当前位置：文档库 › A framework for linguistic relevance feedback in content-based image retrieval using fuzzy logic

A framework for linguistic relevance feedback in content-based image retrieval using fuzzy logic

A framework for linguistic relevance

feedback in content-based image

retrieval using fuzzy logic

Ronald R.Yager *,Frederick E.Petry

Machine Intelligence Institute,Iona College,715North Avenue,New Rochelle,

NY 10801,United States

Department of EE and CS,Tulane University,New Orleans,LA,United States

Received 22September 2003;received in revised form 16January 2005;accepted 6March 2005Abstract

We describe a new approach for exploiting relevance feedback in content-based image retrieval (CBIR).In our approach to relevance feedback we try to capture more of the users ?relevance judgments by allowing the use of natural language like comments on the retrieved https://www.wendangku.net/doc/2912325390.html,ing methods from fuzzy logic and computational intelligence we are able to re?ect these comments into new targets for searching the image database.Such enhanced information is utilized to develop a system that can provide more e?ec-tive and e?cient retrieval.

Keywords:Content-based image retrieval;Linguistic relevance feedback;Natural language;Fuzzy logic

Corresponding author.Address:Machine Intelligence Institute,Iona College,715North Avenue,New Rochelle,NY 10801,United States.Tel.:+12122492047;fax:+12122491689.E-mail addresses:yager@https://www.wendangku.net/doc/2912325390.html, (R.R.Yager),fep@https://www.wendangku.net/doc/2912325390.html, (F.E.

Petry).

Information Sciences 173(2005)

337–352

https://www.wendangku.net/doc/2912325390.html,/locate/ins

338R.R.Yager,F.E.Petry/Information Sciences173(2005)337–352

1.Introduction

Currently there are available databases of vast amounts of digital image data from many sources resulting in a corresponding di?culty in retrieving de-sired images.To overcome some of this problem,research has been carried out for about the last decade on content-based image retrieval(CBIR).CBIR sys-tems are based on the extraction of various features such as color,texture and shape which are used to index the image database.Retrieval can then be based on a user query to such a database to?nd the images that are most similar to the query based on various similarity metrics.However as has been known for a long period for text-based information retrieval systems,user based relevance feedback is a powerful mechanism to obtain even more satisfactory results. Here there is an interactive process with a user providing evaluation as to the relevance of the retrieval so that a modi?ed query can better approximate the user?s desired results.Recently relevance feedback has been applied in con-tent-based image retrieval.These approaches typically display to the user a small set of currently retrieved images,each of which the user then examines and evaluates as relevant or not(positive/negative feedback).The system then uses this evaluation to modify its subsequent search and returns the next set of images based on this user feedback.The process is iterated until the user is ?nally su?ciently satis?ed with the retrieval result(s).In this paper we describe an approach to relevance feedback for CBIR that captures more of the users relevance judgments by allowing natural language type comments on the re-trieved images and analyze some of the major issues that arise in such an approach.

Such enhanced information can be utilized to develop a system that pro-vides both more e?ective and more e?cient retrieval.It is well recognized that for a variety of critical current areas ranging from medical and environ-mental to military and homeland security issues,there is a need for greatly enhanced capabilities for user interactions with vast image databases.One notable such capability,particularly in the light of our great concern with homeland security and preventing terrorism,is as an aid to law enforcement authorities in?nding suspects in databases.That is such a system can help in the interaction between a computer and a person trying to help make identi-?cation.

In order to clarify the overall intention of our approach,consider an ana-logy(limited)with the process used by a police sketch artist.In this situation a person would verbally describe to the artist some characteristics and features of their mental image of,say,a bank robber.As the sketch artist proceeds,the person(i.e.,user)provides re?ning verbal comments on the sketch(image)such as for color—‘‘make the eyes bluer’’,or on texture—‘‘make the hair thicker.’’It is in this spirit of natural language like feedback that we envision the working of system we describe in the following.

R.R.Yager,F.E.Petry/Information Sciences173(2005)337–352339 2.Background

The basic approach to content-based retrieval is to use a similarity matching of a given target image or image speci?cation to a database of images.There have been a variety of approaches to the representation of the images in the image database,but the most common approach has been to represent images by one or more feature vectors.Two commonly used features are color and texture.

A major problem in CBIR is the gap between the typically low-level features extracted from images and the high-level semantics that a user has and can use in describing an image[24].A solution to this problem was to adapt the idea of relevance feedback from document retrieval[10]to content-based image retrie-val[7,9].The nature of the feedback provided by a user could be selection of only positive or relevant examples of images(basically allowing density estima-tion)[8].More generally some systems permit both positive and negative feed-back examples(hence a classi?cation problem)[14,15]and even further descriptions such as degrees of relevance or irrelevance[9]or a‘‘comparative judgment?rather than just a de?nite hit or miss[1].Clearly as additional infor-mation can be captured from a user and e?ectively utilized,the gap in retrieval semantics can be reduced.So an important aspect of our framework is the extensions of approaches such as the above to permit natural language like user feedback.Such additional higher-level interactions of the user with the CBIR system provides considerable valuable information for e?cient image retrieval.

3.Overall relevance feedback model

The purpose of the system is to retrieve,from an image database,an image corresponding to the image that the user of the system has in their mind.We shall refer to the desired image as the mental image.An important part of our system is the natural language like relevance feedback by the user.Fig.1 describes our overall approach to the image retrieval process.

We represent the images contained in the database by vectors whose compo-nents are the feature values associated with the image being represented.The current manifestation of the users mental image is captured by a target vector whose components are the same features used to represent the images but whose values are the current expression of the users mental image.In the con-tent-based image retrieval box of the preceding?gure a vector matching pro-cess between the current target vector and the image vectors is implemented.

Assume that as a result of this matching process we retrieve a collection of N images from the database.These images are then presented to the user for feed-back and comment with respect to their matching the user?s mental image.This feedback information is used to modify the vectors corresponding to the

340R.R.Yager,F.E.Petry/Information Sciences173(2005)337–352

retrieved images.These modi?ed vectors are then used to provide a new target vector(s).The new target vector is then input to the content-based image retrie-val box resulting in an updated collection of images presented to the user.The process continues until the user is satis?ed with images presented to them.

A central focus of our research is on the development of tools for the formal representation of natural language like user feedback and its subsequent use in the modi?cation of the vectors corresponding to the retrieved images.The attainment of such a capability allows the user to provide more natural feed-back information and has the potential to speed up the process of convergence in the search space by allowing the user to more pointedly direct the search process.

Consider that the user of such a system has a mental image,corresponding to the real image they want to retrieve,consisting of two red round objects on a blue background.Assume we are at the point where the system has retrieved an image,which is presented to the user.We want to allow the user to make obser-vations about the image such as:

The objects are little too far apart.

The red should be brighter.

The texture of the blue background should be more?ne.

R.R.Yager,F.E.Petry/Information Sciences173(2005)337–352341 Make the diameter of the left ball about a quarter of an inch larger.

Using observations such as these we shall modify the feature vector of the current observed image and then use this modi?ed vector as part of a new search step to help create a new target vector.

In our work we use an approach based on Zadeh?s paradigm of computing with words[20,21]which makes considerable use of fuzzy sets and the related machinery of the theory of approximate reasoning[22,18].Our choice of a fuzzy sets based approach is because of the bridge that fuzzy sets provide between linguistic expression and the formal mathematical representation needed for modi?cation of the retrieved images vectors.In particular,with the aid of fuzzy technology the types of directive sentences used above can be used to induce the appropriate changes in the vector corresponding to the image being observed.

4.Linguistically guided image vector modi?cation

In the following we describe the general procedure that forms the basis of the process of modifying the component values of the feature vector of a re-trieved image in accordance to the feedback comments made by the user. Let f be some feature associated with the user?s mental image.Assume in the current image presented to the user this feature has value V.In our approach we view the observation made by the user as expressing a relationship,usually fuzzy,between the current value of the feature f and a desired value of the fea-ture f that we denote as V0.Fig.2will be useful in understanding our procedure.

Here we?rst translate the user observation into a formal representation using what Zadeh calls generalized constraint language[21].This representa-tion essentially uses a fuzzy set framework to express the observations of the user.Once having this we manipulate the current observed value of the feature f,V,along with the user?s observations to obtain a new preferred value for the feature V0.

More speci?cally we view the observation made by the user as a fuzzy rela-tionship between the current value of the feature f and a desired value.Thus if X is the domain of the feature f we translate the users observation into a

342R.R.Yager,F.E.Petry/Information Sciences173(2005)337–352 relationship R on the space X·X.This relationship is more speci?cally a fuzzy relationship in the space X·X.In this relationship for any pair x,y2X, R(x,y)indicates the degree to which an original value x of the feature is com-patible with the modi?ed value y.Thus the translation step essentially consists of obtaining the relationship implicit in the user?s comments.

We note that while this translation process is necessarily imperfect it is no di?erent than the task faced by one human in trying implement the desired changes expressed by another human as for example a police artist would face in trying to capture a mental image of a witness to a crime.Furthermore since the overall retrieval process is essentially a feedback process,as has been dem-onstrated with the successful application of fuzzy control[6],these types of sys-tems are generally insensitive to membership grades of the fuzzy sets used.

One focus for future work is the development of a standardized vocabulary and associated fuzzy sets for the representation of requisite terms.Terms related to changes in size,color and location are of particular importance.We note here the recent work with Gardenfors[3]on conceptual spaces as being particularly relevant to this task.

5.Approximate reasoning:basic concepts

The operations needed in the modi?cation of the observed features make use of the fuzzy set based theory of approximate reasoning[18,19,22].In the fol-lowing we describe the necessarily ideas from this theory to support our work.

The primary elements of an approximate reasoning(AR)representation are a collection of variables V j called the atomic variables.It is information about the value of these variables and relationships between the values of these vari-ables that constitute the knowledge of interest.Associated with each variable V j is a set X j indicating the allowable values for the variable.This set is called its domain or universe.A joint variable is any tuple of one or more distinct atomic variables;(V2,V5)and(V1,V2,V6)are examples of joint variables.Asso-ciated with any joint variable is a set consisting of the Cartesian product of the domains of the variables making up the joint variable.

A proposition in the theory of approximate reasoning is an object of the form V is A,where V is a variable(atomic or joint)and A is a fuzzy subset of the domain of V.In approximate reasoning knowledge is represented by propositions of the preceding form.A proposition involving only one variable is called a canonical proposition while those involving two or more variables are called relational propositions.

The fuzzy subsets appearing in the propositions in an AR representation usually take their membership grade in the unit interval I=[0,1].A proposition V is A is called normal if the fuzzy subset A assumes the value one for at least one element in its domain.The propositions in AR are seen as imposing con-

R.R.Yager,F.E.Petry/Information Sciences173(2005)337–352343 straints on the possible values of the variables involved.For example,if A is a crisp subset then the meaning of the proposition V1is A is to indicate that the value for the variable V1is restricted to be a member of the set A,that is the elements in A are the only possible values for V1.The use of fuzzy subsets pro-vides for a grading associated with this idea.When A is a fuzzy subset an inter-pretation of the canonical proposition V1is A is that A(x)indicates the possibility that x is the value of V1.Some special cases of A are worth pointing out.If A is such that A(x*)=1and A(x)=0for all x5x*then this corre-sponds to the situation when V1is known to have the value x*.This is the case V1=x*.At the other extreme is the situation when A=X1,the domain of V1. This corresponds to the case when we have no knowledge about the value of V1,its value is unknown.

If V=(V1,V2)is a joint variable then the meaning of the proposition V is M is that for any pair(x1,x2),M(x1,x2)is the possibility that simultaneously V1=x1and V2=x2.We note that if M(x1,x2)=1for all pairs,then there is no joint restriction on the pair of variables.

In using approximate reasoning framework in the image retrieval process as proposed we must consider two tasks.The?rst task is translation or knowledge representation.This is the process of taking information expressed in a re-stricted type of natural language and converting into an appropriate represen-tation within the framework of AR.The second task is extracting desired information from this knowledge base.This task is called inference.It involves the manipulation of AR propositions to obtain other AR propositions.We shall here?rst consider the second of these tasks.

The manipulation of knowledge within AR is based on two basic opera-tions,conjoin and containment.The conjoin operation is a generalization of the set operations of conjunction and cartesian product.The conjoin operation provides the AR system with the capability for combining information.

De?nition.Assume V a and V b are two variables(atomic or joint)on the universes X and Y respectively.Let V a is D and V b is E be two propositions. Their conjoin denoted,

V a is D?V b is E;

is the proposition V is F where V is a variable consisting of the union of the atomic variables making up V a and V b.Furthermore F is a fuzzy subset of do-main Z of V such that for each z2Z,F(z)=Min[D(x),E(y)]where x is the por-tion of z corresponding to the domain X and y is the portion of corresponding to the domain Y.

Note.When the two variables being conjoined are the same the conjoin operation reduces to the usual intersection of fuzzy sets V a is D·V a is E=V a is D\E.

344R.R.Yager,F.E.Petry/Information Sciences173(2005)337–352

Note.When the two joint variables being conjoined have no common variables the conjoin operation becomes the Cartesian product V a is D·V b is E=(V a,V b)is D·E.

We now de?ne a special conjoin operation which operation plays a role in making propositions which are not necessarily about the same variables be about the same variable.

De?nition.Assume V a and V b are two joint variables such that V b contains all the variables that are in V a.The cylindrical extension of the proposition V a is F to V b is the proposition V b is F°is de?ned by V b is F°=V a is F·V b is X where X is the domain of the variable V b.

Note.The membership function F°can be obtained directly from the membership function of F by setting F°(y)=F(x)where x is the tuple in the base set of V a that corresponds to the portion of y in this subspace.

We now turn to the second basic operation used in approximate reasoning, containment.Containment plays a fundamental role in inference.

De?nition.Assume V a is D and V b is F are two propositions,we say that V a is D contains V b is F denoted V b is F V a is D if F°(z)6D°(z)for all z where F°and D°are the cylindrical extensions of F and D to the domain of the variable V,the union of the atomic variables in V a and V b.

We now turn to the question of reasoning in approximate reasoning.As noted in AR a proposition is viewed as placing some constraint on the allow-able values for the associated variables.Thus the atomic statement V1is A1 says that the values for the variable V1must be in the set A1,it eliminates ele-ments not in A1as possible solutions.More generally,a statement V a is A places restrictions on the tuple of variables making up the joint variable V a.As-sume we have a knowledge base consisting of a set of propositions,P i,for i=1 to n.Since each proposition P i induces a restriction the totality of our knowl-edge can be seen to be the conjunction(conjoin)of these individual restrictions. Thus the set of propositions{P1,P2,...,P n},our knowledge base,induces a combined restriction KB=P1·P2·ááá·P n.

We note that if the value of V is constrained to lie in the set A then it must also lie in any set containing A.A fundamental inference mechanism in AR is the entailment principle which formalizes this observation.

5.1.Entailment principle

From the knowledge V a is A we can infer V b is B if

V a is A V b is B:

One special case of the entailment principle is the operation of projection. De?nition.Let V1and V2be two atomic variables with domains X and Y.Let V a=(V1,V2)be a joint variable.The projection of the proposition V a is A onto

V1,denoted Proj V

1?V a is A is the proposition V1is B where for x2X,

B(x)=Max y2Y[A(x,y)].

It can be shown that V a is A V1is B.Thus,projection is a special case of entailment.

The use of the entailment principle,along with the process of conjoining of individual pieces of knowledge,premises,leads to a very powerful reasoning mechanism.

We are now in a position to explain the process of obtaining the new desired feature value in our image retrieval model.From the current observed image we obtain the value V of the feature f.Let us denote this as V is A where A is a fuzzy subset over the domain X of the feature.From the comments of the observer we obtain the relationship between the current value of the feature V and the desired feature value V0.We express this relationship in AR using a joint variable as

eV;V0Tis R

where R is a fuzzy subset over X·X.We now take the conjoin of the two pieces of information,which we have(V,V0)is R and V is A.This gives us (V,V0)is E,where E is the fuzzy subset such that E(x,y)=Min[R(x,y),A(x)]. Using the projection operation we obtain V0is F where F is the fuzzy subset such that F(y)=Max x2X[E(x,y)].

In the special case when value of the feature is precisely known,V=x*,then the fuzzy subset A is such that A(x*)=1and A(x)=0for x5x*.In this spe-cial case E(x,y)=0for x5x*and E(x,y)=R(x*,y)for all others.In this spe-cial case we see that F is the fuzzy subset such that F(y)=R(x*,y).As a simple illustration consider the situation in which our feature is the diameter of a circle and we have a users comment‘‘make the diameter much bigger.’’In this case R is the relationship‘‘much bigger.’’In this case for any pair x and y the value R(x,y)indicates the degree to which x can be considered as much bigger than y.Here then if3in.is the diameter of the circle in the image being commented on then the fuzzy subset F,where F(y)=R(3,y),is value of the diameter in the new target image.

6.Fuzzy representation of user feedback

As noted a major step in this approach is the representation of the user?s ver-bal observations about proposed modi?cations in terms of fuzzy relationships, R.R.Yager,F.E.Petry/Information Sciences173(2005)337–352345

346R.R.Yager,F.E.Petry/Information Sciences173(2005)337–352

the translation step from Fig.2.Our approach here is based on Zadeh?s work in developing the paradigm of computing with words[20].Some modi?cations are very easy to express.For example the directive‘‘make the left ball about a quarter of an inch bigger.’’Here we can represent the concept of‘‘about a quarter of an inch’’as a fuzzy subset and then use the capability of fuzzy arith-metic to generate the new value.Other representations require using relation-ships such as bigger or closer or darker.

In developing such a system we must provide procedures for de?ning the de-fault fuzzy subsets used to represent the modi?cations and directives suggested by the user.Here we must keep in mind that the overall process is an iterative one where we are feeding back modi?ed vectors to the search process.In some ways this is very similar to the type of control system in which fuzzy logic con-trol has been very successfully applied[19].In these systems it was realized that the performance of the feedback mechanism was not very sensitive to exact de?nition of the fuzzy subsets.The lesson learned was to use representations that move us in the right direction,of course the better the representation the quicker we move.In addition we learned the usefulness of parameterized membership functions.While we can assume default representations of many of the allowable linguistic terms,learning a user?s more individual de?nition is possible by combing fuzzy representations with neural networks and genetic algorithms[6].

Implicit in our approach has been an assumption that all the features about which the user expresses their observations are the same features as those used to represent the images in the feature vector.In some situations this may not be the case.Some of the comments may involve concepts not appearing in the feature vector.Rather than simply disregarding these com-ments,making use of these requires us to consider how to transform directives expressed in terms of‘‘human’’level observable variables into directives for modi?cation in terms of the types of micro variables of which the feature vec-tor may be composed.In some sense this problem is similar to the problem that arises in neural networks where we must allocate errors on observable variables into changes in internal weights.We recall this problem was solved by the back propagation method created by Werbos[16].With this under-standing we are developing a technique,based on the back propagation me-thod.Thus if f is an observable feature and f háááf q are a collection of related micro features,which are part of the feature vector we‘‘back propagate’’any comment on this observed variable to these hidden features.In order to accomplish this we must determine which micro features a?ect the obser-vable feature.From this we need to obtain some relationship and use it to learn how modify the feature vectors to get a desired change in observable variable.Here we shall obtain a constraint propagation problem.The knowl-edge required to determine the relationship between observable variables and elements in the feature vector is domain dependent.

7.Matching fuzzy targets

So far we have focused on the representation of the modi?cations of the fea-ture vector based on the verbal observations of the user.Our approach also raises some important issues a?ecting the search mechanism.As noted,a result of the modi?cation to a presented feature vector is that we obtain a new feature vector in which some of values may be fuzzy subsets.The use in the image data-base search process of feature vectors whose component values are fuzzy raises some important issues.Some of these are of a pragmatic nature a?ecting our implementation of the system while others are related to our understanding of the system ?s performance.

Consider the situation where in the search step we have a target vector with a component feature that may have a fuzzy value.Assume feature f has a value that is a fuzzy subset A .This must now be matched with a vector in the database corresponding to an https://www.wendangku.net/doc/2912325390.html,ing the similarity relationship that exists on the domain of this feature we obtain a measure of similarity between an object in the database and the target.We note that while our target vector can have fuzzy values the vectors representing the images in the database have ordinary values for their features.In particular if X is the domain of f and A is a fuzzy subset with membership function A (x )then if an image has a value of f equal x *we use the extension principle introduced by Zadeh [23]to obtain a measure of similarity between the image and the target for the feature f .Assume Sim(x ,y )2[0,1]indicates the degree of similarity between the values x and y for this https://www.wendangku.net/doc/2912325390.html,ing this the degree of similarity between a target with fuzzy value A and image with feature value x *will be a fuzzy subset on the unit inter-val expressed as S x 2X A ex TSim ex ;x ?Tn o .That is the measure of similarity for this feature will be a fuzzy subset.We now must combine these fuzzy similarities scores cor-responding to the individual feature to obtain an aggregate similarity relevance score for an image and the current target vector.In order to accomplish this we ?rst use fuzzy arithmetic [19]to obtain the average of the similarities of each of the features.This gives us a unique,although fuzzy value,for the similarity of the image to the fuzzy target.As a ?nal step we make a defuzzi?cation [19]of fuzzy image similarity values.This gives us for each image a scalar value cor-responding to its similarity to the fuzzy target.We now can order these values to select the most appropriate images as is done in the non-fuzzy case.

8.Example of the modi?cation of image color features

Here we present an example of how in a content-based image retrieval sys-tem a user can provide linguistic feedback on retrieved images and this feed-back can be translated into modi?cations of the feature vectors.In particular R.R.Yager,F.E.Petry /Information Sciences 173(2005)337–352347

we focus on one of the most commonly used features—color.It is assumed the system is provided with a default vocabulary of terms appropriate for human usage and the semantics for mapping these terms into the more detailed feature vectors.The building of this vocabulary is an important task in the construc-tion of any real system.

Consider a user who wants to ?nd a forest scene from the image database.They may retrieve a number of scenes not containing su?cient green and so the user desires new images with more green.First let us review how the color fea-ture is represented in most CBIR systems.For color representation the color histogram has been commonly used in many image retrieval systems.Systems that employ image retrieval using color histograms have three basic compo-nents [4]:

1.Color categorization .The color spectrum,SC,used for describing an image,subdivided into n color categories based on the exact system color dis-crimination capabilities and the desired granularity.So we have

SC ?f c i j i ?1ááán g

where c i is the i th color category.

2.Histogram structure .Given an image I ,the histogram of I for the color spectrum SC is

H SC ?I ?f px eI ;c i Tj i ?1ááán g

where px(I ,c i )is the number of pixels that correspond to the category c i of the color space SC.

Note the total number of pixels in the segmentation of the images is PX ?P n i ?1px eI ;c i T.

3.Histogram matching .There is some similarity metric function S and the matching of two images,I and J ,based on color histograms is S (H SC [I ],H SC [J ]).As we have discussed there can be a variety of such matching func-tions.This includes well-established metrics such as histogram intersection distance [13],vector distance [5]or several approaches using a Minkowski dis-tance [11,12].

Now assume that a user has evaluated some retrieved images but envisions a mental image that is considerably greener,commenting to make the image ‘‘much greener.’’Note it is possible to grid an image so that the user may spec-ify changes to the top left or bottom of the image for example.

There are two parts to the built-in default vocabulary provided to the user:color terms :blue ;green ;yellow ;...;

modifier terms

:much ;slightly ;less ;...

The instruction given by the user to make the image ‘‘much greener’’contains one term from each category.We now describe how we can process this instruction to get new images.348R.R.Yager,F.E.Petry /Information Sciences 173(2005)337–352

R.R.Yager,F.E.Petry/Information Sciences173(2005)337–352349 Let us discuss the semantics for the color terms?rst.We have a default high-level color spectrum divided into m terms denoted t j.Each of these terms corresponds to some mapping into the color categories found in the feature histogram.Of course the underlying set of color categories,SC,actually used in the color feature representation in the image database is typically a much ?ner granulation than the higher-level set of default color terms provide for the user interface.This implies that m(n.It is reasonable based on human factors study of color perception to use around10or so high-level color terms for user image evaluation and feedback[17].In some systems such as QBIC[2] there may be over200color categories used in the image?s color histogram. Thus it is required to provide the relationship between the coarser-grained high-level and low-level color representations.To achieve this,for each high-level color term t j,a fuzzy subset de?ning a correspondence to the lower level categories c i must be developed:

t j?f c1=l1j;c2=l2j;c3=l3j;...;c n=l nj g

where l ij is the membership of the category c i in the fuzzy set corresponding to t j.For example if the user term is blue,then for various low-level categories of red,membership is zero;some categories of purple may have a non-zero mem-bership;and most of the blue categories would have a membership of1.This can be represented,for example,as a trapezoidal membership function for each of the default color terms as seen in Fig.3.

Next we need to provide representations for the modi?er terms.Here for illustrative purposes we take a simpli?ed approach by specifying for each modi-?er in the vocabulary a?xed ratio of increase or decrease,r.We now use this to implement the user?s instructions by generating a new histogram by modify-ing the histogram of the original image which the user viewed.This new histo-gram will serve as the target for the next round of retrieving images from the image database.As a?rst step we obtain px0(T,c i),the raw pixel count for cat-egory c i for the new target.In particular

px0eT;c iT?pxeI;c iT?e1tl ij?rT

350R.R.Yager,F.E.Petry/Information Sciences173(2005)337–352

where,since the color term speci?ed is t j,we use the membership of c i in t j,l ij, to weight the modi?er ratio r(note r will be negative for modi?ers indicating a decrease).

The new raw pixel counts px0(T,c i),may not sum to the true number of pix-els,PX,in the image segmentation.Thus after these initial modi?cations of the histogram a normalization step is required.Let the new total be

X n

px0eT;c iT

PX0?

i?1

So the normalization is N=PX/PX0if PX0>PX,and vice versa.Finally we obtain the new color histogram for the target of the next round as

H SC?T ?fepxeT;c iT?e1tl ij?rTT?N j i?1ááán g

A similar approach can be taken for other features such as texture and shape to develop the system as described,although the mapping of high-level descrip-tions to actual image feature representations is more problematic.

9.Working at the pixel level

We recall that for each image in the database we have both a feature vector representation as well as the pixel(JPEG or PGM)representation.The pixel representation is used as the actual image presented to user.In addition it is also the pixel representation that is used to the construct the feature vector rep-resentation.The main bene?t of using the feature vector representation is in the process of matching and search in the database.In reality it is the image(pixel) representation and not the feature vector that the user will actually examine and comment on as to the image?s relevance.

One avenue worth investigating is working at the pixel level.Here we solicit the user comments in terms understandable at the pixel level.We describe our ideas in this direction.In this approach,again using AR,we interpret and translate the user?s comments on observed images into directives to modify the pixel representations of these images.These directives will essentially cor-respond to instructions for changing,for example,the color level of various pixel locations in the image currently being observed.Thus if M is the pixel matrix corresponding to current observed image,after implementing the changes we obtain a new pixel matrix M0.We can apply our feature extraction algorithms on this modi?ed image M0yielding the new target feature vector F0. We now can use F0for searching our image database for the desired image.

We note that since the number of images presented to the user is typically not large,the additional computational burden of having to translate from the modi?ed pixel image matrix back into a feature vector representation is minimal.

R.R.Yager,F.E.Petry/Information Sciences173(2005)337–352351 There are some important advantages to this approach.Since the pixel rep-resentation is closer to human perception,the use of the pixel domain to inter-pret the users verbal directives for modifying the current image may prove to be more convenient than direct modi?cation of the feature vector.As an illus-tration consider the feedback to make the red ball a quarter of an inch larger. Essentially in order to accomplish this in the pixel domain we need to make a change in the pixel values corresponding to those in the neighborhood of the ball from their current values of the ball.

So by using some combination of the pixel domain,the feature vector do-main and any other higher-level image interpretations available we have some representations that can be used for translating verbal observer feedback into operational instructions to enable the modi?cation of the current image into a new target feature vector.It also must be strongly emphasized that the process is an iterative one,thus our goal is to provide target vectors that help move the search process in the correct direction.

10.Conclusion

In the preceding we have outlined a new approach to the retrieval of images from a database.Central to this approach is the use of linguistically expressed relevance feedback by the customer of system.In future work we look forward to the implementation of an image retrieval by ourselves and other researchers to test some of the ideas presented here.

References

[1]I.Cox,https://www.wendangku.net/doc/2912325390.html,ler,T.Minka,T.Papathomas,P.Yianilos,The Bayesian image retrieval

system,PicHunter:theory,implementation and psychophysical experiments,IEEE Transac-tions on Image Processing9(1)(2000)20–37.

[2]M.Flickner,H.Sawhney,W.Niblack,J.Ashley,Q.Huang,B.Dom,M.Goraki,J.Hafner,

D.Lee,D.Petkovic,D.Steele,P.Yanker,Query by image and video content:the QBIC

System,IEEE Computer(1995)23–32.

[3]P.Gardenfors,Conceptual Spaces:the Geometry of Thought,MIT Press,Cambridge,MA,

2000.

[4]Y.Gong,Intelligent Image Databases:towards Advanced Image Retrieval,Kluwer Publishers,

Boston,MA,1998.

[5]J.Hafner,H.Sawhney,W.Equitz,M.Flickner,W.Niblack,E?cient color histogram

indexing for quadratic form distance functions,IEEE Transactions on Pattern Analysis and Machine Intelligence17(7)(1995)729–736.

[6]V.Kecman,Learning and Soft Computing,MIT Press,Cambridge,MA,2001.

[7]T.Kurita,T.Kato,Learning of personal visual impression for image database systems,in:

Proceedings of IEEE International Conference on Document Analysis and Recognition,1993, pp.451–459.

352R.R.Yager,F.E.Petry/Information Sciences173(2005)337–352

[8]C.Meilhac,C.Nastar,Relevance feedback and category search in image databases,in:

Proceedings.IEEE Conference on Multimedia Computation and Systems,1999,pp.245–253.

[9]Y.Rui,T.Huang,M.Ortega,S.Mehrota,Relevance feedback:a power tool in interactive

content-based image retrieval,IEEE Transactions on Circuits and Systems for Video Technology8(5)(1998)644–655.

[10]G.Salton,Automatic Text Processing,Addison-Wesley,Reading,MA,1989.

[11]A.Smeulders,M.Worring,S.Santini,A.Gupta,R.Jain,Content-based image retrieval at

the end of the early years,IEEE Transactions on Pattern Analysis and Machine Intelligence22

(12)(2000)1349–1379.

[12]M.Stricker,M.Orengo,Similarity of color images,in:Storage and Retrieval of Image and

Video Databases III,vol.2,1995,pp.381–392.

[13]M.Swain,B.Ballard,Color indexing,International Journal of Computer Vision7(1)(1991)

11–32.

[14]K.Tieu,P.Viola,Boosting image retrieval,in:Proceedings.IEEE Conference on Computer

Vision and Pattern Recognition,2000,pp.764–769.

[15]S.Tong,E.Chang,Support vector machine active learning for image retrieval,in:ACM

Multimedia Conference,2001,pp.566–572.

[16]P.Werbos,The Roots of Backpropagation,John Wiley&Sons,New York,1994.

[17]G.Wyszecki,W.Stiles,Color Science:Concepts and Methods,second ed.,Wiley,New York,

1982.

[18]R.R.Yager,Approximate reasoning as a basis for computing with words,in:L.A.Zadeh,J.

Kacprzyk(Eds.),Computing with Words in Information/intelligent Systems1,Springer-Verlag,Heidelberg,1999,pp.50–77.

[19]R.R.Yager,D.P.Filev,Essentials of Fuzzy Modeling and Control,John Wiley,New York,

1994.

[20]L.A.Zadeh,Fuzzy logic=computing with words,IEEE Transactions on Fuzzy Systems4

(1996)103–111.

[21]L.A.Zadeh,Toward a perception-based theory of probabilistic reasoning with imprecise

probabilities,Journal of Statistical Planning and Inference105(2002)233–264.

[22]L.A.Zadeh,The concept of a linguistic variable and its application to approximate reasoning:

Part1,Information Sciences8(1975)199–249.

[23]L.A.Zadeh,Fuzzy sets,Information and Control8(1965)338–353.

[24]X.Zhou,T.Huang,Relevance feedback in content-based image retrieval:some recent

advances,Information Sciences148(2)(2002)129–137.

新编英语语法教程(第6版)练习参考答案

新编英语语法教程（第6版）第21讲练习参考答案Ex. 21A was sorry to learn… will be sad to hear… would be very surprised to receive… is happy to have found… was afraid to go… was pleased to hear… am very anxious to meet you. were delighted to receive your telegram. were sensible to stay indoors. clerk was prompt to answer the call. rule is easy to remember. are reluctant to leave this neighbourhood. house is difficult to heat. you ready to leave would be foolish to go out in this weather. is quick to see the point. is very keen to get on. are proud to have him as a friend. was rude not to answer your letter. are happy to have you with us this evening. Ex. 21B decision to resign surprised all of us. showed no inclination to leave.

English Linguistic

Chapter 1 Introduction 1. Linguistic and English Linguistic Linguistic 的目的 aims at developing a theory general linguistic 一般语言学 descriptive linguistic 描述性语言学 general linguistic (一般语言学 ) 为 descriptive linguistic (描述性语言学) 提供了 framework(框架), 这也就是为什么 general linguistic 能够被分析和被描述. General linguistic and descriptive linguistic are complementary to each other (相互补充). English linguistic is a kind of descriptive linguistics. 2.The nature of languages (语言的本质) 1. language is a system 2. language is symbolic 3. language is a system of vocal symbols The system of language is called langue 1. language is a system the speaker ’s speech is called parole Competence is the speaker-hearer’s knowledge of his language Performance is the actual use of language in concrete situation 2. language is symbolic ( 语言是象征意义的) 3. language is a system of vocal symbols ( 语言是一种声音符号) 语言学侧重研究的是 speech 不是 written form Reason 1. Biologically (生物上来讲) 儿童学习说比学习读写早得多. 2. Functionally (功能上来讲) 日常生活中口语使用比书面语频繁的多. 3. Historically ( 历史而言) 口头语使用在书面语之前,当今世界有许多语言并没有留下文字记录 Language is arbitrary Language is creative Language is double-structured Language is changeable de Saussure 的理论 Chomsky 的理论

电大公共行政学小抄名词解释

1.公共行政学:公共行政学是研究公共组织依法处理政务的有效性、公平性、民主性的规律的交叉性与综合性学科。（在这里公共组织主要是指政府，公共行政就是政府行政。） 2.公共行政环境:公共行政环境是指直接或间接地作用或影响公共组织、行政心理、行政行为和管理方法与技术的行政系统内部和外部的各种要素的总和。 3.组织文化:组织文化是指组织在一定的环境中，逐步形成的全体公共组织成员所共同信奉和遵守的价值观，并支配他们的思维方式和行为准则。（组织文化在政府也可以称之为公共行政组织文化，在企业则称之为企业文化。组织文化包括组织观念、法律意识、道德感情和价值观等。） 4.政府职能:政府职能是指政府在国家和社会中所扮演的角色以及所应起的作用。（换句话说，就是指政府在国家和社会中行使行政权力的范围、程度和方式。） 5.市场失效:市场失效是指因为市场局限性和缺陷所导致资源配置的低效率或无效率,并且不能解决外部经济与外部不经济的问题以及社会公平问题。 6.行政体制:行政体制指政府系统内部行政权力的划分、政府机构的设置以及运行等各种关系和制度的总和。 7.地方政府体制:地方政府体制是指地方政府按照一定的法律或标准划分的政府组织形式. 8.行政区划体制:行政区划体制是指根据一定的原则将全国领土划分为若干部分和若干层次的管理区域，并设置相应的行政机关的组织体制。 9.完整制:完整制又叫一元统属制，是指公共组织的同一层级或同一组织内部的各个部门，完全接受一个公共组织或同一位行政首长的领导、指挥和监督的组织类型。 10.分离制:分离制又称多元领导制，是指一个公共组织的同一层级的各个组织部门或同一组织部门，隶属于两个或两个以上公共组织或行政首长领导、指挥和监督的组织类型。 11.首长制:首长制又称独立制、一长制或首长负责制。它是指行政首长独自掌握决策权和指挥权，对其管辖的公共事务进行统一领导、统一指挥并完全负责的公共组织类型。 12.层级制:层级制又分级制，是指公共组织在纵向上按照等级划分为不同的上下节制的层级组织结构，不同等级的职能目标和工作性质相同，但管理范围和管理权限却随着等级降低而逐渐变小的组织类型。 13.机能制:机能制又称职能制，是指公共组织在横向上按照不同职能目标划分为不同职能部门的组织类型。14.行政领导者:行政领导者是指在行政系统中有正式权威和正式职位的集体或个人。 15.委任制:亦称任命制，是指由立法机关或其他任免机关经过考察而直接任命产生行政领导者的制度。 16.考任制:考任制是指由专门的机构根据统一的、客观的标准，按照公开考试、择优录取的程序产生行政领导者的制度。 17.行政领导权力:行政领导权力是指行政领导者在行政管理活动中,利用其合法地位以不同的激励方式和制约方式,引导下属同心协力达成行政目标的影响力。18..行政领导责任:行政领导责任是指行政领导者违反其法定的义务所引起的必须承担的法律后果。 19.人事行政:人事行政是指国家的人事机构为实现行政目标和社会目标，通过各种人事管理手段对公共行政人员所进行的制度化和法治化管理。20.人力资源:人力资源是指在一定范围内能够作为生产性要素投入社会经济活动的全部劳动人口的总和。它可分为现实的人力资源和潜在的人力资源两部分。 21.程序性决策:也叫常规性决策，是指决策者对所要决策的问题有法可依，有章可循，有先例可参考的结构性较强，重复性的日常事务所进行的决策。 22.非程序性决策:也叫非常规性决策，是指决策者对所要决策的问题无法可依，无章可循，无先例可供参考的决策，是非重复性的、非结构性的决策。 23.危机决策:是指领导者在自然或人为的突发性事件发生后，迅速启动各种突发事件应急机制，大胆预测，做出决定的过程。 24.行政决策参与:是指行政领导者个人或集体在行政决策时，专家学者、社会团体、公民等对决策提出意见或建议的活动。 25.行政执行:行政执行是行政机关及行政人员依法实施行政决策，以实现预期行政目标和社会目标的活动的总和。 26.行政控制:行政控制指行政领导者运用一定的控制手段，按照目标规范衡量行政决策的执行情况，及时纠正和调节执行中的偏差，以确保实现行政目标的活动。27.行政协调:行政协调是指调整行政系统内各机构之间、人员之间、行政运行各环节之间的关系，以及行政系统与行政环境之间的关系，以提高行政效能，实现行政目标的行为。 28.法制监督:法制监督，又称对行政的监督，是指有权国家机关对行政机关及其工作人员是否合法正确地行使职权所进行的监督与控制。 29.舆论监督:舆论监督是指通过在公共论坛的言论空间中所抒发的舆论力量对政府机构和政府官员滥用权力等不当行为的监督与制约。 30.行政立法:行政立法一般是指立法机关通过法定形式将某些立法权授予行政机关，行政机关得依据授权法(含宪法)创制行政法规和规章的行为。 31.行政法规:行政法规是指国务院根据宪法和法律，按照法定程序制定的有关行使行政权力，履行行政职责的规范性文件的总称。 32.标杆管理: 标杆管理是指公共组织通过瞄准竞争的高目标,不断超越自己,超越标杆,追求卓越,成为强中之强组织创新和流程再造的过程. 33.政府全面质量管理:政府全面质量管理是一种全员参与的、以各种科学方法改进公共组织的管理与服务的，对公共组织提供的公共物品和公共服务进行全面管理，以获得顾客满意为目标的管理方法、管理理念和制度。 34.行政效率:行政效率是指公共组织和行政工作人员从事公共行政管理工作所投入的各种资源与所取得的成果和效益之间的比例关系。 35.行政改革:行政改革是指政府为了适应社会环境，或者高效公平地处理社会公共事务，调整内部体制和组织结构，重新进行权力配置，并调整政府与社会之间关系的过程。 36.政府再造:政府再造是指对公共体制和公共组织绩效根本性的转型，大幅度提高组织效能、效率、适应性以及创新的能力，并通过改革组织目标、组织激励、责任机制、权力结构以及组织文化等来完成这种转型过程。

新编英语语法教程

导论———语法层次 0.1 词素 1）自由词素 2）粘附词素 0.2 词 1）简单词、派生词、符合词 2）封闭词类和开放词类 0.3 词组 1）名词词组 2）动词词组 3）形容词词组 4）副词词组 5）介词词组 0.4分句 1）独立分句和从属分句 2）简单分句和复杂分句 3）主句和从句 4）限定分句、非限定性分句、无动词分句0.5 句子 1）完全句和不完全句 2）简单句、并列句、复杂句、并列复杂句第1讲句子结构 1.1 主谓结构和句子分析 1）主语和谓语 2）句子分析 1.2 基本句型及其转换与扩大 1）基本句型 2）基本句型的转换与扩大第2讲主谓一致（一） 2.1指导原则 1）语法一致 2）意义一致和就近原则 2.2 以-s 结尾的名词作主语的主谓一致问题1）以-s结尾的疾病名称和游戏名称 2）以-s结尾的学科名称 3）以-s结尾的地理名称 4）其他以-s结尾的名词 2.3 以集体名词作主语的主谓一致问题 1) 通常作复数的集体名词 2）通常作不可数名词的集体名词 3）既可作单数也可作复数的集体名词 4）a committee of 等+复数名词

第3讲主谓一致（二） 3.1 以并列结构作主语的主谓一致问题 1）由and/both... And 连接的并列主语 2）由or/nor/either...or 等连接的并列主语 3）主语+as much as 等 4）主语+as well as 等 3.2 以表示数量概念的名词词组作主语的主谓一直问题1）以表示确定数量的名词词组作主语 2) 以表示非确定数量的名词词组作主语 3.3 其他方面的主谓一致问题 1）以名词性分句作主语的主谓一致问题 2）以非限定分句作主语的主谓一致问题 3）关系分句中的主谓一致问题 4）分裂句中的主谓一致问题 5）存在句中的主谓一致问题第4讲 4.1 名词分类和名词词组的句法功能 1）名词分类 2）名词词组的句法功能 4.2 名词的数 1）规则复数和不规则复数 2）集体名词、物质名词、抽象名词、专有名词的数4.3 单位词 1）一般表示个数的单位词 2）表示形状的单位词 3）表示容积的单位词 4）表示动作状态的单位词 5）表示成双、成对、成群的单位词第5讲 5.1 名词属格的构成、意义和用法 1）名词属格的构成 2）名词属格的意义 3）名词属格的用法 5.2 独立属格和双重属格 1）独立属格 2）双重属格第6讲限定词（一） 6.1限定词与三类名词的搭配关系 1）能与三类名词搭配的限定词 2）只能与单数名词搭配的限定词 3）只能与复数名词搭配的限定词

电大专科《公共行政学》名词解释简答题题库及答案(试卷号：2202)

电大专科《公共行政学》名词解释简答题题库及答案（试卷号：2202）盗传必究一、名词解释 1．政府职能：是指政府在国家和社会中所扮演的角色以及所应起的作用。 2．行政区划体制：是指根据一定的原则将全国领土划分为若干部分和若干层次的管理区域，并设置相应的行政机关的组织体制。 3．完整制：又叫一元统属制，是指公共组织的同一层级或同一组织内部的各个部门，完全接受一个公共组织或同一位行政首长的领导、指挥和监督的组织类型。 4．行政效率：是指公共组织和行政工作人员从事公共行政管理工作所投入的各种资源与所取得的成果和效益之间的比例关系。 5．市场失效：是指因为市场局限性和缺陷所导致资源配置的低效率或无效率，并且不能解决外部性问题以及社会公平问题。 6．行政体制：指政府系统内部行政权力的划分、政府机构的设置以及运行等各种关系和制度的总和。 7．程序性决策：也叫常规性决策，是指决策者对所要决策的问题有法可依，有章可循，有先例可参考的结构性较强，重复性的日常事务所进行的决策。 8．行政法规：是指国务院根据宪法和法律，按照法定程序制定的有关行使行政权力，履行行政职责的规范性文件的总称。 9. 管理幅度：是指领导机关或领导者直接领导下属的部门或人员的数额。 10．行政决策参与：是指行政领导者个人或集体在行政决策时，专家学者、社会团体、公民等对决策提出意见或建议的活动。 11．电子政府：是指在政府内部采用电子化和自动化技术的基础上，利用现代信息技术和网络技术，建立起网络化的政府信息系统，并利用这个系统为政府机构、社会组织和公民提供方便、高效的政府服务和政务信息。 12. 公共行政学：是研究公共组织依法处理政务的有效性、公平性、民主性的规律的交叉性与综合性学科。 13. 行政领导责任：是指行政领导者违反其法定的义务所引起的必须承担的法律后果。 14. 风险型决策：是指决策者对决策对象的自然状态和客观条件比较清楚，也有比较明确的决策目标，但是实现决策目标结果必须冒一定风险。 15. 事前监督：是指在某种公共行政管理活动开展之前，监督部门围绕公共行政管理主体的行政行为进行的监督检查。 16. 政府再造：是指对公共体制和公共组织绩效根本性的转型，大幅度提高组织效能、效率、适应性以及创新的能力，并通过改革组织目标、组织激励、责任机制、权力结构以及组织文化等来完成这种转型

新编英语语法教程期末考试试卷

大学英语语法模拟试题 1. Mr. and Mrs. Burns feel more comfortable on a ship than they would be if they ______________any other way. A. would travel B. travelled C. are travelling D. have travelled 2. We______________ that Jim be there. A. hope B. wish C. expect D .ask 3. Lucy is glad she didn’t stay on the farm. She______________ bored. A. may be B. will be C. could be D. might have been 4. The dean of studies would have come to see you had it been possible, but he ______________so busy then. A. had been B. was C. were D. would be 5. They would certainly have come and helped us ______________time. A. did they have B. had they had C. had they have D would they have 6. If you were in better health, we______________ you to join in the work last week. A. would have allowed B. would allow C. should allow D. had allowed 7. She cried for her______________ lover. A. departed B. being departed C. departing D. having departed 8. ______________ in an important examination, one of the students in his class lost interest in his work A. Failing B. Failed C. Having been failed D. Having Failed 9. If it______________ tomorrow, I’ll stay at home. A. rained B. will rain C. had rained D. rains 10. “It’s getting very late.”“Yes, it’s time______________.” A. that we left B. we leave C. we’ll leave D. we have left a bus to go there, but he preferred to walk.

What is linguistic

What is linguistics? I think that what should we realize is the fundamental things of linguistics about is that language is more than just a big bound of words. It’s easy to think that what are we doing, what are we speaking, is producing words that we learn when we were kids and now we know them, and now we string them together and found some sort of ways. Maybe there were some wrong ways to string them together, you learn at school. But basically what we doing is using big bound of words. Now if you really think about it. Words alone can’t be even most of what using a language in terms of what we are interested about using languages. And that’s because you think about it. Imagine if you in heeled a Russian dictionary, that’s say that you for all the attendance purpose knew every single word in the language. If you think about it really, even then you would be unable to have any kind of meaning for conversation of language. And 3 and 4 years olds would like be the years hard of you and that’s because obviously, there is also an issue how you put the words together in order to convey meaning. There’s also a said say, if a language is more than just a big bucket of words, then as Saussure’s grammar’s concern, it’s the matter of setting whether the words fit into one of a part of speech. So the idea is that when you got these parts of speech, then

《新编英语语法教程》语法术语精编

《新编英语语法教程》主要章节语法术语Introduction: Grammatical Hierarchy (导论—语法层次) 2. Parts of speech (word class) 3. Phrases 词组 4. Clause 分句 5. Sentence 句子 1. Morpheme 词素 Free morpheme 自由词素 Bound morpheme 粘附词素 Allomorph 词素变体 Noun phrase Verb phrase Adjective phrase Adverb phrase Preposition phrase Conjunction

Lecture 1 Sentence Structure（L1）Sentence elements: S (subject) 主语V (predicate verb) 谓语动词 O (object) 宾语 C (complement) 补足语 A (Adverbial) 状语 1. Two ways of sentence analysis 1) SVO Sentence Clause NP VP NP Subject Predicate verb Object All the man have done their best. Sentence = Subject + Predicate (Predicate Verb + Object, Complement, Adverbial, etc.) ●句子由主语和谓语构成，进一步把谓语剖析为谓语动词、宾语、补语、状语等。 2) Subject + Predicate (= operator + predication) Sentence Clause Subject Predicate Operator Predication All the man have done their best. ●句子由主语和谓语构成，进一步把谓语剖析为操作词（operator）和述谓成分（predication）。 2. Basic clause types SVC, SV, SV A, SVO, SVOA, SVOC, SV oO Lecture 2 Subject-Verb Concord （L2-3） Guiding principles: Grammatical Concord Notional Concord Principle of Proximity 语法一致原则意义一致原则就近原则 Nominal clause Non-finite clause Relative clause Cleft sentence Existential clause 名词性分句非限定分句关系分句分裂句存在句Lecture 3 Noun and Noun Phrase（L4-5） 1. Classification of nouns

公共行政学名词解释

1.委员会制委员会制是指在公共组织中，由两个人以上掌握决策权和指挥权，按照多数原则进行决策的公共组织类型。 2.层级制层级制又分级制，是指公共组织在纵向上按照等级划分为不同的上下节制的层级组织结构，不同等级的职能目标和工作性质相同，但管理范围和管理权限却随着等级降低而逐渐变小的组织类型。 3.机能制机能制又称职能制，是指公共组织在横向上按照不同职能目标划分为不同职能部门的组织类型。 4.战略管理公共组织的战略管理是指对公共组织在一定时期的全局的、长远的发展方向、目标、任务和政策，以及资源调配做出的决策和管理艺术。 5.政府再造的含义政府再造是指对公共体制和公共组织绩效根本性的转型，大幅度提高组织效能、效率、适应性以及创新的能力，并通过改革组织目标、组织激励、责任机制、权力结构以及组织文化等来完成这种转型过程。政府再造就是用企业化体制取代官僚体制，即创造具有创新习惯和持续改进质量能力的公共组织和公共体制，而不必靠外力驱使。企业家政府是政府再造的重要内容。企业家政府是指具有企业家精神的行政管理者，用企业的管理方式，以低成本高产出为目标，敢于冒风险、敢于创新、敢于打破僵化官僚体制，取得高绩效的政府。

企业家政府重视政府的成本效益，重视创新与改革，强调利用市场机制和竞争，强调对执行者授权，主张顾客导向，主张放松规制。 6.目标管理的涵义和特点目标管理是以目标为导向，以人为中心，以成果为标准，而使组织和个人取得最佳业绩的现代管理方法。目标管理的特点是以人、工作和成果为中心的现代管理方法。 7.行政规章的含义行政规章是指特定的行政机关根据法律和法规，按照法定程序制定的具有普遍约束力的规范性文件的总称。行政规章简称规章。 8.事前监督；事前监督是指在某种公共行政管理活动开展之前，监督部门围绕公共行政管理主体的行政行为进行的监督检查。 9.行政评估的含义行政评估是指对行政执行活动的进展情况和效果进行评价和总结，包括行政执行过程评估和行政执行效果评估两个方面。一般意义上所说的行政评估主要是指行政执行效果评估。 10.行政领导权力的概念行政领导权力是指行政领导者在行政管理活动中，利用其合法地位以不同的激励方式和制约方式，引导下属同心协力达成行政目标的影响力。 11.完整制完整制又叫一元统属制，是指公共组织的同一层级或同一组织内部的各个部门，完全接受一个公共组织或同一位行政首长的领导、指挥和监督的组织类型。

Cognitive Linguistic

Linguistic Coursework ---Cognitive Linguistics

Linguistic course/work -----Cognitive Linguistics Introduction Cognitive Linguistics is a new approach to linguistics which appeared in the late 1980s and it has grown rapidly at home and abroad, and has gradually turned into the major school of linguistics. To cognitive linguists, language not only enables communication, but also reflects mankind’s conceptual world. In other words, linguistic categories not only enable us to communicate, but also impose a certain way of understanding of the world. It integrates the research ways of language typology and functional linguistics, depicting and elaborating the constitution of human language.Cognitive Linguistics presents a forum for high-quality linguistic research on topics which investigate the interaction between language and cognition. Compared with linguistic structuralism, which sparked off substitution drills, and speech act theory, which initiated a complete reorganization of teaching strategies, the impact of cognitive linguistics is much less revolutionary. Yet the influence of cognitive linguistics may prove very valuable, because it lends theoretical support to a number of accepted teaching approaches in the fields of both vocabulary and grammar. The most influential linguists focusing centrally on cognitive principles and organization were Wallace Chafe, Charles Fillmore, George Lakoff, Ronald Langacker, and Leonard Talmy etc. Masterpieces including: Handbook of Pragmatics、What Categories Reveal about the Mind、Metaphors we Live by、Foundations of Cognitive Grammar、An introduction to Cognitive Linguistics etc. At present, the study of cognitive linguistics is very active in Europe and the United States .Cognitive linguistics in the two research center in the United States has formed two different schools: Berkeley School( Lakoff、Fillmore、Kay、Sweetser) and San Diego School(Langacker、Fauconnier). The Foundations of Cognitive Grammar was written by Langacker, father of Cognitive Linguistics. This book introduces a new and fundamentally different conception of language structure and linguistic investigation.

公共行政学作业3答案

公共行政学作业3答案一、名词解释 1、行政监察管辖：行政监察管辖，是指对某个监督对象确定由哪一级或者哪一个行政监察机关实施监督和哪一级或者哪一个行政监察机关对哪些特定监督事项有权进行管辖的法律制度。 2、招标性采购：是指通过招标的方式，邀请所有的或一定范围的潜在的供应商参加投标，采购主体通过某种事先确定并公布的标准从所有投标商中评选出中标供应商，并与之签订合同的一种采购方式。 3、标杆管理：是指一个组织瞄准一个比其绩效更高的组织进行比较，以便取得更好的绩效。 4、行政诉讼：所谓行政诉讼，就是公民或法人对行政机关或行政工作人员就违法行政行为向司法机关提起诉讼。俗称“民告官”。二、单项选择题 1．整个行政执行过程中最具实质意义的、最为关键的阶段是( C )。 A．协调阶段 B．总结阶段 C．实施阶段 D．准备阶段 2．对具有公务员身份的中国共产党党员的案件，需要给予处分的，由( D )给予处分。

A．检察机关 B．行政监察机关 C．党的纪律检查机关 D．党的纪律检查机关和行政监察机关 3．国家预算中占主导地位的是( A )。 A．中央预算 B．县级预算 C．省级预算 D．市级预算 4．根据《立法法》，行政法规和规章应当在公布后的( B )天内报有关部门备案。 A．15 B．30 C．45 D．60 5．批准是一种约束力较强的( A )监督方式。其内容包括：要求监督对象报送审批材料、审查和批准(含不批准)三个基本步骤。 A．事先 B．事中 C．事后 D．全面 6．从20世纪( C )年代开始，西方发达国家相继开始进行行政改革，然后许多发展中国家因为实行市场化也进行不同程度的行政改革。 A．50 B．60 C．70 D．80 7．为了解决在实施决策的过程中出现的而一时又难以查清原因的问题的决策方案，称为( D )。 A．积极方案 B．追踪方案 C．应变方案 D．临时方案 8．行政决策体制的核心( D )。

新编英语语法教程期末考试试卷.doc

大学英语语法模似试题 1. Mr. and Mrs. Burns feel more comfortable on a ship than they would be if they any other way. A. would travel B. travelled C. are travelling D. have travelled 2. We that Jim be there. A. hope B. wish C. expect D .ask 3. Lucy is glad she didn^ t stay on the farm. She bored. A. may be B. will be C. could be D. might have been 4. The dean of studies would have come to see you had it been possible, but he so busy then. A. had been B. was C. were D. would be 5. They would certainly have come and helped us time. A. did they have C. had they have 6. If you were in better health, wc. A. would have allowed C. should allow 7. She cried for her A. departed B. being departed C. departing D. having departed _____________ in an important examination, one of the students in his class lost interest in his work A. Failing B. Failed C. Having been failed D. Having Failed 9. Tf it _____________ tomorrow, P 11 stay at home. A. rained B. will rain C. had rained D. rains 10. “Tt’ s getting very late. n “Yes, it’ s time __________________________ . ” A. that we left B. we leave C. we^ 11 leave D. we have 1 eft 1 Ule ______________ a bus to go there, but he preferred to walk. A. should have taken B. could take C. could have taken IX hadn^ t taken 12. _____________ the English examination I would have gone to the concert last Sunday- A. Tn spite of B. But for C. Because of D. As for 13- _____________ , we can hardly get to the station by sixclock. A. As it wi 11 be B. As it seemed C. As it is D. As if it seems 14. We hung out a lantern lest he ___________________ lost in the mist. A. gets B. get C. will get D. got 15- Tt is quite natural that such fears _______________________ . A. rise B. should arise C. should rise D. are arisen 16. Pm sorry to _________________ your private thoughts, but T think we should get on with some work. A. break in B. break on C. break in on D. break out 17. Do you think he will_________________ a cook wearing that hat? pass for B> pass as C. pass through D. pass on B. had they had D would they have ___________ you to join in the work last week. B. would allow D. had allowed 1 over. 8.

语言学论文Linguistic and Language Learning

Linguistic and Language Learning Nowadays, in China, more and more people begin to learn English as a foreign language in order to have the ability to communicate with foreigners.For them, the first thing they should do is to learn linguistic---the guide for our language learning. On one hand, linguistics is the scientific study of human language. It can be broadly divided into three subfields of study: language form, language meaning, and language in context. With language, we can communicate to others about many things, through spoken word or in writing. That is how history is recorded, spoken or written. But language is very complex; the function of linguistics is to identify the elements in order to analyze it. For example, when we talk about a sentence we try to analyze it from four aspects: sound system, grammatical system, lexical system and discourse system. On the other hand, linguistics is the foundation of language. Without linguistics, there wouldn’t know what happened during a particular period in time. People learn to speak and process spoken language more easily and much earlier through the learning of linguistics. The language teaching practice or learning must take into account the regularities. When we learn a language, we should follow the linguistics rules.Without a solid foundation of linguistics it is hard for students to reach high levels.