Automatic Construction of Japanese WordNet

合集下载

金融用语缩写词

金融用语缩写表ABC: Agricultural Bank of China 中国农业银行ACD: Automatic Call Distribution 自动呼叫分配ACH: Automated Clearing House 自动清算所ADC: Acquisition, development and construction收购、开发与建设AMA: Advanced measurement approaches高级计量法AMC: Asset Management Company 资产管理公司APR: Annual percentage rate 年百分率APS: Assured Payment System 受保支付系统ASA Alternative standardized approach另外一种标准法ATM：Automatic Teller Machine 自动柜员机BIN: Bank Identification Number 发卡行机构代码BOC: Bank of China 中国银行CCB: China Construction Bank 中国建设银行CCF: Credit conversion factor信用转换系数CBRC: China Banking Regulatory Commission 中国银行业监督管理委员会CD: Certificate of Deposit 大额存单CD: Cash Dispenser: 自动提款机CDR: Cumulative default rate积累违约率CF: Commodities finance商品融资CHIBOR: China Inter-bank Offered Rate 中国银行间同业拆借利率CPI: Consumer Price Index 消费者物价指数CRM: Credit risk mitigation信用风险缓释技术CRM: Customer Relationship Management 客户关系管理CSR: Customer Service Representative 客服代表CTI: Computer Telephony Integration 计算机电话集成CVV: Card Verification Value Visa组织采用的卡片校验值DVP: Delivery Versus Payment 券款对付结算EAD: Exposure at default违约风险暴露ECA: Export credit agency出口信贷机构ECAI: External credit assessment institution外部信用评估机构EDC: Electronic Data Capture 转账POSEDI: Electronic Data Interchange 电子数据交换系统EL: Expected loss预期损失EMV: Euro Card/MasterCard/Visa IC运用技术标准EMEAP (Executives’Meeting of East Asia and Pacific Central Banks) 东亚及太平洋地区中央银行行长会议组织FDI: Foreign Direct Investment 外商直接投资FMI: Future margin income未来利差收入GDP: Gross Domestic Product 国内生产总值GNP: Gross National Product 国民生产总值HVCRE: High-volatility commercial real estate高波动性商业房地产ICBC: Industrial and Commercial Bank of China 中国工商银行IFTS: Interbank Fund Transfer System 行际资金转账系统IMF: International Monetary Fund 国际货币基金组织IOU: I Owe You 打白条IPRE: Income-producing real estate 创造收入的房地产IRB: approach Internal ratings-based approach内部评级法IVR Interactive Voice Response 交互式语音应答LGD: Loss given default违约损失率MDB Multilateral development bank多边发开银行MIS: Management Information System 管理信息系统NIF: Note issuance facility票据发行工具NPLs: Non-performing Loans 不良贷款NPAs: Non-performing Assets 不良资产OECD: Organization for Economic Cooperation and Development 经济合作和发展组织PBC: People’s Bank of China 中国人民银行PBX Private Branch eXchange 专用分组交换系统PIN: Personal Identification Number 个人身份识别码POS: Point of Sale 销售点终端PSE: Public sector entity共公部门企业QFII: Qualified Foreign Institutional Investor 合格的境外机构投资者RCCs: Rural Credit Cooperatives 农村信用合作社ROA: Return on Assets 资产回报率ROE: Returns on Equity 股权收益率RPI: Retail Price IndexRTGS: Real Time Gross Settlement 实时全额结算SAFE: State Administrative of Foreign Exchange 国家外汇管理局SDPC: State Development Planning Commission 国家发展计划委员会SET: Secure Electronic Transfer 安全电子交易协议SEZs: Special Economic Zones 经济特区SSL: Secure Sockets Layer 安全套接层协议SOCBs: State-owned Commercial Banks 国有独资商业银行SOEs: State-owned Enterprises 国有企业SWIFT: Society for Worldwide Interbank Financial Telecommunication 环球银行间金融通信系统TBS Telephone Banking System 电话银行系统TVEs: Township and Village Enterprises 乡镇企业UCCs: Urban Credit Cooperatives 城市信用合作社VAT: Value-added Tax 增值税WTO：World Trade Organization 世界贸易组织世界各国常用货币名称一览货币全称英文缩写中译名Afghani Af 阿富汗尼African Financial CommunityFranc CFAF （非洲金融共同体）法郎®1Albanian Lek Lek 阿尔巴尼亚列克Algerian Dinar DA 阿尔及利亚第纳尔Argentine Austral $a 阿根廷奥斯特拉尔Australian Dollar $A 澳大利亚元Austrian Schilling Sch 奥地利先令Bangladesh Taka TK 孟加拉国塔卡Belgium Franc BF 比利时法郎Brazilian Cruzado CZ$ 巴西克鲁扎多Brunei Dollar BR$ 文莱元Burmese Kyat Can$ 缅甸元Canadian Dollar Can$ 加拿大元Cooperation Franc CFAF 中非金融合作法郎®2 Chilean Peso CH$ 智利比索Chinese Renminbi Yuan RMB￥中国人民币元Colombian Peso Col$ 哥伦比亚比索Cuban Peso Cub$ 古巴比索Danish Krone DKr 丹麦克郎Deutsche Mark DM 德国马克Dutch florin Guilder 荷兰盾Egyptian Pound LE ￡E 埃及镑Equatorial Guinea Ekucle EK 赤道几内亚埃奎勒Ethiopian Birr Br 埃塞俄比亚比尔Ethiopian Dollor $Eth 埃塞俄比亚元Finish Markka Fmk 芬兰马克French Franc F.Fr FF 法国法郎Ghanaian Cedi￠加纳塞地Guinean Peso GP （几内亚比绍）几内亚比索Guinean Syli GS 几内亚西里Hong Kong Dollar HK$ 港元Indian Rupee Rs. 单数Re 印度卢比Indonesian Rupiah Rp 印度尼西亚盾（卢比）Iranian Rail Rls 伊朗里亚尔Iraqi Dinar ID 伊拉克第纳尔Israel Pound I￡以色列镑Italian Lira Lit 复数Lire 意大利里拉Japanese Yen ￥日元Jordanian Dinar JD 约旦第纳尔Kenya Shilling KSh 肯尼亚先令Korean won W 朝鲜圆元Korean (Pok) Won 韩国元Kuwaiti Dinar KD 科威特第纳尔Lao Kip K 老挝基普Lebanese Pound ￡L 黎巴嫩镑Libyan Dinar LD 利比亚第纳尔Macau Pataca Pat 澳门地区澳门元Malaysian Dollar M$ 马来西亚元Malaysian Ringgit M$ 马来西亚林吉特Mexican Peso Mex$ 墨西哥比索Mongolian Tugrik Tug 蒙古图格里克Morocco Dirham DH 摩洛哥迪拉姆Mozambique Excudo M Esc 莫桑比克埃斯库多Nepalese Rupee NRs 尼泊尔卢比New Taiwan Dollar 新台币New Zealand Dollar NZ$ 新西兰元Nicaraguan Cordoba C$ 尼加拉瓜科多巴Nigerian Naira ＃尼日利亚奈拉Nowegian Krone NKr, 复数Kroner 挪威克朗Pakistan Rupee PRs 巴基斯坦卢比Panamanian Balboa B 巴拿马波亚Peruvian Indi 秘鲁印地Philippine Peso P 菲律宾比索Pound Sterling ￡或￡Stg 英镑Russian Rouble R 或 rub, Rbl 俄罗斯卢布（独联体）Saudi Riyal SRls 沙特里亚尔Sierra Leonean Leone Le 塞拉利昂利昂Singapore Dollar S$ 新加坡元Somali Shilling Sh. So 索马里先令Southern Yemen Dinar YD 也门第纳尔Sri Lanka Rupee SRs 斯里兰卡卢比Swedish Krona SKr, 复数Kronor 瑞典克朗Syrian Pound LS 叙利亚镑Tanzanin Shilling Tsh 坦桑尼亚先令Thai Baht B 泰国铢Tunisian Dinar D 突尼斯第纳尔Uruguayan New Peso Nur$ 乌拉圭新比索U.S.Dollar U.S.$ 美元Venezuelan Bolivar $b 委内瑞拉玻利瓦Vietnamese dong D 越南盾Zaire Z 扎伊尔Zambian Kwacha K 赞比亚克瓦查Zimbabwe Rhodesian Dollar ZR$ 津巴布韦罗德西亚元®1共同采用此种货币的非洲国家计有：Benin 贝宁Burkina Faso (原名 Upper Volta) 布基纳法索（原名上沃尔特）Cote D' Ivoire [Ivory Cost] 科特迪瓦[象牙海岸]Mali 马里Niger 尼日尔Senegal 塞内加尔Togo 多哥®2共同采用此种货币的国家计有：Cameroon 喀麦隆Central Africa 中非（共和国）Chad 乍得Congo 刚果Gabon 加蓬。

Visual Computer manuscript No. (will be inserted by the editor) Convex Contouring of Volume

Visual Computer manuscript No.(will be inserted by the editor)Convex Contouring of Volumetric Data Tao Ju1,Scott Schaefer1,Joe Warren1Department of Computer Science,Rice UniversityThe date of receipt and acceptance will be inserted by the editorAbstract In this paper we present a fast,table-driven isosurface extraction technique on volumetric data.Un-like Marching Cubes or other cell-based algorithms,the proposed polygonization generates convex negative space inside individual cells,enabling fast collision detection on the triangulated isosurface.In our implementation, we are able to perform over2million point classiﬁca-tions per second.The algorithm is driven by an auto-matically constructed look-up table that stores compact decision trees by sign conﬁgurations.The decision trees determine triangulations dynamically by values at cell ing the same technique,we can perform fast, crack-free multi-resolution contouring on nested grids of volumetric data.The method can also be extended to ex-tract isosurfaces on arbitrary convex,space-ﬁlling poly-hedra.Keyword:contour,polygonization,implicit modeling 1IntroductionRecent advances in hardware technology of3D scanning and sensoring has brought about the generation of large-scale volumetric data such as MRI scans,CT scans and geological images.A common approach to visualize these volume datasets is to represent the data as implicit func-tions and construct polygonal approximations of isosur-faces,i.e.locus of points with some given function value. This process is often referred to as contouring.The con-toured surface partitions the whole volume into negative space(locus of points with lower function values)and positive space(locus of points with higher function val-ues).In many interactive applications,such as computer gaming and real-world simulations,navigation is con-ﬁned to the negative space,therefore fast operations for checking the side of the main subject(e.g.the navigator) with respect to the contoured surface becomes critical.These operations are often referred to as collision detec-tions.Traditionally,contouring algorithms consider the data volume in uniform cubical cells and polygonize isosur-faces in each non-empty cell,i.e.,cells that are inter-sected by the isosurface.By doing so,the entire nega-tive space is decomposed into sub-spaces within individ-ual cells.Assuming that the negative space models free space,checking whether a point lies inside the free space can be greatly accelerated if the negative sub-space is convex within the enclosing cell.Note that this point-classiﬁcation is the fundamental operation for computing collision detection with other convex objects.For exam-ple,an edge lies in the free space if for each cell it passes through,both endpoints of the line segment within that cell lie in the cell’s negative space(ﬁgure1left).This observation is not true if the negative space within the cell is non-convex(ﬁgure1right).Fig.1Edge classiﬁcation in a signed square with two dif-ferent contours.The dashed lines indicate the contours,the gray areas represent negative space,and the dark gray lineis the edge to be checked.The most widely used cell-based algorithm is the March-ing Cubes(MC)algorithm introduced by Lorensen and Cline[7].MC generates triangles on isosurfaces within each cell based on a pre-computed table of positive/negative patterns(referred to as sign conﬁgurations hereafter)2Tao Ju etal.Fig.2Triangulations from theof cell corners.MC became popular for its fast polygo-nization and easy implementation due to its table-drivenmechanism.However,for some sign conﬁgurations,MCdoes not produce polygonizations that are consistentin topology with neighboring cells,and thus result insurface discontinuities[4].This drawback has sparkledextensive research on dis-ambiguation solutions[1],[4],[5],[10]and strategies to generate topologically correctpolygonizations[9].Although these solutions generatetopologically consistent isosurfaces,the resulting nega-tive space inside each cell is sometimes disconnected,andthus not convex.In this paper,we propose a fast table-driven cell-basedcontouring method that generate topologically consis-tent contours,while preserving convexity of the nega-tive space within each cell.The algorithm depends on acomposite look-up table that associates each sign con-ﬁguration with a compact decision tree for dynamic tri-angulation of isosurface patches.The reason for the useof a decision tree is that the actual triangulation de-pends not only on the signs,but also on the magnitudeof corner values.The decision trees are pre-computed toensure a minimal number of tests in determining eachtriangulation.This technique can be extended without diﬃculty tomulti-resolution contouring.In most multi-resolutionapproaches,direct application of the cell-based contour-ing algorithm to a grid of non-uniform cells can resultin surface cracks,i.e.discontinuity between iso-surfacesgenerated from neighboring cells at diﬀerent resolutions.Various multi-resolution frameworks have been proposedfor contouring on non-uniform grids[3],[8],[11],[12],[14],yet they involve special crack-patching strategies.Bloomenthal[2]proposes an adaptive contouring methodwhich requires run-time face tracing to maintain con-sistent contours for neighboring cells.We show that byconstructing look-up tables for transition cells using theabove algorithms,the same table-driven contouring methodcan be applied to non-uniform data to generate crack-free isosurfaces.The remainder of this paper is organized as follows.Afterreviewing the table-driven Marching Cubes algrorithm,we present the convex contouring algorithm on uniformgrids.Then we explain the automatic construction oflook-up tables in more detail.Next we extend the pro-posed technique to non-uniform grids to produce crack-free contour surfaces.We conclude by discussing otherpossible extensions and applications.2Marching Cubes and Look-up TablesMarching Cubes is a popular algorithm for extractinga polygonal contour from volumetric data sampled ona uniform3D grid.For each cell on the grid,the edgesintersected with the iso-surface are detected from signsat cell corners.The algorithm then forms triangles byconnecting intersections on those edges(referred to asedge intersections hereafter).To speed up the process,ituses a look-up table that establishes triangulations foreach sign conﬁguration.Since there are8corners in acell,the look-up table contains256entries,four of whichare shown inﬁgure2.Using the look-up table,the Marching Cubes algorithmcontours a cell in two steps:1.Look up the triangulation in the table by the signconﬁguration,2.For each triangle,compute the exact location of eachvertex(edge intersection)from values at the cell cor-ners by linear interpolation.The Marching Cubes algorithm is fast because it uses ta-ble look-up to build polygonal contours.Unfortunately,the contoured surface generated by the original look-uptable in[7]may contain holes,since the triangular con-tour within each cell is not always consistent with thatof the neighboring cells(ﬁgure3).Although this problem wasﬁxed in later work,it re-veals another drawback of manually created tables.Theentries in the Marching Cubes’look-up table are con-structed by identifying15topologically distinct sign con-ﬁgurations and triangulating each case by hand.The lackof automation makes it susceptible to errors and diﬃ-cult to adapt to other polyhedrons.As we shall see,thelook-up tables used in convex contouring are constructedalgorithmically based on the topology and geometry ofgiven polyhedrons,and therefore minimizes possible er-rors.Convex3Fig.3Two the Marching Cubes topology.3Convex Contouring Using Look-up Tables The goal of convex contouring is to extract polygonal contours that enclose convex negative spaces within each cell.In this section,we will introduce the concept of con-vex contours and describe the proposed polygonization technique based on a pre-computed look-up table.Ex-amples of convex contouring will be presented together with performance comparison with the Marching Cubes algorithm.3.1Convex contourAssuming that the underlying implicit function is trilin-ear inside each cell,the convex hull of the negative space in a cell is the convex hull of all negative cell corners and edge intersections.The convex contour is the part of this convex hull that lies interior to the cell.In 2D,for exam-ple,the convex contour consists of interior line segments connecting edge intersections (i.e.,dashed lines in ﬁg-ure 1left).In 3D,the convex contour consists of interior triangles whose vertices are edge intersections (ﬁgure 4left).Together with the triangles that ﬁll the negative regions on cell faces (ﬁgure 4right),they constitute the convex hull of the negative space inside thecell.Fig.4Convex contour on the convex hull of the negative space.Note that the convex contour in a cell is outlined by linear convex contours on the faces of the cell (high-lighted in ﬁgure 4).Since the linear convex contour in a2D square is uniquely determined given the signs at the 4corners (allowing movement of the edge intersections along the edges),the polygonal convex contours from neighboring 3D cells always share the same linear con-tour on the common face.Hence convex contours always form topologically consistent isosurfaces.In ﬁgure 5,we show the convex contours for the same cells from ﬁgure 2.In comparison,we observe that the convex contour always encloses a connected,convex neg-ative space within the cell,whereas the contour from the Marching Cubes algorithm does not.Note that a convex contour is sometimes composed of multiple connected components,as shown in the rightmost cell of ﬁgure 5.Each of these connected piece of the contour is called a patch ,which may contain multiple holes (such as the center left cell in ﬁgure 5).Each hole on the patch is surrounded by a ring of linear convex contours that can be detected by the signs at the corners.Hence a look-up table for patch boundaries can be constructed automat-ically for each sign conﬁguration.3.2Triangulation and Decision treesUnfortunately,although the boundary of the patches on the convex contour is unique for each sign conﬁguration,the actual polygonization is not.In fact,diﬀerent scalar values at cell corners,which determine the location of edge intersections,may modify the shape of the convex hull and result in diﬀerent triangulation of the convex contour.As illustrated in ﬁgure 6,two cells that share the same sign conﬁguration,but with diﬀerent scalars at cell corners,result in diﬀerent triangulated convexcontours.Fig.6Triangulation in cells with diﬀerent scalar values at corners.Edge intersections are indicted by gray dots.To determine the correct triangulation based on the lo-cation of edge intersections,one could apply a full-scale convex-hull algorithm to compute the convex hull of the negative space.However,such algorithms are too gen-eral for our purposes,since we want only the part of this convex hull that lies interior to the cell.Instead,we can take advantage of the fact that the topology of the boundary is known for each patch on the convex contour.In fact,we can construct a set S of all possible triangu-lations for each patch.For example,ﬁgure 6shows the4Tao Ju etal.Fig.5Convex contours in cells withonly two possible triangulations for the convex contour ofthat sign conﬁguration,since the contour contains a sin-gle4-sided patch.According to Euler’s polygon divisionproblem[6],there are(2n−4)!(n−1)!(n−2)!ways to triangulate an−sided patch into n−2triangles.In fact,this size ofS can be further reduced if we realize that the verticesof these triangles are restricted toﬁxed cell edges,hencesome triangulations will never appear on the convex hullof negative space.We will discuss this process in detailwhen we describe how the look-up table is constructed.Now we can think of the triangulation problem as thefollowing:given the exact locations of the edge inter-sections,choose an appropriate triangulation from S sothat the triangles satisfy the convex hull property(i.e.,all edge intersections and negative corners lie on one sideof the triangle).Hence we need a fast method to diﬀer-entiate the correct triangulation from others by lookingat the edge intersections.Assuming that triangles on theconvex contour face inside the convex hull of the nega-tive space,we can do this by the following4-point test:given four distinct edge intersections V1,V2,V3,V4,if V4lies on the front-facing side of the triangle(V1,V2,V3),then the inverted triangle(V1,V3,V2)does not belongto the convex contour(ﬁgure7left).Similarly,triangles(V1,V2,V4),(V2,V3,V4)and(V3,V1,V4)do not satisfythe convex hull property.Otherwise,by symmetry,weclaim that triangles(V1,V2,V3),(V1,V4,V2),(V2,V4,V3)and(V1,V3,V4)do not lie on the convex contour(ﬁgure7right).Note that if all vertices lie on the same plane,either choice can be made.V1V2V3V4V1V2V3V4Fig.7Four-point test with vertices V1,V2,V3,V4.Each4-point test on the edge intersections rules outthose from the set S of all possible triangulations thatcontain any of the four back-facing triangles.Since anytwo triangulations diﬀer at least in the triangles sharedby one of the boundary edge,the correct triangulationcan be distinguished from every other triangulation in Sthrough appropriate4-point tests.For best performance,a decision tree can be built to distinguish each triangula-tion through a minimal set of tests.An example of sucha decision tree is shown inﬁgure8for a5-sided patch.At each node,the remaining triangulations are shownand the4-point test is represented by indices of the fourvertices(the order is shown at the top left corner).If thefourth vertex lies on the front-facing side of the triangleformed by theﬁrst three vertices(in order),we take theleft branch.Otherwise,we follow the right branch.Theprocess stops at a leaf node,where a single triangulationisleft.2,4,5,13,4,5,12,3,4,52,3,5,11,2,3,412345Fig.8A decision tree using4-point test for5-sided patches.Since the construction of decision trees is based on theoriginal set S,they can be pre-computed and optimizedfor every patch in each sign conﬁguration.As we shallsee,the maximum depth of all these decision trees is5and the average tree depth is1.88.In other words,thecorrect triangulation of the convex contour in a cubic cellcan be determined by performing a maximum of5point-face trials on the edge intersections,and on average nomore than2trials.3.3The look-up tableThe look-up table contains256entries,one for each signconﬁguration of a cubic cell.In each entry,the look-uptable stores the decision trees computed for each patchConvex Contouring of Volumetric Data5 Fig.9Two screen shots of a real-time navigation application using a convexly contoured terrain.by traversing the nodes in the decision trees in pre-order. Each non-leaf node contains a4-point test,and each leaf node stores a triangulation.The edge intersections in4-point tests and triangulations are represented by indices of the edges on which they lie.Two example entries in this table are shown in table1,with their correspond-ing sign conﬁgurations and edge indexing drawn on the right.3.4Contouring by table look-upsIn comparison with the Marching Cubes algorithm,con-vex contouring extracts the polygonal contour in a cell in two steps:1.Look up the decision trees(one for each patch)in thetable by the signs at the cell corners,2.For each decision tree,perform4-point tests on spec-iﬁed edge intersections until arriving at a single tri-angulation.Fig.10Convex contouring on volumetric data.Inﬁgure10,two sets of volumetric data generated by scan-conversion of polygonal models are contoured us-ing the new method.Observe that the edges on the sur-face are manifold and the contour is crack-less.Since the negative space inside each non-empty cell is convex, collision detection can be localized into cells and there-fore become independent of the grid size.Figure9shows two screen shots of a real-time navigation program in which the movement of the viewer is conﬁned within the negative space.The terrain is an iso-surface constructed using convex contouring on a256cubic grid.On a con-sumer level PC machine(dual1.5GHz processors with 2.5G memory),we achieved over2million point classiﬁ-cations per second.As we mentioned before,the average number of tests used to determine triangulation for each patch is less than2.Hence we can perform convex contouring on vol-umetric data with speed comparable to the Marching Cubes algorithm.In table2,the performance of con-vex contouring is compared with that of the Marching Cubes for contouring the terrain inﬁgure9on diﬀer-ent grid sizes.We also compared the average number of triangles generated in each cell in both methods.Notice that convex contouring generates on average only about 2%more triangles than the Marching Cubes algorithm. These extra triangles are needed to preserve the convex-ity of the negative space.Grid Size Marching Cubes Convex Contouring 1283125ms141ms2563781ms907ms51232109ms2437msAvg.Triangles 3.181 3.257Table2Comparison of total contouring time and average number of triangles per cell in convex contouring and the Marching Cubes.6Tao Ju et al.Index Table Entry Cell Conﬁguration171{1,9,12,7}→4-point test{{1,9,7},{9,12,7}}→Triangulation{{1,9,12},{1,12,7}}123456789101112125{1,3,6,12}→4-point test in parent node{1,12,10,2}→4-point test in left child{{1,3,12},{1,12,2},{3,6,12},{12,10,2}}{{1,3,12},{1,12,10},{1,10,2},{3,6,12}}{1,12,10,2}→4-point test in right child{{1,3,6},{1,6,12},{1,12,2},{12,10,2}}{{1,3,6},{1,6,12},{1,12,10},{1,10,2}}123456789101112Table1Two example entries in the look-up table.Each entry is a list of triangulations(each stored as a list of triangles) and4-point tests(each stored by the indices of the vertices)traversed from the decision tree in pre-roder.4Automatic Construction of Look-up TablesIn this section,we will review the table construction pro-cess in more detail.For a given sign conﬁguration,we ﬁrst detect the patch boundaries as a group of rings of linear contours.Then,for each patch,we construct the set of all possible triangulations that could take place on the convex hull of the negative space.Finally,an opti-mal decision tree is built for every patch detected on the contour.The look-up table can be found on the web at /jutao/research/contour tables/.4.1Detection of patch boundariesA patch on the convex contour in a cubic cell is bounded by linear convex contours on cell faces.These linear con-tours form single or multiple rings that surround the ”holes”of the patch.Since the linear convex contours are unique on each cell face for a given sign conﬁgura-tion,a single ring can be constructed using the following tracing strategy:starting from a cell edge that exhibits a sign change(where an edge intersection is expected) and facing the positive end,look for the next edge with a sign change by turning counter-clockwise on the face boundary.Repeat the search process until it returns to the starting edge,when a closed ring is formed(seeﬁgure 11).Notice that face-tracing produces oriented linear convex contours on each cell face.Each linear contour on the cell face is directed so that the negative region on that face lies to its left when looking from outside.There-fore the triangles on the convex contour of the cell that share these linear contours will face towards the nega-tive space.The orientation of the rings are importantfor linear contours already built,and dashed arrows indicate the tracing route.determining the set of possible triangulations that share the same patch boundary.Similar tracing techniques have been described by nu-merous authors in[2],[9],[13],in which convexity of the negative region on each cell face is preserved.These algo-rithms construct a single ring for each patch boundary. However,the problem remains on how to group multiple rings to form the boundary of a multiple-genus patch (such as the center left cell inﬁgure5).Although such patches arise in only4cases among256sign conﬁgura-tions,they may appear much more often in other non-cubic cells(such as transition cells in multi-resolution grids,see Section6).We need to be able to identify these cases automatically from the sign conﬁguration and group the rings appropriately.By deﬁnition,a patch is a continuous piece on the con-vex hull of negative space.Hence it also projects onto a continuous piece of regions on the cell faces.These regions are positive areas that surround positive cor-Convex Contouring of Volumetric Data7ners connected by cell edges.Therefore the boundary of each patch on the convex contour isolates a group of positive corners that are inter-connected by cell edges. The previous face-tracing algorithm could be modiﬁed so that rings constructed around a same edge-connected component of positive corners are grouped to form the boundary of a single patch.This rule is demonstrated in ﬁgure12.In the top left cell,two rings of linear contours form the boundary of two patches,due to the presence of two isolated positive corners.In the top right cell,where there is only one edge-connected component of positive corners,the two rings are grouped to form the bound-ary of a single cylinder-like patch.The connectivity of the positive corners in these two cells are illustrated at the bottom ofﬁgure12.In contrast,face-tracing without ring-grouping would give the same result in both cells, thus violating the convex hull property in the secondcell.Fig.lighted)to form boundaries of patches.Positive corners are colored gray.Bottom:Connectivity graph of positive corners 4.2Pre-triangulation of Convex ContourThe set of possible triangulations for a patch with an oriented boundary topology can be enumerated by re-cursive algorithms.However,this often results in redun-dant triangulations that never appear on the convex con-tour.For example,the genus-2patch in the center left cell inﬁgure5can be triangulated in only one way on the convex hull of the negative space,regardless of the magnitudes of scalar values at the corners.In contrast, brute-force enumeration would return21possible trian-gulations for an arbitrary genus-2patch with two trian-gular holes.The key observation is that,the patches on the convex contour are not arbitrary patches,their ver-tices(edge intersections)are restricted toﬁxed edges on the cell.These spacial restrictions limit the number of possible triangulations that could occur on the convex contour.For fast polygonization at run-time,we hope to pre-triangulate each patch as much as possible during table construction,based only on the sign conﬁguration. By the convex hull property,the half-space on the front-facing side of a triangle on the convex contour must con-tain(or partially contain)every other cell edge that ex-hibits a sign change.Note that the three vertices of the triangle can move only along threeﬁxed cell edges,this half-space is always contained in the union of the half-spaces formed when the three vertices are at the ends of their cell edges.Hence we have a way to identify triangles that will not appear on the convex contour:given three cell edges(C1,C2),(C3,C4)and(C5,C6)(C i are cell cor-ners),construct border triangles,i.e.,triangles formed by one end of each edge in order(such as(C1,C3,C5)).If there is an edge on the cell exhibiting a sign change that lies completely to the back of all non-degenerate border triangles,any triangle whose vertices belong to these three edges(in order)will not lie on the convex con-tour.This idea is illustrated inﬁgure13.The highlighted cell edge on the left exhibits a sign change,and lies to the back of all the border triangles constructed from the three dashed edges(E1,E2,E3).Hence the dashed trian-gle formed by intersections on those edges never appears on the convex contour.In contrast,the highlighted cell edge on the right lies partially to the front of at least one of the border triangles formed by the edges(E1,E2,E3), hence the dashed triangle may exist in the triangulation of the convex contour.E1E2E3E1E2E3Fig.13Identifying triangles that do not lie on the convex contour.By eliminating triangulations that contain these ineli-gible triangles,we can trim down the space of possible triangulations.For example,the number of remaining triangulations for theﬁrst three cells inﬁgure5are re-spectively4,1and4.4.3Construction of decision treesEven after pre-triangulation,some patches still have many potential triangulations on the convex hull of the nega-tive space.To speed up the polygonization at run-time,8Tao Ju et al.we can pre-compute a set of4-point tests on the vertices of the patch(edge intersections)to distinguish between the remaining triangulations.These tests can be orga-nized in a decision tree structure,introduced in the last section.Diﬀerent sets of tests result in diﬀerently shaped decision trees.To obtain optimal performance,we imple-ment a search algorithm that looks for the optimal set of tests that produces a decision tree with the small-est depth.Since each of the two outcomes of a single test eliminates a non-intersecting subset of the remain-ing triangulations,the minimal depth of the correspond-ing decision tree is lower bounded by log2N,where N is the total number of triangulations.For example,the depth of an optimized decision tree for a patch of length 4,5and6are1,3and5respectively.Further computa-tion reveals that the maximum length of a patch(which can not be pre-triangulated)in a cubic cell is6;hence any patch can be triangulated within5point-face trials on theﬂy by walking down the pre-computed decision tree.On average,however,it only takes1.88tests to de-termine the triangulation of a patch,due to infrequent occurrence of large patches and the reduced triangula-tion space as a result of pre-triangulation.5Multi-resolution Convex ContouringIn volume visualization,the number of polygons gener-ated by uniform contouring easily exceeds the capacity of modern hardware.For real-time applications,it is often advantageous to display the geometry at diﬀerent lev-els of detail depending on the distance from the viewer. This technique has the advantage that it speeds up the rendering process without sacriﬁcing much visual accu-racy.By using a view-dependent approach,the grid is contoured at diﬀerent resolutions depending on the dis-tance from the navigator.In particular,we can create a series of nested bounding boxes centered at the viewer, with the grid resolution decreasing by a factor of2.A2D example of this multi-resolution framework is shown inﬁgure14left,in which a circle is contoured using cell grid at two diﬀerent resolutions.When the coarse cells at the top meet theﬁne cells at the bottom,the con-tour in the coarse cells need to be consistent with the contour from the neighboringﬁne cells on the common edges.We call these coarse cells transition cells,which are adjacent to cells at aﬁner resolution.A2D transition cell thus hasﬁve corners andﬁve edges,as shown in the middle ofﬁgure14.By connecting edge intersections on each cell edge,the transition cells can be contoured in a way similar to a regular4-corner cell,yielding consistent contours with the adjacentﬁne cells.Some of the con-toured example are shown inﬁgure14right.Notice that the convexity of the negative region is still preserved in each transition cell.In3D,a transition cell between two resolutions is either adjacent to twoﬁne cells on an edge,or adjacent to four ﬁne cells on a face(seeﬁgure15left).These two types of cells can be regarded as convex polyhedrons with9 corners(ﬁgure15center)and13corners(ﬁgure15right) respectively.Since the previous discussion on regular cells applies to any convex polyhedron,we can also build look-up ta-bles and perform convex contouring on these transition cells.In this way,crack-free surfaces can be contoured on nested grids within the same framework as uniform contouring.5.1Convex contours in transition cellsAs in a regular cell,the convex contour inside a tran-sition cell is outlined by the linear contours on the cell faces.Since the linear convex contour is unique on each face for a given sign conﬁguration,the boundary of patches on the convex contour can be pre-computed using the proposed face-tracing algorithm.At the top ofﬁgure16, the oriented boundaries of patches in diﬀerent transition cells are detected and drawn as dashed arrows.Since con-vex contours from neighboring cells in a multi-resolution grid always share the same linear contour on the com-mon face as their boundaries,topological consistency is preserved everywhere on the contouredsurface.Fig.16Patch boundaries(top)and triangulation(bottom) in three transition cells.By pre-computing optimal decision trees for each patch, the triangulation can be determined by applying succes-sive4-point tests on the edge intersections.At the bot-tom ofﬁgure16,patches detected from the cells on the top are triangulated on the convex hull of the negative space.However,due to the presence of four co-planar faces on a13-corner cell,this convex hull could degener-ate onto a plane(ﬁgure17left).To determine the correct triangulation of the convex contour on the degenerate convex hull,we introduce an outward perturbation to。

Design and Construction of Concrete-Filled Steel Tube Column System in Japan

OUTLINE OF CFT COLUMN SYSTEM
Structural System
Figure 1 shows typical connections between a CFT column and H-shaped beams often used in Japan. The connection is fabricated by shop welding, and the beams are bolted to the brackets on-site. In the case of connections using inner and through-type diaphragms, the diaphragm plates are located inside the tube, and a hole is opened for concrete casting. A cast steel ring stiffener is used for a circular CFT column. In the case of a ring stiffener and an outer diaphragm, there is no object inside the tube to interfere with the smooth casting of the concrete. Concrete casting is usually done by Tremie tube or the pump-up method. High strength and ductility can be obtained in the CFT column system because of the advantages mentioned below. However, difficulty in properly compacting the concrete may create a weak point in the system, especially in the case of inner and through-type diaphragms where bleeding of the concrete beneath the diaphragm may produce a gap between the concrete and steel. There is currently no way to ensure compactness or to repair this deficiency. To compensate, high-quality concrete with a low water-content and a superplasticizer for enhanced workability is used inand Construction of Concrete-Filled Steel Tube Column System in Japan

2009年考研英语真题Text3解析

持续的经济增长促进教育的发展，
即使政府不强制推行教育。
suspect
continue
continual
continuous
怀疑，猜想
继续，连续
不断的，连续的
连续的，持续的
After all, that’s how education got started.
毕竟，最初的教育就是这样产生的
afterall
----a result of the training that U.S. workers received on the job.
然后，研究表明，
丰田、尼桑、本田（三家日本汽车商）的美国工厂
达到日本本土工厂生产率的约95%
这归功于对美国工人进行的在岗培训。
reveal
revelation
counterpart
政治上的政治家政治的政治家政治政见知识分子智力智力聪明聪明的明智的可理解的易懂的传统的常见的会议惯例优先重点促进提升迅速的有进展进步发展progressiveundoubtedlydoubtfortunatefortuneunfortunately进步的先进的毫无疑问地怀疑疑虑幸运的运气命运不幸地wearefortunatethatisitbecausebuildingneweducationalsystemsthereandputtingenoughpeoplethroughthemtoimproveeconomicperformancewouldrequiretwoorthreegenerations
formaleducation
formal
form
educate
education
economist
economic

Development and Impact of the Modern High-speed Train A Review

Transport Reviews, Vol. 26, No. 5, 593–611, September 20060144-1647 print/1464-5327 online/06/050593-19 © 2006 Taylor & FrancisDOI: 10.1080/01441640600589319Development and Impact of the Modern High-speed Train: A ReviewMOSHE GIVONIThe Bartlett School of Planning, University College London, London, UK(Received 8 August 2005; revised 14 November 2005; accepted 23 January 2006)A BSTRACT The inauguration of the Shinkansen high-speed train service between Tokyo and Osaka, Japan, at 210 kph maximum operating speed some 40 years ago marked the comeback of the train as an important passenger mode of transport. Since then high-speed train (HST) services have been introduced in many countries and are planned in many more, and the train has once more become the dominant mode of transport on many routes. This review summarizes the different elements of HST operation with the aim of characterizing HST operation and putting in context its impact in terms of what it is best designed for and what it can deliver. The review concludes that the HST is best designed to substitute conventional railway services on routes where much higher capac-ity is required and to reduce travel time, further improving the railway service, also against other modes, therefore leading to mode substitution. However, the high invest-ment in HST infrastructure could not be justified based on its economic development benefits since these are not certain. Finally, the following definition for HST services is suggested: high capacity and frequency railway services achieving an average speed of over 200 kph.IntroductionTransport technologies seldom make a comeback, save in nostalgia trips for well-heeled tourists. … But there is a spectacular exception: railways,written off thirty years ago as a Victorian anachronism destined to atro-phy before the steady growth of motorway traffic, have suddenly become one of the basic technologies of the twenty-first century.The reason of course is the high-speed train (Banister and Hall, 1993,p. 157)On 1 October 1964, the first high-speed train (HST) passenger service was launched on the Tokaido line between Tokyo and Osaka with trains running at Correspondence Address: Moshe Givoni, Department of Spatial Economics, Free University, De Boele-laan 1105, 1081 HV Amsterdam, The Netherlands. Email: mgivoni@feweb.vu.nl594M. Givonispeeds of 210 kph. This date marks the beginning of the modern HST era. Since then, the HST network has expanded, first in Japan, and later in other countries, and speeds have increased. Today, about 40 years later, the HST is in many respects a distinct mode of transport.There is no single definition for high speed in the context of railway services, although the reference is always to passenger services and not to freight. High speed can relate to the infrastructure capability to support high speed (this might explain the term ‘high-speed rail’ (HSR), in addition to the fact that train and rail (or railway) are often used synonymously), the rolling stock capability to achieve high speed and/or the actual operation speed achieved. The European Union (EU) definition, given in Directive 96/48 (European Commission, 1996a), is 250 kph for dedicated new lines and 200 kph for upgraded lines in respect of the infrastructure capabilities. The same applies to the rolling stock (on specially built and upgraded lines, respectively). With some HSTs operating at speeds of 350 kph, 200 kph might not seem high speed anymore. However, HST operation is not all about (maximum) speed. Other elements are as or even more important in the overall consideration of HST as a mode of transport and in the developments of HST lines. This review examines the different elements of HST operation and puts HST operation in context in terms of what the HST can deliver and what can be expected from the introduction of HST services. It is a synthesis of the existing literature and current state of the art. It begins by describing the main technologi-cal developments in railway technology required to operate HST services (the second section) followed by definition of four models of HST (third section). Next, the development of the HST network across the world is described (fourth section). The focus then shifts to analysing the impacts of HST services, first the transport impact (e.g. effect on travel time and modal share; the fifth section) is considered, followed by the spatial and socio-economic impacts (sixth section) and the environmental impact (seventh section). The cost of HST lines is exam-ined (eighth section) before conclusions are drawn (ninth section). Technological Evolution of the Present HSTTraditionally, a speed of 200 kph was considered as the threshold for ‘high speed’(Ellwanger and Wilckens, 1994), which was achieved in Germany in tests as early as 1903. In 1955, the French set a new speed record of 331 kph and they also hold the current speed record for a ‘steel wheel on steel rail’ train of 515 kph achieved in 1990 by a F rench TGV HST (Whitelegg and Holzapfel, 1993). However, the commercial speed that can be achieved is of greater importance. The maximum operating speed on the Tokaido line now stands at 270 kph (Central Japan Railway Company, 2003), while on the TGV Atlantique line trains operate at a maximum speed of 300 kph. The standard for new lines is even higher, at 350 kph, which is the official maximum operating speed of new HST lines such as the Madrid–Barcelona line (International Union of Railways, 2003). Higher oper-ating speeds seem commercially unfeasible at present due to noise problems, high operating costs and other technical problems.The modern HST uses the same basic technology of a steel wheel on a steel rail as the first trains did at the beginning of the 19th century. Yet, many incremental engineering and technological developments were required in all aspects of train operation to allow trains to run commercially at speeds higher than 200 kph. Although much higher speeds were reached in tests by simply using more powerDevelopment and Impact of the Modern High-speed Train595 to propel conventional trains, these speeds were ‘deemed infeasible for commer-cial application because the fast-moving vehicles damaged the tracks severely’(Raoul, 1997, p. 100). In addition, the increase in the centrifugal forces as speed increases when trains run through curved sections led to discomfort to passen-gers and, furthermore, it is not enough that the train is capable of running at high speed, but the track must supports trains running at high speeds.The main technical challenges in the development of commercial HSTs were to develop a train and track that could maintain stability and the comfort of passen-gers (while the train is running at high speed), maintain the ability to stop safely, avoid a sharp increase in (train) operating costs and (track) maintenance costs, and avoid an increase in noise and vibration to areas adjacent to the line. The solution included, in most cases, building tracks that avoid tight curves; increas-ing the distance between axles in the bogies to help maintain stability and placing the bogies between carriages (and not at the ends of each carriage) to reduce weight by halving the number of bogies required to carry the carriages; improv-ing stability by preventing the cars from pivoting away from one another on curves; designing aerodynamic trains to reduce drag and shaping the train in a way that reduces the noise and vibration it induces; and using lighter and stron-ger materials (Raoul, 1997). In addition, the high speed required improvements to the signalling systems, the introduction of automatic braking/decelerating systems to improve safety, and changes in the operation of trains, e.g. the need to replace roadside signals with signals inside the driver’s cab since at high speeds the trains passed the roadside signals too fast for the driver to see them.Main Models of HSTThe different needs and special characteristics of the different countries pursuing the development of HST operation led to the evolution of different models of HSTs. The Japanese Shinkansen, which was the first modern HST in operation, can be considered as the base model for HST. Subsequently, three other models have evolved.The ShinkansenThe main features of the Shinkansen (‘new trunk line’, and the name given to the Japanese HSTs) evolved from Japan’s unique characteristics which include large metropolitan centres located a few hundred kilometres apart from each other with a high demand for travel between them. F or example, the Tokaido line connects Tokyo, Osaka and Nagoya, Japan’s biggest cities (approximately 30, 16 and 8.5 million inhabitants, respectively), which are a few hundred kilometres apart from each other (Tokyo–Osaka 560 km with Nagoya located on the route 342 km from Tokyo) and generate high demand for travel between them (132 million passengers on the Tokaido Shinkansen in 2002; Central Japan Railway Company, 2003). A unique feature of the Shinkansen is the new dedicated line which in the case of Japan was required since the conventional railway network is narrow gauge and could not support HSTs. This isolates the Shinkansen services from the rest of the railway system in Japan. The geographic features of Japan together with the requirement to avoid tight curves and steep gradients (to allow for high speeds) resulted in many tunnels and bridges along the route, which is typical of the Shinkansen lines. A total of 30% of the Japanese Shinkansen lines596M. Givonirun through tunnels (Okada, 1994) leading to very high construction costs. Furthermore, construction of new lines into the city centres further exacerbates construction costs due to engineering complexity and the high land values in city centres.The TGVThe French TGV (Train à Grande Vitesse), which began operation in 1981, resem-bles the Shinkansen in purpose but differs in design philosophy. The differences are attributable to some extent to overcoming the disadvantages of the Shinkansen and to the different physical characteristics of France and Japan (Sone, 1994). The most significant difference between the TGV and the Shinkansen is probably the ability of the former to operate on conventional tracks as well, which allows the TGV to use the conventional lines as it enters and leaves the city centre, leading to significant cost savings. It also means that the HST can serve regions with no HST infrastructure and specifically serve parts of the network where at present the demand is not high enough to justify the construction of a dedicated line (Bouley, 1986).The Spanish HST, the AVE (Alta Velocidad Espanola, or Spanish High Speed), is a mix of the TGV and Shinkansen models. It uses a TGV-type rolling stocks, but like the Shinkansen it runs on a dedicated line throughout, because the Spanish conventional network is wider then the standard International Union of Railways (UIC) gauge used across most of Europe. This was favoured to allow the AVE to connect with the emerging European, and mainly French, HST network (Gómez-Mendoza, 1993).Germany’s HST, the ICE (Inter-City Express), follows the TGV model of HST, mainly in the compatibility feature. It deviates from the TGV and the Shinkansen models by adopting a mixed-use line, meaning the line is used for both passenger and freight transport (Bouley, 1986). This feature turned out to be a disadvantage since it led to high construction costs (to support the higher load of freight trains) and low utilization of the lines (since freight trains operate at much lower speeds).The Tilting HSTThe Japanese, French, Spanish and German HST trains all use a newly built track on the sections where high speed is achieved, which translates into high construc-tion costs. Yet, on many routes demand is not high enough to justify the cost of constructing new tracks that allow high-speed operation. This problem was solved by the tilting train model of HST, but at the price of lower speeds. To allow higher speeds on conventional lines with tight curves, the train tilts as it passes through curves. By simply tilting the train in tight radius curves (although by a sophisticated computerized mechanism), the discomfort passengers feel from the centrifugal force as the train goes at high speed through curves is solved. ‘The bogies remain firmly attached to the rails while the body of the carriage tilts, and so compensates for centrifugal force’ (Giuntini, 1993, p. 61). This principle is adopted by many countries as a cheaper alternative to the TGV and Shinkansen models of HST. The Swedish X-2000 and the Italian Pendolino (ETR-450) are examples of HSTs running on conventional rail using the tilting mechanism, thus avoiding the price of expensive new tracks, but reaching maximum speed of onlyDevelopment and Impact of the Modern High-speed Train597 210 kph (X-200) or 250 kph (ETR-450). Today, a tilting mechanism is also used on TGV trains, like the TGV Pendulaire, which can reach a maximum speed of 300 kph (International Union of Railways, 2003) and all the new Shinkansen models will adopt a tilting mechanism (Japan Railway and Transport Review, 2005).The MAGLEV HSTIf the tilting train model of HST is considered a downgrade from the Shinkansen and TGV models, mainly in terms of speed, the MAGLEV model of HST is an upgrade. Magnetic levitation (MAGLEV) technology was first tested in the 1970s, but it has never been in commercial operation on long-distance routes. The tech-nology relies on electromagnetic forces to cause the vehicle to hover above the track and move forward at theoretically unlimited speeds. In practice, the aim is for an operation speed of 500 kph (Taniguchi, 1993). In 2003, a MAGLEV test train achieved a world record speed of 581 kph (Takagi, 2005). The special infrastruc-ture required for MAGLEV trains means high construction costs and no compati-bility with the railway network. The MAGLEV is mostly associated with countries like Japan and Germany where MAGLEV test lines are in operation. In Japan, the test line will eventually be part of the Chuo Shinkansen between Tokyo and Osaka connecting the cities in about 1 hour compared with the present 2.5 hours. In China, a short MAGLEV line was opened in December 2003 connecting Shanghai Airport and the city’s Pudong financial district with trains running at maximum speed of 430 kph (International Railway Journal, 2004). However, plans to adopt MAGLEV technology for the planned Beijing–Shanghai route were abandoned in favour of a conventional steel wheel-on-steel rail HST (People’s Daily, 2004). The future of the MAGLEV, it seems, depends on its success in Japan, in the same way the development of the HST depended largely on the success of the first Shinkansen line.The main differences between the four models of HST described above in terms of maximum operating speed, compatibility with the conventional network and construction costs are summarized in F igure 1. In general, the reference in the present paper is to the Shinkansen and TGV models of HST.Development of the HST NetworkDespite the success of the Tokaido line in Japan, the spread of the HST around the world was relatively slow. It took 17 years after the opening of the Tokaido line for the first HST to be introduced outside Japan (in France) and another 7 years for the second European HST (and the world’s third) to begin service in Italy. At present, the HST covers much of Japan and Europe and is being introduced in the Far East. The USA in this respect is lagging behind (see below).In Japan, the success of the Tokaido line prompted the development of a Japanese HST network that was developed over time (Matsuda, 1993). This network (F igure 2) consists today of 2175 km of HST line in operation, with a further 215 km under construction and 349 km at the planning stage (International Union of Railways, 2003). The HST line between Paris and Lyon, which opened in 1981, was France’s first HST. This line also proved to be a success story, and in turn the driving force for developing the French HST (Polino, 1993). At present, the French HST network consists of 1541 km of HST lines in operation, 320 km under construction and another 937 km at the planning stage (International Union598M. GivoniFigure 1.Operating speed, construction cost and compatibility (with the conventional network)characteristics of the four high-speed train modelsof Railways, 2003). Despite the successful introduction of the HST in France, its spread across Europe was relatively slow. In 1988, the Italians introduced the Pendolino tilting train (ETR-450 model) on the Rome–Milan route; and in 2000 Sweden introduced its first HST, the X-2000, also a tilting-train model of HST. Next, in 1991, it was Germany’s turn to introduce the ICE between Hannover and Würzburg; and a year later Spain introduced its first HST service between Seville and Madrid. The UK joined the list of countries with HST lines in 2003 when the first phase of the link from London to the Channel Tunnel (the Channel Tunnel Rail Link, or CTRL) was opened. This line is scheduled for completion in 2007.Figure 2.Japan’s high-speed train networkDevelopment and Impact of the Modern High-speed Train599 When the above European countries were independently pursuing their HST plans, and France was pursuing the expansion of its HST network to neighbour-ing countries, the EU emerged and with it the idea of the Trans-European Networks (European Commission, 2001). The international dimension of the European HST network (F igure 3) increases the scope for HST services since many countries, e.g. Belgium and the Netherlands, do not have domestic routes that can justify HST services.Besides Japan and Europe, HST development takes place mainly in the F ar East. South Korea became the eighth country with trains operating at 300 kph in April 2004 when its HST, the KTX (Korea Train eXpress), which is a TGV model of HST, began operation. In 2005, Taiwan completed its 345-km HST line between Taipei and Kaohsiung, the first country to adopt Shinkansen technology outside Japan. China, as noted above, is also underway to join the countries operating HSTs with a planned line between Beijing and Shanghai (Takagi, 2005).In the USA only one HST line is in operation: the Acela Express tilting train running on the North East Corridor line between Boston and Washington, DC. At present, ten corridors across the USA have been designated for HST operation, but the start of construction still seems far way. The Californian HST, connecting the San F rancisco Bay area with Los Angeles and San Diego, is at the most advanced planning stages (F ederal Railroad Administration, 2005). Current debate in the USA on the future of Amtrak, the US passenger rail company, and the level of subsidies it should receive casts more doubts on the prospects for HST. With no funding for inter-city rail services, there are unlikely to be funding for HST projects, thus the fate of the proposed HST projects depends on state funding and governors’ will rather than federal funding. For an account of HST in the USA, or the lack of it, see Klein (1993) and Thompson (1994).In Australia, following a scoping study on an east coast HST network (2000 km long), connecting Melbourne, Canberra, Sydney and Brisbane (Department of Transport and Regional Services, 2001), the government decided in 2002 toFigure 3.Planned European high-speed train network. Source: International Union of Railways (2003)600M. Givoniconclude its investigation after the study showed such a network would be too expensive and that 80% of the funding would have to be public money (Anderson, 2002).With respect to MAGLEV lines, the Chuo Shinkansen between Tokyo and Osaka is at an advanced planning stage, and it still awaits the government’s green light (Takagi, 2005). There are currently no concrete plans for a MAGLEV line anywhere else.HST as a Mode of TransportThe main reason for the construction of the first Shinkansen and TGV lines was increasing route capacity, and this was also the case for many other lines. In Italy, capacity continues to be the main reason to construct high-speed lines, while the gain in speed will be relatively small, e.g. on the Rome to Naples line. In the UK, spare capacity on the rail network in the 1970s and 1980s was one of the main reasons for not considering HST development when other European countries started developing their HSTs (Commission for Integrated Transport, 2004).HST lines increase the capacity on the route since they usually supplement existing ones and they free capacity on the conventional network for use by freight and regional passenger services. The direct increase in capacity offered by the HST line is due to the higher frequency, which is feasible due to the higher speed and the most up-to-date signalling systems that allow relatively short headway between trains without compromising safety, and due to the use of long trains with high seat capacities. For example, the first Shinkansen, model 0, had a capacity of 1340 seats and Eurostar trains have 770 seats (International Union of Railways, 2003). The combination of a dedicated high-speed line, high-capacity trains, advanced signalling systems and high-speed enabled JR Central to carry on the Tokaido line 362 000 passengers per day on 287 daily services or a total of 132 million passengers in 2002 (Central Japan Railway Company, 2003). Higher frequency due to a higher speed and improved signalling also means that the introduction of tilting trains on existing tracks will lead to increased capacity on the route.Reducing travel time is also an important reason for introducing HST services, although not the main reason in most cases. Before the inauguration of the HST in Japan, it took 7 hours to travel between Tokyo and Osaka on the conventional line; it was then reduced to 4 hours following the inauguration of the Shinkansen and it is 2 hours 30 minutes since 1992 (Matsuda, 1993). If the green light will be given to the MAGLEV Chou Shinkansen, travel time between the cities will be further reduced to only 1 hour. The opening of the Spanish HST between Madrid and Seville reduced travel time from 6 hours 30 minutes to 2 hours 32 minutes (European Commission, 1996b) and there are many other similar examples. These time savings have an economic value which the Japanese estimate at approxi-mately €3.7 billion per year (Okada, 1994).The ability of the HST to cut travel time is determined by the average speed it achieves, which is affected mainly by the number of stops and the different speed restrictions along the route. Therefore, HSTs that have a very high maximum operating speed might still achieve a relatively low average speed and limited travel time savings due to the above. Japan and France provide the fastest services in the world at average speeds of around 260 kph (Takagi, 2005). The NozomiDevelopment and Impact of the Modern High-speed Train601 service on the Tokaido line, which stops only at the main stations along the route, achieves an average speed between Tokyo and Osaka of 206 kph. This is signifi-cantly lower than the maximum speed achieved by these HSTs.HST services can only be attractive on high-demand routes, which is why most HST services will be between city centres. Hence, cities with dense and dominant city centres (in terms of population and/or employment) are more attractive for HST services, unlike large cities which are more dispersed in nature. Since long access journey to the HST station might cancel the time savings the HST service offers (Hall, 1999) there is an incentive to provide more than one station per city. Yet, more stations/stops means a lower average speed and thus a trade-off must be made. Large metropolitan cities that are polycentric in nature can justify more than one HST station, e.g. Tokyo has three stations on the Tokaido line (Tokyo, Shinagawa and Shin-Yokohama). In contrast, the four stations planned on the CTRL (St Pancras and Stratford both in London, and Ebbsfleet and Ashford both in Kent) might not generate enough demand to justify the reduction in average speed and the longer travel time.Shorter travel times and an increased level of service (a higher frequency and also improved travelling conditions) following the introduction of HST lead to changes in the modal share on the route and to the generation of new demand. The modal share the HST captures depends mainly on the travel time it offers compared with other modes, but also on the cost of travel and travel conditions. Most of the demand shifted to the train mode following the introduction of HST services is from the aircraft and, to a lesser extent, from the car, which was the case on the Paris–Lyon and Madrid–Seville routes (Table 1). However, most of the demand for new HST services is demand shifted from the conventional railway, which has negative consequences for the conventional rail network (Vickerman, 1997). For example, on the Sanyo Shinkansen, 55% of the traffic was diverted to the new line from other rail lines (23% from the aircraft, 16% from the car and bus, and 6% new (induced) demand) (Sands, 1993b). In some cases, the traffic genera-tion effect of new HST services is substantial, such as on the Paris–Lyon and the Madrid–Seville lines (Table 1).González-Savignat (2004) observes that on relatively short routes the car has the highest modal share on the route, before the introduction of HST services, e.g. 71 and 82% on the Madrid–Zaragoza and Zaragoza–Barcelona routes, respectively, Table 1.Modal share (%) before and after the introduction of high-speed trainservicesTGV, Paris–Lyon line AVE, Madrid–Seville line Before (1981)After (1984)Change Before (1991)After (1994)Change Aircraft317−244013−27 Train407232165135 Car and bus2921−84436−8 Total10010037a10010035b a Total traffic increased by 37%. A total of 10% is related to the estimated trend of growth and 27% is considered as induced traffic.b Total traffic increased by 35%.Source: European Commission (1996).602M. Givoniboth 300 km apart. In this case, a similar percentage shift of passengers from car and aircraft to the HST means a greater shift from the car in absolute numbers. Furthermore, González-Savignat notes that in the Spanish experience, the social benefits from the diversion of passengers from the car to the HST are larger than diversion of passengers from the aircraft to the HST.In summary, by definition all HST lines fulfil the purposes of increasing the route capacity and reducing travel time. Higher capacity and travel speed lead to changes in the modal share, increasing the share of the train at the expense of the aircraft and the private car and diverting passengers from the conventional train to the HST. In addition, the introduction of HST services also leads to the generation of new demand on the route. All these sum the ‘transport’ effects of the HST. F or more evidence on the transport effect of the HST in Europe, see Vickerman (1997).HST as a Substitute to the AircraftMuch attention is given to the HST as a substitute to the aircraft and as a possible solution to the congestion and environmental problems faced by the air transport industry, although substituting the aircraft is not the main reason for introducing HSTs (for a literature review of the subject, see Givoni, 2005).Due to its speed and the location of most HST stations at the city centre, the HST can offer comparable or shorter travel times than the aircraft on some routes and can therefore substitute it. The travel time advantage depends on the average speed of the HST service and the distance each mode has to cover, since trains do not necessarily follow the direct route. F or example, on a journey between London and Paris, the HST passes almost 500 km while the aircraft only 380 km. In general, on routes of around 300 km, evidence shows that the introduction of HST services almost leads to a withdrawal of aircraft services (e.g. between Tokyo and Nagoya and between Brussels and Paris), while on routes of around 1000 km and above, the HST ceases to be a good substitute for the aircraft (e.g. between Tokyo and F ukuoka, 1070 km, the HST share of the traffic is only 10%). In between these distances, there is usually direct competition between the modes. In most cases, competition is also between the operators, the airlines and the railways. This competition means that HST services are added to the aircraft services and not really substituting them. On the London–Paris route the HST captures about 70% of the market (Eurostar, 2005), but the airlines still offer about 60 flights a day just between London Heathrow and Paris Charles de Gaulle (CDG) (Innovata, 2004), two of the most congested airports and one of the most congested routes in Europe (Central Office for Delay Analysis, 2005). However, the situation is different when HST services are introduced at large hub airports (together with the means for a fast and seamless transfer between the aircraft and the HST). In this case, many airlines choose to replace the aircraft with the HST on some routes, leading to a real mode substitution. Lufthansa adopts such mode substitution on the routes from Frankfurt airport to Stuttgart and Cologne, where the aircraft is substituted by the German ICE HST and operated by Deutsche Bahn. Air F rance uses the HST to replace the aircraft on routes from CDG to Brussels, and on other routes it uses it to complement the aircraft. Furthermore, Airlines such as Emirates, American Airlines and United Airlines use HST services from CDG to complement their flights into Paris (International Air Transport Association, 2003).。

AUTOMATIC CONSTRUCTION

AUTOMATIC CONSTRUCTION OF ENGILSHCHINESE TRANSLATION LEXIOCN FROM PARALLEL SPOKEN LANGUAGE CORPUS
Bo-Xing CHEN Center for Speech Interaction Technology Research Institute of Acoustics, Chinese Academy of Sciences
1.2 Necessity of Constructing Translation Lexicon
At the same time, the automatic construction of translation lexicons (phrase and word alignment) from bilingual text has also made great progress. The algorithm of phrase and word alignment makes use of paraphrase dictionary information, etyma information, co-occurrence probability information and context heterogeneity, etc. or the combinations of this several information. For example, Fung[6] used a method as Pattern Matching to pick-up nouns and proper nouns from bilingual text; Kumano and Hirakawa[8] used the statistic and grammatical information to construct translation lexicons from aligned bilingual text; Wu and Xia[9], Hiemstra[10], Smadja[11] et al. all have studied the statistic method to construct translation lexicons from aligned bilingual text; there into, since 1995 Melamed [12][13][14][15][16][17] began to publish a series of articles on constructing translation lexicons from bilingual text through the method of counting co-occurrence probability. Although there has been a large amount of research for automatic construction of translation lexicons, the research for translation lexicons based on Chinese-English bilingual corpus is not very much, especially rare for spoken language corpus. And

各个学科英文单词

最新各个学科的英文单词中文学科、专业名称英文学科、专业名称哲学 Philosophy哲学 Philosophy马克思主义哲学Philosophy of Marxism 中国哲学Chinese Philosophy外国哲学Foreign Philosophies逻辑学Logic伦理学Ethics美学 Aesthetics宗教课Science of Religion科学技术哲学Philosophy of Science and Technology西方经济学Western Economics世界经济World Economics人口、资源与环境经济学Population,ResourcesandEnvironmental Economics应用经济学Applied Economics公民经济学National Economics经济学Economics理论经济学Theoretical Economics政治经济学Political Economy经济思想史History of Economic Thought经济史History of Economic地区经济学Regional Economics财政学 (含税收学 ) Public Finance (including Taxation)金融学 (含保险学 ) Finance (including Insurance)家产经济学Industrial Economics国际贸易学International Trade劳动经济学Labor Economics统计学Statistics数目经济学Quantitative Economics中文学科、专业名称英文学科、专业名称国防经济学National Defense Economics法学 Law法学 Science of Law法学理论Jurisprudence法律史Legal History宪法学与行政法学Constitutional Law and Administrative Law刑法学Criminal Jurisprudence民商法学 ( 含劳动法学、社会保障法学 ) Civil Law and Commercial Law (including Science of Labour Law and Science of Social Security Law )诉讼法学Science of Procedure Laws经济法学Science of Economic Law环境与资源保护法学Science of Environment and NaturalResources Protection Law国际法学(含国际公法学、国际私法学、国际经济法学、) International law (including International Public law, International Private Law and International Economic Law)军事法学Science of Military Law政治学Political Science政治学理论Political Theory中外政治制度Chinese and Foreign Political Institution科学社会主义与国际共产主义运动Scientific Socialism and InternationalCommunist Movement中共党史 (含党的学说与党的建设 ) History of the Communist Party of China(including the Doctrine of China Party and Party Building)马克思主义理论与思想政治教育Education of Marxist Theory and Education in Ideology and Politics国际政治学International Politics国际关系学International Relations外交学Diplomacy社会学Sociology社会学Sociology人口学Demography人类学Anthropology风俗学 (含中公大众文学) Folklore (including Chinese Folk Literature)民族学Ethnology民族学Ethnology马克思主义民族理论与政策Marxist Ethnic Theory and Policy中国少量民族经济Chinese Ethnic Economics中国少量民族史Chinese Ethnic History中国少量民族艺术Chinese Ethnic Art教育学Education教育学Education Science教育学原理Educational Principle课程与教课论Curriculum and Teaching Methodology教育史History of Education比较教育学Comparative Education学前教育学Pre-school Education高等教育学Higher Education成人教育学Adult Education职业技术教育学Vocational and Technical Education特别教育学Special Education教育技术学Education Technology心理学Psychology根基心理学Basic Psychology展开与心理学Developmental and Educational Psychology 应专心理学Applied Psychology体育学Science of Physical Culture and Sports体育人文社会学Humane and Sociological Science of Sports运感人体科学Human Movement Science体育教育训练学Theory of Sports Pedagogy and Training民族传统体育学Science of Ethnic Traditional Sports文学 Literature中国语言文学Chinese Literature文艺学Theory of Literature and Art语言学及应用语言学Linguistics and Applied Linguistics汉语言文字学Chinese Philology中国古典文件学Study of Chinese Classical Text中国古代文学Ancient Chinese Literature中国现当代文学M odern and Contemporary Chinese Literature中国少量民族语言文学Chinese Ethnic Language andLiterature比较文学与世界文学Comparative Literature and World Literature外国语言文学Foreign Languages and Literatures英语语言文学English Language and Literature俄语语言文学Russian Language and Literature法语语言文学French Language and Literature德语语言文学German Language and Literature日语语言文学Japanese Language and Literature印度语言文学Indian Language and Literature西班牙语语言文学Spanish Language and Literature阿拉伯语语言文学Arabic Language and Literature欧洲语言文学European Language and Literature亚非语言文学Asian-African Language and Literature外国语言学及应用语言学L inguistics and Applied Linguistics inForeign Languages新闻流传学Journalism and Communication新闻学Journalism流传学Communication艺术学Art艺术学Art Theory音乐学Music美术学Fine Arts设计艺术学Artistic Design戏剧戏曲学Theater and Chinese Traditional Opera 电影学Film播放电视艺术学Radio and television Art舞蹈学Dance历史学History历史学History史学理论及史学史Historical Theories and History of Historical Science考古学及博物馆学Archaeology and Museology历史地理学Historical Geography历史文件学 ( 含敦煌学、古文字学 ) Studies of Historical Literature (includingPaleography and Studies of Dunhuang)特意史History of Particular Subjects中国古代史Ancient Chinese History中国近现代史Modern and Contemporary Chinese History世界史World History理学 Natural Science数学 Mathematics根基数学Fundamental Mathematics计算数学Computational Mathematics概率论与数理统计Probability and Mathematical Statistics 应用数学Applied mathematics运筹学与控制论Operational Research and Cybernetics物理学Physics理论物理Theoretical Physics粒子物理与原子核物理Particle Physics and Nuclear Physics 原子与分子物理Atomic and Molecular Physics等离子体物理Plasma Physics凝集态物理Condensed Matter Physics声学 Acoustics光学 Optics无线电物理Radio Physics化学 Chemistry无机化学Inorganic Chemistry剖析化学Analytical Chemistry有机化学Organic Chemistry物理化学 ( 含化学物理 ) Physical Chemistry(includingChemical Physics)高分子化学与物理Chemistry and Physics of Polymers天文学Astronomy天体物理Astrophysics天体丈量与天体力学Astrometry and Celestial Mechanics地理学Geography自然地理学Physical Geography人文地理学Human Geography地图学与地理信息系统Cartography and Geography Information System大气科学Atmospheric Sciences气象学Meteorology大气物理学与大气环境Atmospheric Physics andAtmospheric Environment大海科学Marine Sciences物理大海学Physical Oceanography大海化学Marine Chemistry大海生理学Marine Biology大海地质学Marine Geology地球物理学Geophysics固体地球物理学Solid Earth Physics空间物理学Space Physics地质学Geology矿物学、岩石学、矿床学Mineralogy, Petrology, Mineral Deposit Geology地球化学Geochemistry古生物学与地层学 ( 含古人类学 )Paleontology and Stratigraphy (includingPaleoanthropology)结构地质学Structural Geology第四纪地质学Quaternary Geology生物学Biology植物学Botany动物学Zoology生理学Physiology水生生物学Hydrobiology微生物学Microbiology神经生物学Neurobiology遗传学Genetics发育生物学Developmental Biology细胞生物学Cell Biology生物化学与分子生物学Biochemistry and Molecular Biology 生物物理学Biophysics生态学Ecology系统科学Systems Science系统理论Systems Theory系统剖析与集成Systems Analysis and Integration科学技术史History of Science and Technology工学 Engineering力学 Mechanics一般力学与力学根基General and Fundamental Mechanics固体力学Solid Mechanics流体力学Fluid Mechanics工程力学Engineering Mechanics机械工程Mechanical Engineering机械制造及其自动化Mechanical Manufacture and Automation机械电子工程Mechatronic Engineering机械设计与理论Mechanical Design and Theory车辆工程Vehicle Engineering光学工程Optical Engineering仪器科学与技术Instrument Science and Technology精细仪器及机械Precision Instrument and Machinery测试计量技术及仪器Measuring and Testing Technologies and Instruments资料科学与工程Materials Science and Engineering资料物理与化学Materials Physics and Chemistry资料学Materialogy资料加工工程Materials Processing Engineering冶金工程Metallurgical Engineering冶金物理化学Physical Chemistry of Metallurgy钢铁冶金Ferrous Metallurgy有色金属冶金Non-ferrous Metallurgy动力工程及工程热物理Power Engineering and Engineering Thermophysics工程热物理Engineering Thermophysics热能工程Thermal Power Engineering动力机械及工程Power Machinery and Engineering流体机械及工程Fluid Machinery and Engineering制冷及低温工程Refrigeration and Cryogenic Engineering 化工过程机械Chemical Process Equipment电气工程Electrical Engineering电机与电器Electric Machines and Electric Apparatus电力系统及其自动化Power System and its Automation高电压与绝缘技术High Voltage and Insulation Technology电力电子与电力传动Power Electronics and Power Drives电工理论与新技术Theory and NewTechnology ofElectrical Engineering电子科学与技术Electronics Science and Technology物理电子学Physical Electronics电路与系统Circuits and Systems微电子学与固体电子学Microelectronics and Solid State Electronics第 23页共 50页Technology信息与通信工程Information and Communication Engineering通讯与信息系统Communication and Information Systems信号与信息办理Signal and Information Processing控制科学与工程Control Science and Engineering控制理论与控制工程 Control Theory and Control Engineering检测技术与自动化装置D etection Technology and Automatic Equipment系统工程Systems Engineering模式辨别与智能系统Pattern Recognition and Intelligent Systems导航、制导与控制Navigation, Guidance and Control计算机科学与技术Computer Science and Technology计算机软件与理论Computer Software and Theory计算机系统结构Computer Systems Organization计算机应用技术Computer Applied Technology建筑学Architecture建筑历史与理论Architectural History and Theory建筑设计及其理论Architectural Design and Theory城市规划与设计 (含景色园林规划与设计 ) Urban Planning and Design (including Landscape Planning and Design)建筑技术科学Building Technology Science土木匠程Civil Engineering岩土工程Geotechnical Engineering结构工程Structural Engineering市政工程Municipal Engineering供热、供燃气、通风及空调工程Heating, Gas Supply, Ventilating and Air Conditioning Engineering防灾减灾工程及防护工程Disaster PreventionandReduction Engineering and Protective Engineering桥梁与地道工程Bridge and Tunnel Engineering水利工程Hydraulic Engineering水文学及水资源Hydrology and Water Resources水力学及河流动力学Hydraulics and River Dynamics水工结构工程Hydraulic Structure Engineering第 26页共 50页水利水电工程Hydraulic and Hydro-Power Engineering港口、海岸及近海工程Harbor, Coastal and Offshore Engineering测绘科学与技术Surveying and Mapping大地丈量学与丈量工程Geodesy and Survey Engineering拍照丈量与遥感Photogrammetry and Remote Sensing地图制图学与地理信息工程Cartography and Geographic Information Engineering化学工程与技术Chemical Engineering and Technology化学工程Chemical Engineering化学工艺Chemical Technology生物化工Biochemical Engineering应用化学Applied Chemistry工业催化Industrial Catalysis地质资源与地质工程Geological Resources and Geological Engineering矿产普查与勘探Mineral Resource Prospecting and Exploration地球探测与信息技术Geodetection and Information Technology地质工程Geological Engineering矿业工程Mineral Engineering采矿工程Mining Engineering矿物加工工程Mineral Processing Engineering安全技术及工程Safety Technology and Engineering石油与天然气工程Oil and Natural Gas Engineering油气井工程Oil-Gas Well Engineering油气田开发工程Oil-Gas Field Development Engineering油气储运工程Oil-Gas Storage and Transportation Engineering纺织科学与工程Textile Science and Engineering纺织工程Textile Engineering纺织资料与纺织品设计Textile Material and Textiles Design纺织化学与染整工程Textile Chemistry and Dyeing and Finishing Engineering服饰设计与工程Clothing Design and Engineering轻工技术与工程 The Light Industry Technology and第 29页共 50页Engineering制浆造纸工程Pulp and Paper Engineering制糖工程Sugar Engineering发酵工程Fermentation Engineering皮革化学与工程Leather Chemistry and Engineering交通运输工程Communication and Transportation Engineering道路与铁道工程Highway and Railway Engineering交通讯息工程及控制Traffic Information Engineering & Control交通运输规划与管理 Transportation Planning and Management载运工具运用工程Vehicle Operation Engineering第 30页共 50页船舶与大海工程Naval Architecture and Ocean Engineering船舶与大海结构物设计制造 Design and Construction of Naval Architecture and Ocean Structure轮机工程Marine Engine Engineering水声工程Underwater Acoustics Engineering航空宇航科学与技术Aeronautical andAstronauticalScience and Technology飞翔器设计Flight Vehicle Design航空宇航推动理论与工程Aerospace Propulsion Theory and Engineering航空宇航器制造工程ManufacturingEngineering ofAerospace Vehicle人机与环境工程Man-Machine and Environmental第 31页共 50页Engineering兵器科学与技术Armament Science and Technology武器系统与运用工程Weapon Systems and Utilization Engineering兵器发射理论与技术Armament LaunchTheory andTechnology火炮、自动武器与弹药工程Artillery, Automatic Gun and Ammunition Engineering军事化学与烟火技术Military Chemistry and Pyrotechnics核科学与技术Nuclear Science and Technology核能科学与工程Nuclear Energy Science and Engineering核燃料循环与资料Nuclear Fuel Cycle and Materials核技术及应用Nuclear Technology and Applications第 32页共 50页辐射防护及环境保护 Radiation and Environmental Protection农业工程Agricultural Engineering农业机械化工程Agricultural Mechanization Engineering农业水土工程Agricultural Water-Soil Engineering农业生物环境与能源工程Agricultural Biological Environmental and Energy Engineering农业电气化与自动化AgriculturalElectrification andAutomation林业工程Forestry Engineering丛林工程Forest Engineering木材科学与技术Wood Science and Technology第 33页共 50页林产化学加工工程Chemical Processing Engineering of Forest Products环境科学与工程Environmental Science and Engineering环境科学Environmental Science环境工程Environmental Engineering生物医学工程Biomedical Engineering食品科学与工程Food Science and Engineering食品科学Food Science粮食、油脂及植物蛋白工程Cereals, Oils and Vegetable Protein Engineering农产品加工及贮藏工程Processing andStorage ofAgriculture Products水产品加工及储藏工程Processing and Storage of Aquatic第 34页共 50页Products农学 Agriculture作物学Crop Science作物种植学与耕种学Crop Cultivation and Farming System作物遗传育种学Crop Genetics and Breeding园艺学Horticulture果树学Pomology蔬菜学Olericulture茶学 Tea Science农业资源利用学Utilization Science of Agricultural Resources土壤学Soil Science植物营养学Plant Nutrition植物保护学Plant Protection植物病理学Plant Pathology农业昆虫与害虫防治Agricultural Entomology and Pest Control农药学Pesticide Science畜牧学Animal Science动物遗传育种与繁殖Animal Genetics, BreedingandReproductionScience动物营养与饲料科学Animal Nutrition and Feed Science草业科学第 36页共 50页特种经济动物饲养学 (含蚕、蜂等 ) The Rearing of Special-type EconomicAnimals (including Silkworm, Honeybees, etc.)兽医学Veterinary Medicine根基兽医学Basic Veterinary Medicine预防兽医学Preventive Veterinary Medicine临床兽医学Clinical Veterinary Medicine林学 Forestry林木遗传育种学Forest Tree Genetics and Breeding丛林培养学Silviculture丛林保护学Forest Protection丛林经理学Forest Management野生动植物保护与利用 Wildlife Conservation and Utilization园林植物与赏识园艺Ornamental Plants and Horticulture水土保持与沙漠化防治 Soil and Water Conservation and Desertification Combating水产学Fisheries Science水产养殖学Aquaculture Science捕捞学Fishing Science渔业资源学Science of Fisheries Resources医学 Medicine根基医学Basic Medicine人体解剖与组织胚胎学Human Anatomy, Histology and Embryology免疫学Immunology病原生物学Pathogenic Organisms病理学与病理生理学Pathology and Pathophysiology法医学Forensic Medicine放射医学Radiation Medicine航空航天与航海医学Aerospace and Nautical medicine临床医学Clinical Medicine内科学 (含心血管病学、血液病学、呼吸系病学、消化系病学、内分泌与代谢病学、肾脏病学、风湿病学、传得病学) Internal medicine(including Cardiology,Hematology,Respiratory, Gastroenterology,Endocrinology and Metabolism, Nephrology, Rheuma-tology, Infectious Diseases)儿科学Pediatrics老年医学Geriatrics精神病学Neurology精神病与精神卫生学Psychiatry and Mental Health皮肤病与性病学Dermatology and Venereology影像医学与核医学Imaging and Nuclear Medicine临床查验诊疗学Clinical Laboratory Diagnostics护理学Nursing外科学(含一般外科学、骨外科学、泌尿外科学、胸心血管外科学、神经外科学、整形外科学、烧伤外科学、野战外科学 ) Surgery (General Surgery,Orthopedics,Urology,Cardiothoracic Surgery, Neurosurgery,Plastic Surgery, Burn Surgery, Field Surgery)妇产科学Obstetrics and Gynecology眼科学Ophthalmic Specialty耳鼻咽喉科学Otolaryngology肿瘤学Oncology痊愈医学与理疗学Rehabilitation Medicine & Physical Therapy运动医学Sports Medicine麻醉学Anesthesiology急诊医学Emergency Medicine口腔医学Stomatology口腔根基医学Basic Science of Stomatology口腔临床医学Clinical Science of Stomatology公共卫生与预防医学Public Health and P reventive Medicine流行病与卫生统计学Epidemiology and Health Statistics劳动卫生与环境卫生学Occupational and Environmental Health营养与食品卫生学Nutrition and Food Hygiene儿少卫生与妇幼保健学Maternal, Child and Adolescent Health卫生毒理学Hygiene Toxicology军事预防医学Military Preventive Medicine中医学Chinese Medicine中医根基理论Basic Theories of Chinese Medicine中医临床根基Clinical Foundation of Chinese Medicine中医医史文件History and Literature of Chinese Medicine方子学Formulas of Chinese Medicine中医诊疗学Diagnostics of Chinese Medicine中医内科学Chinese Internal Medicine中医外科学Surgery of Chinese Medicine中医骨伤科学Orthopedics of Chinese Medicine中医妇科学Gynecology of Chinese Medicine中医儿科学Pediatrics of Chinese Medicine中医五官科学Ophthalmology and Otolaryngoloy of Chinese Medicine。

丰田跨国战略研究英文

Vios small car's main rival is wide of
the SDL, Mazda 2, Jin-xiang and AccFast polo.
SDL
is positioned 30 to 35-year-old age group, the cause of an
increase in development period, in a society of "core" class, is the
Unique advertising slogan
“Where there is a road, there is a TOYOTA car”
Manufacturing capability development
TOYOTA has a number of technology patents
Thereby the profit can be increased.
the target cost system
Through value engineering analysis it can ensure that the product attractive enough from price performance ratio
Toyota's car brand
Toyota's market share
In Japan was 45.3%, In the United States there are 16.8%, In the world market share of 12.7%.
Competing products of its
FAW Toyota's main competitors are :

1、下载文档前请自行甄别文档内容的完整性，平台不提供额外的编辑、内容补充、找答案等附加服务。
2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
3、如文档侵犯您的权益，请联系客服反馈,我们会尽快为您处理(人工客服工作时间：9:00-18:30)。

c Construction of Japanese WordNet
Hiroyuki Kaji and Mariko Watanabe
Department of Computer Science, Shizuoka University 3-5-1 Johoku, Hamamatsu-shi 432-8011, Japan kaji@inf.shizuoka.ac.jp
Abstract
Although WordNets have been developed for a number of languages, no attempts to construct a Japanese WordNet have been known to exist. Taking this into account, we launched a project to automatically translate the Princeton WordNet into Japanese by a method of unsupervised word-sense disambiguation using bilingual comparable corpora. The method we propose aligns English word associations with those in Japanese and iteratively calculates a correlation matrix of Japanese translations of an English word versus its associated words. It then determines the Japanese translation for the English word in a synset by calculating scores for translation candidates according to the correlation matrix and the associated words appearing in the gloss appended to the synset. This method is not robust because a gloss only contains a few associated words. To overcome this difficulty, we extended the method so that it retrieves texts by using the gloss as a query and uses the retrieved texts as well as the gloss to calculate scores for translation candidates. A preliminary experiment using Wall Street Journal and Nihon Keizai Shimbun corpora demonstrated that the proposed method is promising for constructing a Japanese WordNet.
Our method calculates the correlations between Japanese translations of an English word and its associated words by aligning English word associations with those in Japanese, where English and Japanese word associations are extracted from the corpora of both languages (See solid arrows in Fig. 1). Given an English word in a synset, it calculates the score for each of its Japanese translation candidates according to the gloss appended to the synset; the score is defined as the sum of correlations between the translation candidate and the associated words appearing in the gloss (See dotted arrows in Fig. 1). It should be noted that the proposed method does not require parallel corpora, whose availability is limited. The method can be applied to a pair of weakly comparable corpora in English and Japanese, e.g., the Wall Street Journal and the Nihon Keizai Shimbun. In addition, it is fully unsupervised; it does not require any manual input such as sense-tagging on the training corpora. It thus has the potential for allowing a large number of English synsets to be automatically translated into Japanese. However, the method has inherent weaknesses in that it lacks robustness because a gloss only contains a few associated words. To overcome this difficulty, we extended it so that it retrieves texts by using a gloss as a query and uses associated words contained in the retrieved texts to determine the translations for the words in the synset to which the gloss is appended (See dashed arrows in Fig. 1).
1 Introduction
The WordNet (Miller, 1990) has become the de facto standard lexical database. A family of WordNets has been developed for a number of European languages as well as a few Asian languages and these will be linked to form a multilingual lexical database. However, no attempts to develop a WordNet for the Japanese language have been reported. Although a few lexical databases including the EDR concept dictionary (Yokoi, 1995) are available for Japanese, their architecture is different from that of the WordNet. Taking into account the importance of linking lexical databases across languages, we launched a project to develop a Japanese WordNet. A number of automatic methods were proposed for the EuroWordNet project (Vossen, 1998) to help to develop WordNets for new languages (Farreres, et al., 1998). However, because of the immature technologies and the limited availability of language resources, WordNets have mainly been developed manually. Current corpus-based natural language processing technologies have made a great deal of progress. Consequently, we took the approach of automatically translating the Princeton WordNet into Japanese. This paper describes the method we propose along with a preliminary experiment, which has proved that our approach is promising for constructing a Japanese WordNet.
2 Approach
A WordNet consists of synsets, i.e., synonym sets, which define word senses, and inter-synset relations such as hypernyms, hyponyms, and others. Therefore, translating a WordNet means translating synsets. Our problem is how to determine appropriate translations for a word depending on the synsets to which it belongs. We focused on the gloss appended to each synset, which provides clues to determine the translations of the words belonging to the synset. We adopted a method of unsupervised word-sense disambiguation using bilingual comparable corpora (Kaji and Morimoto, 2002; 2005) to translate words belonging to a synset. This method, in which each sense of a first-language word is defined by a set consisting of second-language translations, can directly be applied to translation tasks. We assumed that translation candidates for a word would represent different senses.