communicating with each other and rejecting a third one. The first two persons are standing in an open triangle formation commonly used for friendly conversation. It leaves a potential opening for a third person to join them. Such a person attempts to join the group, but the first two respond by rejecting him out. One raises his left arm, forming a barrier. The other deliberately avoids making eye contact with the newcomer. They have formed a closed formation, suppressing any possibility for a new person to join them. The newcomer feels rejected and leaves unhappily. The interpersonal relationships between actors is represented by a structure that encodes the type of the relationship (friendly, hierarchical) and the level of that relationship (from 0 (not friend, subordinate) to 1 (very close friend, superior)). These values are used to evaluate the desire to communicate with another actor and are updated according to the issue of the communication.
The issue of a communication may be for example the transfer of an object from an actor to another one or the rejection of an actor by another one. Figure 5 shows an example.
Fig. 5. A non-verbal communication between two virtual humans
3.8 Our agent-based architecture for Intelligent actors
Automatically animating a virtual human is such a complex task, that one has to decompose it into simpler elements and distinguish between low and high-levels. Saying that a task belongs to low level does not means that it is something easier or faster to achieve! We will rather regroup in the low-level the physical elements (for the virtual human, this includes the body, and basic animations like locomotion or objects interaction), and the behaviour will be handled by the high-level, simulating the virtual human’s brain process. As shown in Figure 6, our low-level component is called Agents’ Common Environment (ACE) and the high-level decisions making (the IntelligentVirtual Agent or IVA).
To simulate the physical word, we have developed the ACE system [24] that understands a set of different commands to be able to control the simulations, like the creation and animation of 3D objects, virtual humans, and smart objects [25] . Virtual humans encapsulate various motion motors and may have facial expression. They are able to repeat a recorded animation (key-frames animation), can walk, or use Inverse Kinematics for better interactions with objects. Finally, they can visually perceive their environment. ACE is mainly coded in C++ to ensure high performances. For convenient user-interaction, it also provides a Python layer which interprets commands on the fly and animates the virtual humans.
Fig.6. Low and high level components in the system
IVA [26] relies on a Beliefs, Desires and Intentions architecture, as described by Rao and Georgeff [27], so that each agent manages its own beliefs and takes decisions by itself. While agents in the low-level have to share a common area and are handled by the Agents Controller, there is no such need for the IVA. Therefore, even if each IVA runs into a separated Java thread over the same application, there is no inter-threads communication.
This is demonstrated in Figure 7: when Agent 1 speaks to Agent 2, the message goes first to the low-level,then the thread for the first agent puts the message inthe shared area. Agent 2 is then able to retrieve the message and hear it.. This is in accordance with real human interaction: we do not have direct connections between
our brains, but we need to use our body for communicating. The knowledge of the agent is decomposed into its beliefs and internal states (like the anxiety for instance), the goals to achieve and the plans, which are specifying a sequence of actions required to reach a specific goal. Based on these, the agent is then able to select the correct actions to perform in order to achieve its goals. Since beliefs and goals are subject to evolve over time, the agent is able to take into account new elements and dynamically react.
Fig. 7. Verbal communication between two IVAs has to go through the low-level
Conclusion
In this paper, we have shown that the simulation of the dynamics of human societies is a very complex task. We have proposed to use a agent architecture to build such a society. Our approach is bottom-up and consists in defining the human entity as a single, autonomous and independent entity (the Agent), and then immerse many of them in a common environment to let them interact and co-operate as the environment evolves. To achieve our goal and create these Virtual Human societies, we extend the concepts we have developed for crowd simulations to encompass more autonomy at the level of individuals.
Acknowledgment
The author is grateful to the people who contributed to this research, especially Marcelo Kallmann, Jean-Sebastien Monzani, Angela Caicedo, Anthony Guye-Vuillème, Branislav Ulicny, Soraia Musse, and Pascal Bécheiraz. The research was sponsored by the Swiss National Research Foundation.
References
1. Farenc, N., Boulic, R., Thalmann, D., "An Informed Environment Dedicated to the
Simulation of Virtual Humans in Urban Context", Proc. Eurographics '99, Milano, Italy, pp.309-318, 1999.
2. LeBon, G., Psychologie des Foules, Paris:Alcan, 1895.
3. Freud, S., Group Psychology and Analysis of the Ego, London: International
Psychoanalytical Press, 1921.
4. Smelser, N., Theory of collective behavior, Routledge & Keganpaul, London, 1962.
5. McPhail, C., The Myth of Maddening Crowd, NY:Aldine De Gruyter, 1991.
6. Character Studio 3, https://www.doczj.com/doc/574222838.html,/docs/char_st_WHT.pdf
7. Varner, D., et al., "UMSC Small Unit Leader Non-Lethal Trainer", in Proc. ITEC'98, 1998
8. Williams, J.,R., A Simulation Environment to Support Training for Large Scale Command
and Control Tasks, PhD thesis, University of Leeds, 1995.
9. Still, G., K., Crowd Dynamics, PhD thesis, Warwick University, 2000
10. Helbing, D., Molnar, P., "Self-Organization Phenomena in Pedestrian Crowds", Pages
569-577 in: F. Schweitzer (ed.) Self-Organization of Complex Structures. From Individual to Collective Dynamics, Gordon and Breach, London, 1997.
11. McPhail, C., Powers, W.,T., and Tucker, C.,W., "Simulating individual and collective
actions in temporary gatherings", Social Science Computer Review, 10(1):1-28, Spring, 1992.
12. Sime, J., D., "Crowd psychology and engineering: Designing for people or ballbearings.",
in R.A. Smith and J.F. Dickie, eds., Engineering for Crowd Safety, p. 119-130, Elsevier, 1993
13. Musse S.R., Thalmann D., A Behavioral Model for Real-Time Simulation of Virtual
Human Crowds, IEEE Transactions on Visualization and Computer Graphics, Vol.7, No2, 2001, pp.152-164.
14. Emering L., Boulic R., Molet T., Thalmann D., Versatile Tuning ofHumanoid Agent
Activity, Computer Graphics Forum
15. Musse, S.R., Babski, C., Capin, T. and Thalmann, D. Crowd, Modelling in Collaborative
Virtual Environments. ACM VRST ‘98, Taiwan
16 Schweiss E., Musse S.R., Garat F.. Thalmann D., An Architecture to Guide Crowds Using
a Rule-Based Behaviour System, Proc.Agents 99.
17. Ingrand, F.F., Georgeff, M.P., and Rao, A.S. (1992). An architecture for real-time
reasoning and system control. IEEE Expert/Intelligent Systems, 7(6):34–44.
18. Botelho, L.M. and Coelho, H. (1998). Artificial autonomous agents with artificial
emotions. In Proceedings of the Second International Conference on Autonomous Agents, pages 449–450, New York. ACM Press.
19. Forsyth, D. R. (1990). Group dynamics (2nd ed.). Pacific Grove, CA: Brooks/Cole.
20. Thomas F., Johnston O.. Disney Animation: The Illusion of Life. Abbeville Press, New
York, 1981.
21 Thalmann D., A New Generation of Synthetic Actors: the Interactive Perceptive Actors,
Proc. Pacific Graphics ‘96 Taipeh, Taiwan, 1996, pp.200-219.
22 Renault O., Magnenat-Thalmann N., Thalmann D., A Vision-based Approach to
Behavioural Animation, Journal of Visualization and Computer Animation, Vol.1, No1, 1990, pp.18-21.
23. Argyle M., Bodily Communication, Methuen and Co Ltd. (1975)
24 Kallmann M, Monzani J.S., Caicedo A., Thalmann D., ACE: A Platfor for the Real time
Simulation Of Virtual Human Agents in Real Time, Proc. Eurographics CAS 2000, pp.73-
85.
25 Kallmann M., D. Thalmann D., A behavioral interface to simulate agent-object
interactions in real-time. In IEEE Computer Society Press, editor, Proc. Computer Animation 99, pages 138–146, 1999.
26 Monzani J.S., Caicedo A., Thalmann D., Integrating Behavioural Animation Techniques,
Proc. Eurographics 2001.
27 Rao A.S., Georgeff M.P., Modeling rational agents withing a bdi-architecture. In J. Allen,
R. Fikes, and E. Sandewall, editors, Proceedings of the Third Inter-national Conference on Principles of Knowledge Representation and Reasoning. Morgan Kaufmann, 1991.
机器翻译技术介绍
常宝宝 北京大学计算语言学研究所 chbb@https://www.doczj.com/doc/574222838.html,
什么是机器翻译
研究目标：研制出能把一种自然语言（源语言）的文 本翻译为另外一种自然语言（目标语言）的文本的计 算机软件系统。 制造一种机器，让使用不同语言的人无障碍地自由交 流，一直是人类的一个梦想。 随着国际互联网络的日益普及，网上出现了以各种语 言为载体的大量信息，语言障碍问题在新的时代又一 次凸显出来，人们比以往任何时候都更迫切需要语言 的自动翻译系统。 但机器翻译是一个极为困难的研究课题，无论目前对 它的需求多么迫切，全自动高质量的机器翻译系统 (FAHQMT)仍将是人类一个遥远的梦。
机器翻译的基本方法
机器翻译的基本方法 ? 基于规则的机器翻译方法 ? 直接翻译法 ? 转换法 ? 中间语言法 ? 基于语料库的机器翻译方法 ? 基于统计的方法 ? 基于实例的方法 ? 混合式机器翻译方法
目前没有任何 一种方法能实现机 器翻译的完美理 想，但在方法论方 面的探索已经使得 人们对机器翻译问 题的认识更加深 刻，而且也确实带 动了不少不那么完 美但尚可使用的产 品问世。
20世纪90年代以前，机器翻译方法的主流一直是基于规则的方 法，不过，统计方法后来居上，目前似乎已成主流方法，从学术 研究的角度看，更是如此。(Google translate)
机器翻译的基本方法
20世纪90年代以前，机器翻译方法的主流一直是基于 规则的方法，因此基于规则的方法也称为传统的机器 翻译方法。 直接翻译法 ? 逐词进行翻译，又称逐词翻译法(word for word translation) ? 无需对源语言文本进行分析 ? 对翻译过程的认识过渡简化，忽视了不同语言之间 在词序、词汇、结构等方面的差异。 ? 翻译效果差，属于早期过时认识，现已无人采用 How are you ? 怎么 是 你 ? How old are you ? 怎么 老 是 你 ?
Study on Layout Optimization of Passenger Station based on Multi-objective Planning Abstract—This paper study on the optimal layout of passenger station and the basic principle of selection location. There are both quantitative index and qualitative index existing in the evaluation indexes of the layout of passenger station. To unify the evaluation indexes, this paper establishes a multi-objective planning optimization model of the layout of passenger station, by using the basic idea of multi-objective planning and introducing the effect of targets of the evaluation indexes. In Nanning railway terminal for example, four evaluation indexes: construction investment、operation cost in railway terminal、cost of passenger trip and the other social benefits are determined. Calculate respectively all effect of targets of the evaluation indexes and then determine the optimal scheme of the layout of passenger station in Nanning railway hub by the total effect of targets after being weighted. Keywords-component; multi-objective; passenger station; Layout Optimization In the past, paying much attention to improving line carrying capacity and neglecting the carrying capacity of hubs or stations in China leaded to the shortage of passenger capacity and the difficulty in meeting the demand of transport market. At present, China is in the primary stage of high-speed railway construction. High-speed railway being introduced into existing railway terminal will break the transport pattern of existing railway hub and will dramatically change the original railway hub transportation structure、transportation operation mode and transportation path. It is an important question in railway terminal planning to reasonably determine the locations and the number of the high-speed passenger stations in hub and optimize the layout of passenger stations in terminal. I Layout Optimization model of Passenger Station in railway hub A. Layout of passenger station in railway hub and basic principle of selection location （1）For human-oriented, make the trip distance and time of passenger from the starting point to passenger station shortest、the travel cost lowest、the riding most convenient and reduce the passenger transfer or make it more convenient. (Including
机器翻译研究现状与展望1 戴新宇，尹存燕，陈家骏，郑国梁 （南京大学计算机软件新技术国家重点实验室，南京 210093） （南京大学计算机科学与技术系，南京 210093） 摘要：本文回顾机器翻译研究的历史，介绍典型的机器翻译方法，包括：基于规则、基于统计以及基于实例的机器翻译方法；针对机器翻译的研究现状，详细介绍和分析了基于混合策略的机器翻译方法，对统计以及机器学习方法在机器翻译中的应用进行了描述；论文还介绍了当前机器翻译评测技术；最后对机器翻译进行总结和展望。 关键字：机器翻译，基于规则，基于统计，基于实例，混合策略，机器学习 Machine Translation:Past,Present,future Dai Xinyu, Yin Cunyan, Chen Jiajun and Zheng Guoliang (State Key Laboratory for Novel Software Technology, Department of Computer Science & Technology Nanjing University, Nanjing 210093) Abstract：This paper firstly presents the history of machine translation, and introduces some classical paradigms of machine translation: RBMT, SBMT and EBMT. Secondly, we introduce the recent research on machine translation, and describe the hybrid strategies on machine translation in detail, and discuss the applications of machine learning for machine translation. We also analyze the current techniques about evaluation on machine translation. Finally, we draw a conclusion and prospect on the research of machine translation. Keywords：Machine Translation, RBMT, SBMT,EBMT, HSBMT, Machine Learning. 1本论文工作得到863课题资助（编号：2001AA114102, 2002AA117010-04） 戴新宇博士生，主要研究自然语言处理；尹存燕助教，主要研究自然语言处理；陈家骏教授，博士生导师，主要研究自然语言处理、软件工程；郑国梁教授，博士生导师，主要研究软件工程。
本科毕业设计（论文） 外文翻译 专业名称：土木工程 年级班级： 学生姓名： 指导教师： 二○年月日
Urban transportation Planning An urban transportation system is basic component of an urban area's social，economic，and physical structure. Not only does the design and performance of a transportation system provide opportunities for mobility，but over the long term，it influences patterns of growth and the level of economic activity through the accessibility it provides to land. Planning for the development or maintenance of the urban transportation system is thus an important activity，both for promoting the efficient movement of people and goods in an urban area and for maintaining the strong supportive role that transportation can play in attaining other community objectives. There are several basic concepts about an urban transportation system that should be kept in mind. Most important，a transportation system in an urban area is defined as consisting of the facilities and services that allow travel throughout the region，providing opportunities for:(I)mobility to residents of an urban area and movement of goods and (2) accessibility to land .Given this definition，an urban transportation system can be further characterized by three major components: the spatial configuration that permits travel from one location to another; the transportation technologies that provide the means of moving over these distances; and the institutional framework that provides for the planning, construction, operation, and maintenance of system facilities. The Spatial Configuration of a Transportation System One way to describe the spatial dimension of an urban transportation system is to consider the characteristics of individual trips from an origin to a destination. For example, a trip can consist of several types of movement undertaken to achieve different objectives. Travelers leaving home might use a local bus system to reach a suburban subway station(a trip collection process)，proceed through the station to the subway platform (a transfer process)，ride the subway to a downtown station (a line-haul process)，and walk to a place of employment (a distribution process). Similarly，one can view a home-to-work trip by car as consisting of similar segments，with the local street system providing the trip collection process, a freeway providing the line-haul capability，a parking lot in the central business district serving as a transfer point，and walking，as before，serving the distribution function. The facilities and services that provide these opportunities for travel，when
英文翻译 Flexible pavement design Generally speaking,pavements(and bases) may be divided into two broad classifications or tipes:rigid and flexible. As commonly used in the United States,the term “rigid pavement”is applied to wearing surfaces constructed of Portland-cement concrete. A pavement constructed of concrete is assumed to possess considerable flexural strength that will permit it to act as a beam and allow it to bridge over minor irregularities which may occor in the base or subgrade on which it rests;hence the term “rigid”.Similarity,a concrete base that supports a brick or block layer might be described as “rigid”. All other types of pavement have traditionally been classed as “flexible”.A commonly used definition is that “a flexible pavement is a structure that maintains contact with and distributes loads to the subgrade and depends on aggregate interlock,particlefriction,and cohesion for stability”.Thus,the classical flexible pavement include primarily those pavement that are composed of a series of granular layers topped by a relatively thin high-quality bituminous wearing surface .Typically,the highest-quatily materials are at or near the surface.It should be pointed out that certain pavementsthat have an asphalt surface may behave more like the classical “rigid”pavement,for example, pavement that have very thick asphalt surface or that have base courses composed of aggregate treated with asphalt,cement, or lime-fly ash. However,for convenience of presentation,these pavements will be considered to be in the flexible class. The structure of flexible pavement is composed of a “wearing surface”, base, subbase(not always used), and subgrade . The wearing surface and the base often comprise two or more layers that are somewhat different in composition and that are put down in separate construction operations.On many heavy-duty pavements,asubbase of select material is often placed between the base and subgrade.the wearing surface may range in thickness from less than 1 in. in the case of a bituminous surface used for low-cost, light-traffic loads to 6 in. or more of alphalt
机器翻译综述 1.引言 1.1机器翻译的历史 现代机器翻译的研究应该是从20世纪50年代开始，但是早在这以前很多人已经提出了相应的想法，甚至是远在古希腊时期就有人提出要用机器来进行语言翻译的想法。 在1946年，美国宾夕法尼亚大学的两位科学家设计并制造了世界上第一台电子计算机。与此同时，英国工程师同美国洛克菲勒基金会副总裁韦弗在讨论计算机的应用范围时，就提出了利用计算机实现语言的自动翻译的想法。在1949年，韦弗发表了一份名为《翻译》的备忘录，正式提出了机器翻译问题。他提出了两个主要观点： 第一，他认为翻译类似于解读密码的过程。 第二，他认为原文与译文“说的是同样的事情”，因此，当把语言A翻译为语言B时，就意味着从语言A出发，经过某一“通用语言”或“中 间语言”，可以假定是全人类共同的。 在这一段时间由于学者的热心倡导，实业界的大力支持，美国的机器翻译研究一时兴盛起来。 1964年，美国科学院成立语言自动处理咨询委员会，调查机器翻译的研究情况，给出了“在目前给机器翻译以大力支持还没有多少理由”的结论，随后机器翻译的研究就陷入了低潮期。直到70年代以后机器翻译的研究才重新进入了一个复苏期，随后机器翻译的发展又迎来了繁荣期 1.2机器翻译的主要内容 经过50多年的发展，在机器翻译领域中出现了很多的研究方法，总结如下：●直接翻译方法 ●句法转换方法 ●中间语言方法 ●基于规则的方法 ●基于语料库的方法 基于实例的方法（含模板、翻译记忆方法） 基于统计的方法 在当前的研究中，更多的是基于统计的方法进行的，因为基于统计的方法可以充分的利用计算机的计算能力，并且并不需要过多的语言学知识作为支撑，可以让更多的计算机科学家投入到实用系统的研究中，极大的促进了统计机器翻译的发展。 下面对各个方法逐一的进行介绍。
中英文对照外文翻译文献 (文档含英文原文和中文翻译) 英文原文： The Basics of a Good Road We have known how to build good roads for a long time. Archaeologists have found ancient Egyptian roadsthat carried blocks to the pyramids in 4600 BCE. Later,the Romans built an extensive road system, using the same principles we use today. Some of these roads are still in service. If you follow the basic concepts of road building, you will create a road that will last. The ten commandments of a good road are: （1）Get water away from the road （2）Build on a firm foundation （3）Use the best materials （4）Compact all layers properly （5）Design for traffic loads and volumes
（6）Design for maintenance （7）Pave only when ready （8）Build from the bottom up （9）Protect your investment （10）Keep good records 1．Get water away from the road We can’t overemphasize the importance of good drainage.Engineers estimate that at least 90% of a road’s problems can be related to excess water or to poor waterdrainage. Too much water in any layer of a road’sstructure can weaken that layer, leading to failure. In the surface layer, water can cause cracks and potholes. In lower layers it undermines support, causing cracks and potholes. A common sign of water in an asphalt road surface is alligator cracking — an interconnected pattern of cracks forming small irregular shaped pieces that look like alligator skin. Edge cracking, frost heaves, and spring breakup of pavements also point to moisture problems. To prevent these problems remember that water: ? flows downhill ? needs to flow someplace ? is a problem if it is not flowing Effective drainage systems divert, drain and dispose of water. To do this they use interceptor ditches and slopes,road crowns, and ditch and culvert systems. Divert —Interceptor ditches, located between the road and higher ground along the road, keep the water from reaching the roadway. These ditches must slope so they carry water away from the road. Drain —Creating a crown in the road so it is higher along the centerline than at the edges encourages water to flow off the road. Typically a paved crown should be 1?4" higher than the shoulder for each foot of width from the centerline to the edge. For gravel surfaces the crown should be 1?2" higher per foot of width. For this flow path to work, the road surface must be relatively water tight. Road shoulders also must be sloped away from the road to continue carrying the flow away. Superelevations (banking) at the outside of curves will also help drain
现代城市交通规划的思想发展·文献综述 所谓交通规划，是指根据特定交通系统的现状与特征，用科学的方法预测交通系统交通需求的发展趋势及交通需求发展对交通系统交通供给的要求，确定特定时期交通供给的建设任务、建设规模及交通系统的管理模式、控制方法，以达到交通系统交通需求与交通供给之间的平衡，实现交通系统的安全、高效与节能、环保。【1】 1、现代城市交通规划的理论发展 城市交通规划以城市总体规划为前提，在城市规划中逐渐发展、不断成熟。最初的交通仅限为道路的铺设、交通工具的开发，20世纪末机械开始代替人力，交通运输发展迅猛，交通规划开始进入人们视线……随着城市和交通运输的发展，路网设计的重要性日益突出，1925年法国勒·柯布西耶在《光明城市》【2】中分析了交通问题的产生原因，提出了立体交通的构想，1933年的《雅典宪章》【3】也进一步反映了设计者们对交通规划的重视……到了现代，交通规划已不局限于路网的设计，美国学者肯尼思·科尔森（Kenneth Kolson）在《大规划》（Big Plans，2001）【4】一书中写道：设计师们的注意力不仅仅集中于交通峰值、出入口布局、停车位计算等总体规划的关键参数，而更多地关注市民出行的特征和感受，并以此调整城市区域内大型交通运输的主导模型。从“车为本”到“人为本”，是现代交通规划理念的一大进步。 2、以交通线为骨架的线形城市思想 1882年，轨道交通发展迅猛，西班牙工程师索马里·马塔提出“线形城市”【5】的概念，其重要原则就是城市建设的一切其他问题均以城市运输问题为前提，通过交通运输线组成城市的网络。索马里大胆地提出城市不再是分散在不同地区的点，而是由一条铁路和道路干道相串联在一起的、连绵不断的城市带，并且这个城市是可以贯穿整个地球的。这一理论在在上海世博会期间由乌克兰建筑设计师伊格尔·斯维斯顿【6】进一步完善、展示，他对于具体的城市功能区域安排提出了自己的想法。 这样一个以交通线为轴线的线形城市形态是否可行、建设范围在多大较为科
智能交通信号灯 摘要：信号控制是一种必要的措施以确保的质量和安全,交通循环。现在的信号控制的进一步发展具有极大的潜力来减少运行时间、车辆、事故成本和整车排放。检测的发展和计算机技术改变了交通信号控制从定时开环规定自适应反馈控制。目前的自适应控制方法,像英国、瑞典MOV A SOS)和英国(孤立的信号(area-wide又控制),采用数学优化与仿真技术来调整信号波动的时间观察到的交通流实时的。优化是通过改变时间和周期长度的绿色的信号。在area-wide交叉口控制偏移是之间也发生了变化。已经开发为几种方法确定最优周期长度和最小延迟在十字路口,但基于不确定性和严格的交通信号控制的本质,全局最优是不可能找到的。 1．引文：由于越来越多的公众意识的环境影响道路交通许多当局现在所追求的政策来:，管理供求?拥挤，影响模式和路径选择?;贯彻“三个代表”重要思想,提高公共汽车?有轨电车和其他公共服务车辆;设施提供更好的、更安全,骑自行车和行人的道路使用者等脆弱;降低汽车排放?、噪声和视觉入侵;为所有道路改善安全?用户群。 在自适应交通信号控制的弹性增强的增加的数量在周期层叠的绿色阶段,从而使数学优化非常复杂和困难。因为这个原因,自适应信号控制在大多数情况下不是建立在精确的优化上,而是建立在绿色的扩展原理。在实践中,遵循的均匀性是最主要的交通信号控制安全的原因。这一规定的限制的周期时间和相位的安排。因此,在实践中是交通信号控制的针对性的解决方案和调整的基础上由交通规划者。现代可编程信号控制器以大量的可调参数是非常适合这一过程。对于好的结果,一个经验丰富的策划人和微调领域中是必要的。模糊控制已经被证明是成功的,在这些问题中,精确的数学建模是困难的或不可能的,但一名有经验的人可以控制的工艺操作。因此,交通信号控制是一种适合于任务特别为模糊控制。事实上,最古老的文化之一的潜力的例子是一个模拟的模糊控制在一个inter-section交通信号控制的两个单向的街道。即使在这个非常简单的情况下,模糊控制是至少在作为一个良好的传统的自适应控制。一般而言,模糊控制是发现在复杂问题都优于用多目标决策。在交通信号控制多种交通流竞争来自同一时间和空间,而且不同的优先选择往往不同交通流或车辆组。此外,优化标准,包括几个同时喜欢平均和最大车辆和行人延误、最大队列长度和百分比停止的车辆。所以,它很可能是很有竞争力的模糊控制在复杂真实的十字路口的地方传统的优化方法的使用是有问题的。 2．模糊逻辑：介绍了模糊逻辑,并成功地应用于大范围的自动控制任务。最大的好处模糊逻辑是有机会模型与不确定的模糊决策。此外,模糊逻辑有能力理解语言指令和控制策略的基础上产生的先验的沟通。这一点在利用模糊逻辑来控制理论的基础上,是模仿人类专家控制的知识,而不是为了构建过程本身。的确,模糊控制已经被证明是成功的,在这些问题中,精确的数学建模是困难的或不可能的,但一名有经验的操作员可以控制的过程。一般而言,模糊控制是发现在复杂问题都优于多目标决策。 目前,有大量的基于模糊推论系统技术。不过它们当中的主要部分,受含糊不清的根基;即使它们大都是古典数学方法表现更好,他们还带有黑色的盒子,如德模糊化,这是很难证明数学或逻辑的。例如,如果-然后模糊规则,它们在核心的模糊推理系统,经常报道的工作方式,是Ponens概括规则推理机制的经典,但随便起来就不是这样的,这之间的关系,这些规则和多值逻辑是任何已知的复杂和人工。此外,专家系统的性能应相当于人类专家:它应该得到同样的结果,专家给,但提醒当控制问题是如此模糊,专家是不确定适当的行为。现有的模糊专家系统很少满足这第二种情况。 然而,很多研究观察,模糊推理的方法是基于相似。Kosko,举个例子,写的模糊隶属……代表的相似性定义对象特性的imprecisely。以这句话严重,我们学习系统的多值等价,即模糊相
驱动桥设计 随着汽车对安全、节能、环保的不断重视，汽车后桥作为整车的一个关键部件，其产品的质量对整车的安全使用及整车性能的影响是非常大的，因而对汽车后桥进行有效的优化设计计算是非常必要的。 驱动桥处于动力传动系的末端，其基本功能是增大由传动轴或变速器传来的转矩,并将动力合理地分配给左、右驱动轮，另外还承受作用于路面和车架或车身之间的垂直力力和横向力。驱动桥一般由主减速器、差速器、车轮传动装置和驱动桥壳等组成。 驱动桥作为汽车四大总成之一，它的性能的好坏直接影响整车性能，而对于载重汽车显得尤为重要。驱动桥设计应当满足如下基本要求： 1、符合现代汽车设计的一般理论。 2、外形尺寸要小，保证有必要的离地间隙。 3、合适的主减速比，以保证汽车的动力性和燃料经济性。 4、在各种转速和载荷下具有高的传动效率。 5、在保证足够的强度、刚度条件下，力求质量小，结构简单，加工工艺性 好，制造容易，拆装，调整方便。 6、与悬架导向机构运动协调，对于转向驱动桥，还应与转向机构运动协调。智能电子技术在汽车上得以推广使得汽车在安全行驶和其它功能更上一层楼。通过各种传感器实现自动驾驶。除些之外智能汽车装备有多种传感器能充分感知交通设施及环境的信息并能随时判断车辆及驾驶员是否处于危险之中，具备自主寻路、导航、避撞、不停车收费等功能。有效提高运输过程中的安全，减少驾驶员的操纵疲劳度，提高乘客的舒适度。当然蓄电池是电动汽车的关键，电动汽车用的蓄电池主要有：铅酸蓄电池、镍镉蓄电池、钠硫蓄电池、钠硫蓄电池、锂电池、锌—空气电池、飞轮电池、燃料电池和太阳能电池等。在诸多种电池中，燃料电池是迄今为止最有希望解决汽车能源短缺问题的动力源。燃料电池具有高效无污染的特性，不同于其他蓄电池，其不需要充电，只要外部不断地供给燃料，就能连续稳定地发电。燃料电池汽车(FCEV)具有可与内燃机汽车媲美的动力性能，在排放、燃油经济性方面明显优于内燃机车辆。
中文2942字 本科生毕业设计（论文）外文资料翻译翻译资料名称（外文）： Highway and Airport Pavement Design 翻译资料名称（中文）： 公路和机场路面设计 学院：建筑工程学院系土木工程 专业：土木工程（道桥） 班级： 学号： 姓名： 指导教师： 完成日期：2012 年02 月20 日
Highway and Airport Pavement Design（Excerpt） T. F. Fwa National University of Singapore 1.Introduction Pavements are designed and constructed to provide durable all-weather traveling surfaces for safe and speedy movement of people and goods with an acceptable level of comfort to users. These functional requirements of pavements are achieved through careful considerations in the following aspects during the design and construction phases: (a) selection of pavement type, (b) selection of materials to be used for various pavement layers and treatment of subgrade soils, (c) structural thickness design for pavement layers, (d) subsurface drainage design for the pavement system, (e) surface drainage and geometric design, and (f ) ridability of pavement surface. The two major considerations in the structural design of highway and airport pavements are material design and thickness design. Material design deals with the selection of suitable materials for various pavement layers and mix design of bituminous materials (for flexible pavement) or portland cement concrete (for rigid and interlocking block pavements). These topics are discussed in other chapters of this handbook. This chapter presents the concepts and methods of pavement thickness design. As the name implies, thickness design refers to the procedure of determining the required thickness for each pavement layer to provide a structurally sound pavement structure with satisfactory performance for the design traffic over the selected design life. Drainage design examines the entire pavement structure with respect to its drainage requirements and incorporates facilities to satisfy those requirements. 2．Pavement Types and Materials 2.1 Flexible versus Rigid Pavement Traditionally, pavements are classified into two categories, namely flexible and rigid pavements. The basis for classification is the way by which traffic loads are transmitted to the subgrade soil through the pavement structure. As shown in Fig. 2.1, a flexible pavement provides sufficient thickness for load distribution through a multilayer structure so that the stresses and strains in the subgrade soil layers are
一个汉英机器翻译系统的 计算模型与语言模型* 刘群+詹卫东++常宝宝++刘颖+ (+中国科学院计算技术研究所二室北京100080) (++北京大学计算语言学研究所北京100871) 摘要：本文介绍我们所设计并实现的一个汉英机器翻译系统。在概要介绍本系统的主要目标和设计原则的基础上，着重说明系统的计算模型和语言模型，最后给出实验结果和进一步的打算。 关键词：自然语言处理机器翻译中文信息处理 一、引言 我国的机器翻译研究近年来取得了很大的发展。特别是英汉机器翻译系统的研制已经取得了较大的成功，达到了初步实用的阶段。相对而言，汉英机器翻译的研究却进展比较缓慢，离实用化还有相当的距离[1]。我们的目的是利用目前最新的计算机软件技术、相对成熟的机器翻译方法和先进的汉语语法理论，构造一个初步实用的汉英机器翻译系统。本文将对我们所开发的系统所采用的计算模型和语言模型作一个总体性的介绍，而不涉及过多的细节。 下面我们简要介绍一下本系统的几个主要设计原则： ⑴采用成熟的技术 我们的目的是构造一个真正实用的汉英机器翻译系统，因而在可供选择的若干技术路线面前，我们将尽量选用比较成熟的技术，而在现有技术难以解决问题时再尝试一些新技术。 ⑵开放的体系结构 开放的体系结构主要体现在系统的实现上所采用的软件构件技术[8]。整个系统采用一些相对独立的软件构件组成，因而可以方便地对系统进行修改、维护和扩充。翻译的过程严格按照独立分析、独立生成的原则进行组织，每一阶段的算法相互独立，对其中一个阶段算法的修改不会对其他算法造成影响。 ⑶方便的调试环境 本系统强调为语言工作者提供一个方便的调试环境。系统提供多窗口图形界面的知识库调试工具，支持课题组中多人同时通过网络对一个知识库进行操作。提供对翻译过程直观显示，用户可以清晰地看到翻译过程的每一步操作。提供翻译出错原因查找机制，用户 *本项目的研究受到863-306资助，合同号为863-306-03-06-2