What You See Is What You Meant: direct knowledge editing with
Contents
1. When satisfied that the goal is correctly defined the author pro ceeds to specify the method or methods by which appointments can be scheduled Opening the anchor these methods yields a single option labelled methods in effect the only choice here is whether to develop this optional anchor at all Since a method has several attributes all of which must be expressed through anchors the feed back text grows suddenly more complicated 5 To schedule the appointment e Before starting follow this procedure Then proceed as follows 1 Do this action by using these methods 2 Next step To quit follow this procedure e Next method Since the material will no longer fit into a single sentence the gen erator chooses a different pattern in which the methods are presented in bulleted paragraphs introduced by a TO phrase that presents the goal The rephrasing of the goal is instructive it shows that the au thor has been choosing the meaning not the words It is the generator not the author that decides how the procedure should be worded as a result the wording may change without any intervention from the author as the editing of the knowledge proceeds The model provides for more than one method since sometimes a goal can be achieved in several ways Since the author has decided that there will be at least one method the components of the first method so far undefined are shown by suitable anchors the optional anchor N2. Interpreted features These are not directly controlled but may be recovered by interpreting lower level features In a word proces sor syntactic and semantic features can be derived by interpreting the character sequence This raises the issue of who or what re covers the interpreted features A program cannot utilize syntactic 678 or semantic features encoded in a character sequence unless it has the ability to parse and understand the language Thus the seman tic features encoded in a document might be useful only to human observers 3 Presentational features These are introduced by the program in order to display the other features to the user For instance a text editor that only saves the character sequence must add graphi cal features in order to print the document or to display it on the screen We can now express the similarities and differences between text editing and WYSIWYM editing What they have in common is that the editor displays graphical graphemic syntactic and semantic fea tures this is why they look alike The difference is that a word pro cessor allows direct control over graphemic features the character sequence while WYSIWYM editing allows direct control over se mantic features the knowledge In WYSIWYM editing the graphi cal graphemic and syntactic features are all presentational The user may influence them e g by choosing French in preference to En glish but cannot control them directl
3. amp Sons Ltd Such graphical tools have the advantage of allowing what we will call direct knowledge editing as opposed to text editing When you edit a network diagram the basic operations have a direct se mantic interpretation for instance you create an object belonging to one of the permitted categories or you extend a labelled arc from one object to another thus encoding one of the permitted relation ships between the two objects By contrast when you encode the knowledge by writing a program in a text editor the basic operations of inserting and deleting characters have no direct semantic interpre tation to extract the knowledge the text must be parsed and inter preted and even experienced programmers are likely to make syntax errors Although graphical editing tools make life easier for the knowl edge engineer they fail in the objective of making knowledge edit ing accessible to domain experts Network diagrams correspond too directly to the underlying knowledge formalism they cannot be un derstood and developed without training in such technical notions as object attribute and value restriction Empirical studies have reported high error rates by domain experts using graphical object oriented modelling tools 3 and a clear advantage of text over graph ics for understanding nested conditional structures 9 1 2 Natural language input Some researchers have tried an alternative solut
4. comprises a single procedure object for which the attributes i e the goal and the methods for achieving it are undefined From this model the generator produces a single sentence feedback text 2 Do this action by using these methods The feedback text has special features related to its authoring func tion e The phrase this action is coloured red indicating that it marks a mandatory choice a location where information must be added In this case the red phrase represents the undefined goal of the procedure Since colour is unavailable we reproduce red phrases in bold face in this paper e The phrase these methods is coloured green indicating that it marks an optional choice a location where information may be added In this case the green phrase represents the undefined methods for achieving the goal We reproduce green phrases in italics e Both coloured phrases are mouse sensitive By clicking on either phrase the author opens a pop up menu from which a concept may be selected We refer to the mouse sensitive coloured phrases as anchors by analogy with the links in a hypertext Anchors can be developed in any order we will assume in this illustration that the author decides to define the goal of the procedure first followed by the methods for achieving it To begin the process of defining the goal scheduling an appoint ment the author clicks on the red anchor In response the system displays a po
5. constrained to focus on syntactic development of the current sentence rather than on semantic development of the knowledge 1 3 Anew solution We have seen that direct knowledge editing is preferable to text edit ing since it avoids any need for automatic parsing and interpretation on the other hand presentation in natural language is preferable to presentation in a network diagram or in a programming language especially if we want the editing tool to be accessible to domain ex perts as well as to knowledge engineers WYSIWYM editing attempts to have your cake and eat it by a technique that combines direct knowledge editing with presentation in natural language text The method of presenting an object oriented model through a text has been used before see 12 for a review what is novel about WYSI WYM is that the author can actually build and edit the model by in teracting with the feedback text Before explaining the technique in detail we will make some pre liminary comments in order to identify the niche that WYSIWYM edit ing occupies in the space of possible editors 2 TYPES OF EDITOR At first sight WYSIWYM editing looks like word processing because the edited material is presented as a formatted document To under stand the difference we need to look more closely at the nature of word processing When you edit a document with a word processor you aim to record various kinds of information At a basic level you de
6. of feedback texts e g how best to word the anchors nor confirmed that adequate response times could be obtained for full scale applications However if satisfactory large scale implementa tions prove feasible the method brings many potential benefits e A document in natural language possibly accompanied by dia grams is the most flexible existing medium for presenting in formation We cannot be sure that all meanings can be expressed clearly in network diagrams or other specialized presentations we can be sure they can be expressed in a document e Domain experts understand natural language much better than they understand network diagrams e Authors require no training in a controlled language or any other presentational convention Apart from the expense of initial train ing this means that there is no problem of having to relearn the conventions when a knowledge base is re examined after a delay of months or years e Since the knowledge base is presented through a document in nat ural language it becomes immediately accessible to anyone pe ripherally concerned with the project e g management public relations domain experts from related projects Documentation of the knowledge base often a tedious and time consuming task becomes automatic e The model can be viewed and edited in any natural language that is supported by the generator further languages can be added as needed When supported by a multilingual natural la
7. to French becomes a matter of presentational convenience akin to switching from small to large font The program performs no parsing or interpretation of any kind since the highest level the knowledge is directly controlled by the user and hence there are no interpreted features 3 EDITING MEANING At all levels editing consists of modifying a current configuration ei ther by inserting material at a given location or by deleting material previously inserted The initial minimal configuration must have at least one location as material is added new locations become avail able In a text editor the initial configuration is an empty string with a single location and hence a single cursor position if you type the letter A there are now two locations before and after the letter add B and there are three locations and so forth R Power D Scott and R Evans In a semantic network the configuration comprises objects of var ious types the labelled nodes in a network diagram depending on its type an object may have various attributes the labelled arcs in a diagram the value of an attribute is another object the node to which the arc points The basic editing operation is that of adding a new object of a specified type as the value of an attribute of an existing object A location can be thought of as an attribute that cur rently has no value As an initial minimal configuration we need at least one fixed attri
8. What You See Is What You Meant direct knowledge editing with natural language feedback Richard Power and Donia Scott and Roger Evans Abstract Many kinds of knowledge based system would be easier to develop and maintain if domain experts as opposed to knowledge engineers were in a position to define and edit the knowledge From the viewpoint of domain experts the best medium for defining the knowledge would be a text in natural language however natural lan guage input cannot be decoded reliably unless written in controlled languages which are difficult for domain experts to learn and use WYSIWYM editing is an alternative solution in which the texts em ployed to view and edit the knowledge are generated not by the user but by the system The user can add knowledge by clicking on an chors in the text and choosing from a list of semantic alternatives each choice directly updates the knowledge base from which a new text is then generated 1 KNOWLEDGE EDITING Many applications require editing of information expressed in a knowl edge representation formalism Expert systems are an obvious ex ample others are systems for generating documents 4 8 and for encoding design specifications 5 With currently available support tools knowledge editing has to be performed by knowledge engi neers who are familiar with the representation formalism the knowl edge cannot be directly inspected and modified by domain experts or other i
9. bute that cannot be deleted The new object added as the value of this root attribute will have attributes of its own so that further locations become available The basic idea of WYSIWYM editing is that a special kind of nat ural language text is generated in order to present the current config uration of a semantic network This feedback text includes generic phrases called anchors which mark attributes that have no value The anchors represent the locations where new objects may be added By opening a pop up menu on an anchor you obtain a list of short phrases describing the types of objects that are permissible values of the attribute if you select one of the options a new object of the specified type is added to the semantic network A new feedback text is then generated to present the modified configuration including the attributes of the new object As more information is added about a new object it will be repre sented by longer spans of text comprising whole sentences or per haps even several paragraphs These spans of text are also mouse sensitive so that the associated semantic material can be cut or copied The cutting operation removes the network fragment that was previ ously the value of an attribute and stores it in a buffer where it re mains available for pasting into another suitable location When the text is regenerated an anchor will show that the attribute now has no value and the span of text t
10. concept e g dialogue box thus redefining the action as one of opening the Appointment Editor dialogue box the effect on the text would be that both the sentences expressing this action would change This reinforces the point that the author is editing meaning not text To complete the model the author should develop the red anchor this action which is eventually replaced by the phrase Choose the Appointment option from the Edit menu At this point the model is potentially complete since it contains no red anchors Note that R Power D Scott and R Evans the model was also potentially complete for texts 4 and 6 When a model is potentially complete the author can switch the modality from Feedback to Output in order to obtain a text which sim ply presents the knowledge base without indicating the locations at which further information may be added The generator will now produce text 1 which was used at the start of this section to indicate the desired content Note that this output text has been completely regenerated it was not obtained merely by omitting the green anchors from text 7 In particular since the method for opening the Appointment Editor win dow can now be expressed by a phrase there is no need to defer it to a separate paragraph 5 SIGNIFICANCE WYSIWYM editing is a new idea that requires practical testing We have not yet carried out formal usability trials nor investigated the design
11. edge editor see 12 10 for more details of the architecture As an example of WYSIWYM editing we will describe a session in which a technical author uses DRAFTER 2 in order to define the knowledge underlying a brief passage of software documentation 679 We will suppose that the author is producing a tutorial guide to the OpenWindows Calendar Manager and is currently working on a sec tion that explains how to schedule an appointment The procedure for scheduling an appointment requires various data to be entered in a dialogue box called the Appointment Editor win dow With some simplifications it could be expressed by the follow ing text which we quote here in order to clarify the author s task 1 To schedule an appointment Before starting open the Appointment Editor window by choosing the Appointment option from the Edit menu Then proceed as follows 1 Choose the start time of the appointment 2 Enter the description of the appointment in the What field 3 Click on the Insert button The author s aim is to introduce this information into the domain model If this is done successfully the system will be able to generate the above text or an equivalent one in English French and any other supported language Assuming that the model is to be built from scratch the author begins by selecting the option New from the main menu Since the system is specialized for defining procedural models a new model
12. er and Donia Scott A support tool for writing multilingual instructions in JJCAI 95 pp 1398 1404 1995 9 M Petre Why looking isn t always seeing readership skills and graphical programming Communications of the ACM 38 6 33 42 1995 10 R Power and D Scott Multilingual authoring using feedback texts in Proceedings of the 17th International Conference on Computational Linguistics and 36th Annual Meeting of the Association for Computa tional Linguistics Montreal Canada 1998 11 Stephen Pulman Controlled language for knowledge representation in Proceedings of the first international workshop on controlled lan guage applications Katholieke Universiteit Leuven Belgium 1996 12 D Scott R Power and R Evans Generation as a solution to its own problem in Proceedings of the 9th International Workshop on Natural Language Generation Niagara on the Lake Canada 1998 13 D Skuce and T Lethbridge CODE4 A unified system for manag ing conceptual knowledge International Journal of Human Computer Studies 42 413 451 1995 14 Harry Tennant The commercial application of natural language inter faces in COLING 806 1986 R Power D Scott and R Evans
13. ext method at the bottom can be developed if the author wishes to define further methods A method comprises a precondition optional a sequence of steps obligatory and an interrupt procedure optional Each step is a pro cedure because in addition to a goal it may have methods of its own The precondition is a task that should be performed before the steps the interrupt procedure provides a way of abandoning the method if you have second thoughts Since there must be at least one step the first step is shown by a sentence with anchors for goal and methods Further steps can be added by opening the optional anchor Next step The same pattern is 680 thus used for a sequence of methods and a sequence of steps except that the former is presented by a bulleted list and the latter by an enumerated list Since the basic mechanism should now be clear we now jump to a later stage in which most of the information has been defined the only missing piece is the method for opening the Appointment Editor window 6 To schedule the appointment e Before starting open the Appointment Editor window by using these methods Then proceed as follows 1 Choose the start time of the appointment from this object by using these methods 2 Enter the description of the appointment in the What field by using these methods 3 Click on the Insert button by using these methods 4 Next step To quit follow this procedure e Next method To add the
14. fine a sequence of characters These characters make up words and sen tences which in turn express concepts and ideas The word proces sor may also allow you to influence the graphical appearance of the document e g by changing the font or the size of the characters if it supports WYSIWYG you will be able to see on the screen the appearance of the document when it is printed on paper Following the usual distinctions in linguistics we can identify four levels of information each level having its own characteristic fea tures 1 Graphical level the two dimensional bitmap 2 Graphemic level the character sequence 3 Syntactic level the words and sentences 4 Semantic level the meaning The visual display in a word processor shows all these levels How ever the editing operations mainly concern the graphemic level the insertion or deletion of characters This is the only level at which the user exerts direct control The graphical level can be influenced by altering fonts page sizes etc but the user cannot directly draw the characters The syntactic and semantic levels can be influenced only through the mediation of the graphemic level To summarize this situation we can distinguish three kinds of fea tures 1 Directly controlled features These are linked to the basic edit ing operations For instance in a word processor the character se quence is controlled directly by positioning the cursor and hitting keys 2
15. hat previously represented this value will no longer appear 4 ILLUSTRATION OF wysiwyM EDITING Our first application of WYSIWYM editing was in the context of the DRAFTER project 8 which developed a system to support the pro duction of software documentation in English and French The sys tem includes a knowledge editor with which a technical author can define the procedures for using common software applications such as word processors and diary managers in this way the author builds the domain model from which a text generator produces instruc tions in English and French that describe these procedures The eventual aim of such systems is to support the technical authors who produce tutorial guides and user manuals for software applications by automatically generating routine procedural passages in many lan guages In DRAFTER 1 the first version of the system knowledge edit ing was performed through a graphical interface in which objects in the knowledge base were presented through nested boxes with brief linguistic labels Since authors found these diagrams hard to inter pret and modify we decided to explore the new idea of presenting the growing domain model through a natural language text thus ex ploiting the multilingual generator to support knowledge editing The result was a completely re engineered system DRAFTER 2 in which the generator not only produces the final output texts but also sup ports a WYSIWYM knowl
16. ion in which the au thor defines the knowledge base by writing a natural language text which the system automatically interprets and encodes in the knowl edge representation formalism This method is obviously attractive since it would allow domain experts to specify a knowledge base in the most natural way the problem is whether it is feasible Unfor tunately information extraction from free text is unreliable for the foreseeable future natural language input can be used only if the au thor adheres to a controlled language such as Attempto Controlled English 2 or Computer Processable English 11 Many benefits of natural language input are therefore lost The domain expert can de fine a knowledge base only after training in the controlled language and even after training the author may have to try several formula tions before finding one that the system will accept Natural language input implies text editing rather than direct knowl edge editing the author defines a character sequence that must be parsed and interpreted to extract the knowledge An interesting inter mediate technique which might be called syntactic editing is used in the NLMenu system 14 sentences are composed not by typing in characters but by choosing at each stage from a list of the continua tions allowed by the grammar In this way the system guarantees that every input sentence can be parsed and interpreted the disadvantage is that the author is
17. k Kees van Deemter for useful comments on an earlier draft REFERENCES 1 AECMA AECMA Simplified English A guide for the preparation of aircraft maintenance documentation in the International Aerospace Maintenance Language AECMA Brussels 1995 2 Norbert Fuchs and Rolf Schwitter Attempto controlled english ace in Proceedings of the first international workshop on controlled lan guage applications Katholieke Universiteit Leuven Belgium 1996 3 Y Kim Effects of conceptual data modelling formalisms on user vali dation and analyst modelling of information requirements PhD thesis University of Minnesota 1990 4 Richard I Kittredge and Alain Polgu re Generating extended bilin gual texts from application knowledge bases in International Work shop on Fundamental Research for the Future Generation of Natural Language Processing Kyoto Japan pp 147 160 1991 5 B Macais and S Pulman A method for controlling the production of specifications in natural language The Computer Journal 38 4 1995 6 Robert MacGregor and Raymond Bates The LOOM knowledge repre sentation language in Proceedings of the Knowledge Based Systems Workshop St Louis April 21 23 1987 7 S Paley Generic knowledge base editor user manual Technical re port SRI International California 1996 8 C cile Paris Keith Vander Linden Markus Fischer Anthony Hartley Lyn Pemberton Richard Pow
18. last piece of information the method for opening the Ap pointment Editor window the author develops the anchor these meth ods third line This poses a problem for the generator since as we have seen the material for an expanded method will not fit into a sin gle sentence The problem is solved by deferring the procedure for opening the Appointment Editor window to a separate paragraph 7 To schedule the appointment e Before starting open the Appointment Editor window Then proceed as follows 1 Choose the start time of the appointment from this object by using these methods 2 Enter the description of the appointment in the What field by using these methods 3 Click on the Insert button by using these methods 4 Next step To quit follow this procedure e Next method To open the Appointment Editor window e Before starting follow this procedure Then proceed as follows 1 Do this action by using these methods 2 Next step To quit follow this procedure e Next method As aresult of this reorganization of the text the action of opening the window has to be expressed twice in the first paragraph it serves as the precondition in the procedure for scheduling an appointment in the last paragraph it serves as the goal of a sub procedure Of course this does not mean that there are now two actions The author might decide to cut one of the phrases the Apointment Editor window in order to replace window by another
19. nguage gen eration system as in DRAFTER 2 WYSIWYM editing obviates the need for traditional language localisation of the human computer interface New linguistic styles can also be added e g a terminol ogy suitable for novices rather than experts e Asaresult WYSIWYM editing is ideal for facilitating knowledge sharing and transfer within a multilingual project Speakers of sev eral different languages could collectively edit the same knowl edge base each user viewing and modifying the knowledge in his her own language e Since the knowledge base is presented as a document large knowl edge bases can be navigated by the methods familiar from books and from complex electronic documents e g contents page in dex hypertext links obviating any need for special training in navigation e For systems in which information must be retrieved from the knowl edge base by complex queries WYSIWYM editing can be used in order to formulate queries as well as to edit the knowledge base e For systems which generate technical documentation WYSIWYM editing ensures that the output texts will conform to desired stan 681 dards of terminology and style For instance the generation rules could be tailored to meet the constraints of controlled languages such as AECMA 1 ACKNOWLEDGEMENTS We thank all members of the DRAFTER team who contributed to our ideas on knowledge editing especially C cile Paris and Keith Vander Linden We also than
20. nterested parties Much recent research has sought to sim plify knowledge editing e g by graphical browsers or by input in controlled languages We propose here a new knowledge editing method called wysI WYM editing WYSIWYM editing allows a domain expert to edit a knowledge base reliably by interacting with a feedback text gener ated by the system which presents both the knowledge already de fined and the options for extending it Knowledge is added by menu based choices which directly extend the knowledge base the result is immediately displayed to the author by means of an automatically generated natural language document thus What You See Is What You Meant 1 1 Graphical tools Most knowledge bases are coded in object based formalisms with objects classified by a conceptual hierarchy so that attributes can be inherited as in such languages as LOOM 6 Encoding the knowl edge is often a programming task but some knowledge editing tools provide graphical browsers in which relations among objects are shown by network diagrams examples are the Generic Knowledge Base Editor 7 and the CODE4 Knowledge Management System 13 1 Information Technology Research Institute University of Brighton Lewes Road Brighton BN2 4GJ UK Firstname Lastname itri bton ac uk 1998 R Power D Scott and R Evans ECAI 98 13th European Conference on Artificial Intelligence Edited by Henri Prade Published in 1998 by John Wiley
21. p up menu listing the available action concepts choose click from which the author should select schedule DRAFTER 2 now up dates its model by filling the goal attribute with a scheduling action R Power D Scott and R Evans which includes a new attribute for the event to be scheduled as yet undefined From the updated model a completely new feedback text is generated incorporating the information just defined the schedul ing action along with a new red anchor indicating that the author must choose which kind of event is to be scheduled 3 Schedule this event by using these methods Although the anchors can be developed in any order it would be most logical for the author to continue defining the goal by click ing on this event and choosing the appropriate concept in this case appointment 4 Schedule the appointment by using these methods The goal is now completely specificed since it is shown by a phrase with no red anchors If an error has been made e g choosing meet ing instead of appointment the author can undo the mistaken choice by opening a pop up menu on the relevant span of text in this case the meeting and selecting the option Cut This will bring back text 3 so that the correct choice can be made from the anchor this event If the goal were completely wrong the author could cut the span Schedule the meeting undoing both choices and returning to text 2
22. y e g by typing in specific characters Under this scheme we can imagine four kinds of natural language editor according to the level that is directly controlled by the user 1 Handwriting Editor The user directly controls the graphical level by writing the document by hand using a pen based input device In order to utilize information at other levels e g to print a typed version of the document the editor would have to perform char acter recognition 2 Text Editor The user directly controls the graphemic level by typ ing in characters at the cursor This brings a cost and a benefit the user loses control over the detailed shaping of the characters but reliably encodes a character sequence that the editor can utilize e g by sending it through email or varying the font sizes or col umn widths 3 NL Menu The user controls the syntactic level by choosing from a list of words or phrases that are permissible extensions of the current sentence In this way the user loses direct control over the character sequence but reliably encodes a linguistic structure from the sub language that the editor can parse so that the pro gram can reliably derive the meaning 4 WYSIWYM editing The user controls the semantic or knowledge level by choosing conceptual extensions of the current meaning from pop up menus All other features become presentational so that the user has no direct control over wording switching from English
Download Pdf Manuals
Related Search