Tight Coupling:
Guiding User Actions in a Direct Manipulation Retrieval System

Christopher Ahlberg & Staffan Truvé ;

Department of Computer Science & SSKKII

Chalmers University of Technology

S-412 96 Göteborg, Sweden

Phone: +46 31 7725410

Email: {ahlberg,truve}@cs.chalmers.se

Tight coupling is a strategy in the design of query mechanisms in direct manipulation query systems. Tight coupling helps users navigate toward high precision queries in a space of database queries, avoid empty query results, and quickly narrow down the number of possible and meaningful queries. Tight coupling of a query mechanism is defined as: the results of user operations (querying, zooming, panning) on query devices (starfields, rangesliders, alphasliders, and toggles) are reflected in all query devices by visual feedback and physical constraints on meaningful query settings. An intuitive design of tight coupling is presented, along with a formal description. The design is based on a Model of Information Exploration - MIE. An empirical study in which subjects interacted with a tightly coupled interface to a jobs/skills matching database confirms the power of tight coupling.

Keywords: Information exploration, visualization, dynamic queries, database query, query refinement, tight coupling

1 Introduction

A major difficulty for users of traditional query systems based on boolean logic is controlling the amount of output from a query, observed for example by Borgman (1986) and Salton (1989). Both increasing and decreasing the result set is reported to be difficult in many systems (Borgman, 1986). A constant cause for irritation and errors is when users either retrieve thousands of documents or none at all.

To battle this all-or-nothing phenomenon several projects have tried to create incremental query systems, where queries are created step-by-step, with continuous feedback in the query process. Relevance feedback is one approach to iterative querying that has received a lot of attention. In an iterative fashion, users receive a ranked list of matching documents as query result, mark relevant documents and then perform a new search with a more relevant query. This process is referred to as the relevance feedback loop (Salton, 1989; Sanderson & Rijsbergen, 1991).

The Rabbit prototype from Xerox Parc allowed users to query a database by incrementally modifying templates presented by the system - the thousands of items that might fulfill a query are never presented, just an example which can be further modified (Williams, 1984). This approach reminds users of likely terms to use in their descriptions and guides them in the reformulation of queries. Godin et.al (1989) presents a system where a lattice representing possible queries can be traversed and thereby the query is manipulated in a controlled way. Each traversal represents the addition or removal of a query term.

Dynamic queries also allow users to incrementally create queries (Ahlberg, Williamson & Shneiderman, 1992)(Shneiderman, 1994). This is done by manipulating query devices (Ahlberg & Truvé, 1995) such as rangesliders, toggles, and alphasliders (Ahlberg & Shneiderman, 1994b)(Figure 13). However, even in a dynamic queries system the all-or-nothing phenomenon causes problems. The FilmFinder (Ahlberg & Shneiderman, 1994a), a typical dynamic queries system, includes 1838 film titles, 5468 actors, 1463 directors, etc. As the conjunction of all the query predicates set by the query devices decides the query result, only a very small fraction of all the possible query sequences (totalling >1015) leads to anything but an empty query result.

In practice, for interaction with rangesliders (i.e. a slider selecting films of length 60 to 120 minutes) this is not a major problem as the query range is incrementally increased or decreased and users receive continuous feedback regarding the size of the query result in a visualization. However, for alphasliders and toggles, which change between distinct single values (Bergman --> Bertolucci), nearly all combinations create an empty query result. Not only do those empty query results create annoyance when searching for specific objects, they also hinder users from effectively using alphasliders for examining trends for categorical variables (e.g. in the FilmFinder context, when exploring how popularity correlates with drama directors that got an award).

Figure 1. The FilmFinder prototype presented at CHI'94 (Ahlberg & Shneiderman, 1994). Films are presented in a starfield (interactive scatterplot) with the year of production and the popularity of the films as axes. Users perform queries by manipulating query devices such as rangesliders, alphasliders, and toggles.

2 Tight coupling

To create a smooth and effective direct manipulation based query environment, a query interface can be designed to be tightly coupled (Ahlberg & Shneiderman, 1994). This paper focuses on tight coupling of the query mechanism of dynamic queries systems (and other similar direct manipulation query systems). Tight coupling is defined as:

Tight coupling: results of user operations (querying, zooming, panning) on query devices (starfields, rangesliders, alphasliders, and toggles) are reflected in all query devices by visual feedback and physical constraints on meaningful query settings.

Query devices and their related query formulation mechanisms are designed to interact with each other by restricting users to query criteria that lead to non-empty query results - which also provides users with important feedback about the state of the query mechanism. This interaction should be rapid, incremental, and reversible (following principles of direct manipulation). A tightly coupled query can be regarded as a series of filters selecting a subset of a database. For each new filter that is added, users can only select filter values letting through at least one database object still existing after the last filter (Figure 2).

Figure 2. Intuitive view of a tightly coupled dynamic queries system with objects holding attributes of two types; figures and letters. After a filter has been applied only those values existing in objects which passed the previous filters are possible to select as filter values, thus avoiding empty query results.

However, this intuitive view of the query does have disadvantages. Primarily, it imposes a unnecessarily strong sequentiallity on the query. After having selected one filter value users cannot change their mind and select another value on the same level, as only those filter values existing in objects below are selectable after the manipulation.

2.1 The nature of information exploration systems

The task carried out while interacting with an information exploration system such as the FilmFinder is often exploratory in nature. The query is not well defined and constantly changes. Users must be able at any time to add new constraints to the query, but also remove others. The order of the sequence of added query constraints should therefore not be rigid and should be possible to change at any time.

To achieve an exploratory environment, while retaining the benefits of a tightly coupled interface, a query mechanism based upon a simple Model of Information Exploration (MIE) has been developed. Norman's model of how humans perform tasks, the Seven Stages of Action, is a good start (Norman, 1986). MIE does not attempt to be a predictive model for information visualizations like Card's Cost-of-Knowledge Function (Card, Pirolli & Mackinlay, 1994). Rather, it is intended to facilitate reasoning about the process of information exploration.

When performing a search, whether for a specific item, to confirm a hypothesis, or to explore trends and anomalies, users start off with a more or less well-defined goal. From this goal, an intention to act is formed and a sequence of manipulations is selected - based on the affordances of the available widgets. Notice that the sequence of manipulations may very well consist of only a single widget manipulation before evaluation of the result is done (Figure 3).

The execution of the sequence of manipulations affects the visualization and the query mechanism - and thereby triggers users to perceive the new state of the visualization and interpret the results. New pieces of information encountered will provide new directions to follow and give a new conception of the query.

Figure 3. Model of information exploration (MIE) - based upon Norman's Seven Stages of Action model.

If the evaluation of the interpretation conforms with the exploration goal the search might stop here, but probably the query will continue to shift (Bates, 1989). The close connection between search and visualization has been confirmed by our earlier experiments (Ahlberg et.al, 1992)(Williamson & Shneiderman, 1992), and also by other researchers (Rao, Card, Johnson, Klotz & Trigg, 1994).

2.2 Support from tight coupling

Tight coupling aids users through the query process in several ways:

In terms of MIE, it is important that users in the Selection and Manipulation stages can browse the range of the query device currently in focus and try different selections, without the range for that particular query device being updated. More concretely, in the film domain, after a user has selected Bergman as director, the query devices should be updated to only include those attribute values existing in Bergman's films. However, the range of the director alphaslider should not be updated since users may still want to change their selection (the view in
(Figure 2) implies this undesirable behavior). Instead, updating should be done when users indicate they want to proceed with the exploration - for example by starting manipulating another query device.

Accordingly, the view of the tightly coupled dynamic queries system in (Figure 2) needs to be refined into (Figure 4). Users are allowed to iteratively manipulate each query device, while receiving feedback in both the display and the other query devices. The range of the manipulated query device remains constant until another query device is manipulated. A formal description in terms of browsing a lattice consisting of tuples of selected database objects and sets of possible filter values can be found in the Appendix.

Figure 4. Refined view of tight coupling. At each level users can change their choice of filters before the range of possible values for the manipulated filter is updated.

Finally, a modified MIE can be presented, fitting well with the view of how a tightly coupled information exploration environment should be designed, as shown in (Figure 4). Instead of regarding the seven stages of actions as a single loop, granularity is increased through the introduction of two loops in the model (Figure 5). The inner loop maps to users manipulating a single widget and trying out different values for the corresponding query component. In the lattice structure in the Appendix this corresponds to traversing the structure horizontally. The outer loop maps to the larger activity, i.e. exploring a database, consisting of a series of manipulations of various widgets. In the lattice structure this corresponds to traversing the structure vertically.

Figure 5. Modified model of information exploration (MIE).

3 Design issues for tight coupling

The essence of tight coupling is to make more of the interior state of a query mechanism explicit to users. However, this must be done with caution - too much feedback will do more damage than good and interfere with the perception and interpretation stages of MIE (Figure 5). Below, feedback designs for a number of different widgets and situations usually occurring in information exploration systems are presented. The examples all come from the Information Visualization and Exploration Environment (IVEE) (Ahlberg & Wistrand, 1995), a dynamic queries systems generator which automatically creates an information exploration environment from a database.*1



3.1 Alphasliders

The widget that first indicated the need for tight coupling in a dynamic queries system was the alphaslider
(Figure 6) (Ahlberg & Shneiderman, 1994b).

Figure 6. The alphaslider widget. The index below the slider indicates the distribution of the strings over the slider. This particular alphaslider allows selection from two thousand filmtitles.

The index below the slider indicates the distribution of the strings over the slider, i.e. the set of selectable filter values in (Figure 4). The tight coupling updates this set as users manipulate other widgets, to reflect the current contents of the set of selectable filter values. For example in (Figure 7) the director slider has been set to Bergman, and accordingly the title slider has been updated to only include Bergman's titles. Refer to (Figure 13) for an example application using alphasliders and the other widgets below.

Figure 7. The same alphaslider as in (Figure 6) but updated to only hold Bergman's titles.

3.2 Rangesliders

Rangesliders are used for selecting ranges of attributes, primarily integer attributes although other datatypes certainly are possible (Figure 8).

Figure 8. Rangeslider for an integer attribute.

The rangeslider should effectively reflect the state of the query mechanism after a widget manipulation. A possible feedback design is where the position of the dragboxes show the minimum and maximum of the currently selected set of objects - this solution is used in IVEE (Figure 9).

Figure 9. Rangeslider where positions of dragboxes have been updated to reflect the minimum and maximum length of the currently selected set of objects.

Reflecting the maximum and minimum would be even more useful if the slider contained a density plot showing the distribution of data over the slider, such as done in Eick's data visualization sliders (Eick, 1994). A variation is to instead change the minimum and maximum values of the slider, but informal user testing showed this to be less useful.

Selecting from a continuous slider such as the one in (Figure 8) can actually generate an empty query result - even with the tight coupling scheme described above. Imagine only a few points being available for selection, as in (Figure 10). Moving the dragboxes to the gray state will cause an empty query result as no database object holds values in that range.

Figure 10. When slider dragboxes are moved into the gray state the selected range excludes all objects - causing an empty query result.

Designing for this situation is necessary. In terms of MIE (Figure 5) users should be aided in their interpretation of the query state and subsequent selection of actions - i.e. realizing that manipulating this slider is the only way to escape from the empty query result state. This is effectively presented by graying out and invalidating all query devices but the rangeslider causing the empty result.

3.3 Toggles

The last category of widgets to be discussed here is toggles. These are conceptually close to the alphasliders, although they allow the selection of multiple values. Tight coupling of the query mechanism should in this case make the toggles reflect which values are selectable - i.e. exists in some object remaining after the last query. This is effectively done by graying out the appropriate values (Figure 11).

Figure 11. Toggles after being updated.

3.4 Granularity of incrementality

Above, feedback design for various widgets has been discussed. An important issue to resolve is when appropriate feedback should be provided. Earlier it was noted that the following is essential to tight coupling: rapid, incremental, and reversible interaction between widgets. That operations should be rapid and reversible seems obvious. More interesting is at what granularity they should be incremental. Should manipulation of a slider cause all other widgets to update their ranges and visually reflect this for every mouse movement?

Early user testing showed this not to be an effective design - the visual chaos resulting from every widget reflecting its constant change in query range drew users focus from the relevant parts of the display. Therefore the recommendation for interaction between query devices to be rapid, incremental, and reversible should be adjusted to rapid, incremental upon mouse release, and reversible.

3.5 Zooming issues

In information exploration environments where geographic (spatial) and abstract visualizations are important components, zooming is an equally important operation, not only as a way of focusing geometrically on interesting objects, but to visually search and explore regions of interest in the data set - i.e. zooming is essentially a query operation. Zooming is important as it allows explorations of very large datasets where visualizations otherwise would have been unmanageable. Important work on zooming data sets are for example Bederson and Hollan's Pad++ (1994) and Jog and Shneiderman's zoombars (1994). Zoombars are used in the IVEE system (Figure 12), placed below and to the left of the visualization (Figure 13).

Figure 12. Zoombar widget which can control the size and position in one dimension of the view window in a visualization.

Realizing that zooming is a query operation leads to the suggestion that zooming should be integrated in the tight coupling concept. Zooming in or out should lead to users only being able to pose queries ranging over objects currently in view. Following the discussion above on incrementality of feedback on query operations, zooming should not affect the range of the query devices until users finish the operation, i.e. release the mouse button, to minimize visual chaos.

The zoombars placed on each side of the visualization, such as in (Figure 13), not only allow users to zoom in and out in the visualization but also provide effective feedback on the size and position of the current view of the visualization. IVEE also allows users to zoom with the mouse buttons for zooming in both dimensions of the visualization simultaneously. The feedback from the zoombar is then even more important. The tight coupling between the zoombars and the starfield visualization allows them to complement each other effectively.

The inverse effect of zooming affecting the query mechanism is queries affecting the zooming. It can easily be imagined how posing a query could automatically zoom in the visualization to an appropriate view after each manipulation of a widget. However this could create just the visually chaotic situation we want to avoid. Even if this scheme would be appropriate in some situations (such as when zooming the objects will reveal more semantic information about them), it will not affect the tight coupling. The operation of zooming into the currently selected objects could be attached to a separate widget.

4 Empirical study

To evaluate the tight coupling concept, an empirical study where subjects interacted with two designs of a dynamic queries interface was conducted. The two designs differed only in whether the query mechanism was tightly coupled (as defined above) or not. The study focused on investigating what kinds of error situations can be avoided with tight coupling.

A job seeking database from the Swedish Employment Service - a large government run national employment agency - was used for the study. The database held jobs/persons from the Göteborg region - about 1500 of each. Subjects searched the database using dynamic queries interfaces. Note that in the study all texts in the interfaces were in Swedish.

Figure 13. A dynamic queries interface to a job seeking database.

4.1 Apparatus

A Silicon Graphics Indy Workstation with a mechanical mouse was used for the study. The IVEE system (Ahlberg & Wistrand, 1995) was used for constructing the dynamic queries interfaces (Figure 13). IVEE allowed both interfaces to be created rapidly without any specialization of the code.

4.2 Participants

Eight subjects participated in the study, four from two local employment agencies and four students from Chalmers University of Technology. Six of the subjects were female and two male. All subjects had previous computer experience, the subjects from the employment agencies mainly from daily work with the AF90-system, a text-based query system for matching people with employers.

4.3 Tasks

Subjects were presented with a set of four matched tasks for each interface. The tasks were to, from a written description of the needs of for example a company, find a number of persons matching their needs. Each task set also included a question focusing on trends of various attributes. Sample tasks were:

4.4 Procedure

Each session lasted for about one hour. Subjects received both written and oral instructions, performed a few training tasks, and filled out an initial questionnaire. The subjects were seated at the computer in a quiet room and were asked to think out aloud while completing the tasks. A small cassette recorder with a built in microphone was used to record the subjects utterances. The think out aloud approach was chosen as the main object of the study was to qualitatively investigate error situations.

All subjects were exposed to both interfaces. The presentation order of the interfaces was counter-balanced. An experimenter was present during the session to answer questions, prompt subjects to continue to think out loud if they fell silent, and take notes. After completing the tasks, subjects filled out a short questionnaire regarding their subjective satisfaction.

4.5 Results and Discussion

The problems and error situations subjects ran into during the study were classified into the following categories of reasons for errors and problems (the involved stages in MIE (Figure 5) are given within parentheses and in italics):

4.5.1 Widget manipulation problems

Manipulating the alphasliders was a smooth operation for most subjects. However for the non-tight coupling condition, when there were many items in the lists attached to the alphaslider it was difficult to use it for browsing. For the tight coupling condition this was not the case, as users would usually start by selecting one or more categories of professions (the top widget in
(Figure 13)) before starting manipulating the alphasliders. This selection would typically reduce the amount of items attached to the profession and education alphasliders to 10 or 20% of their original size and thereby simplify browsing, both motorically (manipulation of the slider was easier with less items in the list) and visually (perception and interpretation of the visualization was easier as fewer changes would occur due to fewer number of changes of the item selected with the alphaslider).

Some subjects made mistakes through manipulating the wrong dragbox of rangeslider (independently of the tight coupling condition). Their intention to increase or decrease the size of the range was not mapped to the correct manipulation of the slider.

4.5.2 Empty and small query results

Empty query results were a major cause for problems in the non-tight coupling condition. Empty results often occurred from subjects concentrating too much on the query area of the screen (the area to the right in (Figure 13)) and not observing the results in the visualization. Subjects would quickly set four or five criteria leading immediately to an empty result. Obviously this skipping of the perception and interpretation of the visualization was problematic. Empty results also occurred when first manipulating for example the profession alphaslider and then browsing educations - most educations would not match to any profession and leave an empty result - i.e. when browsing the education alphaslider the visualization would be blank for almost all entries. In the tight coupling condition only educations matching entries for the selected professions would be selected and accordingly the results in the visualization would be easier to interpret.

Choosing criteria to relax after a too small or empty query result had been reached was problematic. Subjects could not readily identify criteria and select query devices to adjust or remove to increase the size of the result set. This finding suggests that the introduction of feedback indicating which query devices reduce the size of the result set most would be supportive for users trying to decide which widgets to manipulate.

Subjects were also observed adding more query criteria after an empty result had been reached (in the non-tight coupling condition). They would incrementally add a number of criteria while the result set was monotonically decreasing in size and then when the result set was empty they would attempt to broaden the query through the introduction of even more criteria. This was only observed in the non-tight coupling condition. The feedback in the query devices in the tight coupling condition probably indicated this to be less meaningful - it supported users in the interpretation/evaluation and selection stages.

4.5.3 Tight coupling

An interesting problem was observed due to the tight coupling condition. For one task in each of the matched task sets, subjects were asked to find a person for a job in profession category Y - probably for working in the profession X, where X was actually not part of category Y. Subjects first selected category Y, and in the tight coupling condition they could not find X on the profession alphaslider which caused confusion. In the non-tight coupling condition subjects found the profession X and could more easily realize that X did not belong to profession category Y.

The automatic update of the rangesliders in the tight coupling condition was not trouble free. Some subjects were confused regarding who had updated the slider - the system or themselves. Introduction of appropriate feedback should be useful for this problem.

4.5.4 Terminology & categorization

Other problems occurred, such as problems due to the terminology and categorization of jobs and educations in the database (independently of the tight coupling treatment). This was problematic not only for the subjects previously unexposed to the database (i.e. the students), but also for the employment agency employees who usually are very specialized in what area of jobs and educations they deal with.

4.5.5 MIE & empty query results

In terms of MIE, user errors and problems related to empty query results in the non-tight coupling condition were observed mainly in the Selection and Manipulation of widgets and Interpretation stages. In the tight coupling condition these problems were mostly avoided due to the smaller number of possible actions, and also the extra feedback provided in the query devices. Especially the behavior of quickly manipulating a number of widgets without realizing that the query result was emptied, was avoided in the tight coupling condition. This is not surprising - an incremental query system depends on users observing results incrementally and it is also obvious from MIE that if the perception and interpretation of the visualization is skipped undesirable query results may be the outcome.

4.5.6 Subjective satisfaction

Finally, all users responded that the tight coupling interface was the most effective and easy-to-use. One subject commented [an employment agency employee, regarding tight coupling]: "This is what it is all about - cutting down the result". They also commented generally about the system that it would be much more effective for the typical situation where an agency employee works together with an unemployed person to find a job.

5 Future work and conclusions

We believe tight coupling of the query mechanism in a dynamic queries system is an important usability step. The presented empirical study has verified the power of a tightly coupled query mechanism. Subjects avoided empty query results and problems related to those. However, the empirical study also showed us interesting areas for future work.


This work was in part supported by NUTEK, grant no: 5321-93-2760, and Arbetsmiljöfonden, grant no: 94-0525. Staffan Truvé was in part supported by Carlstedt Research & Technology AB.

We want to thank Ben Shneiderman, Ninad Jog and other members at the Human-Computer Interaction Lab at University of Maryland for a very fruitful collaboration on DQ-issues. We also want to thank Carl-Martin Allwood, Johan Hagman, Andy Moran, Ben Shneiderman, Anselm Spoerri, and Erik Wistrand for helpful comments on this paper.


Ahlberg, C., Williamson, C., Shneiderman, B. (1992), Dynamic Queries for Information Exploration: An Implementation and Evaluation. Proceedings ACM CHI'92: Human Factors in Comp. Systems, pages 619-626.

Ahlberg, C., Shneiderman, B. (1994a), Visual Information Seeking: Tight Coupling of Dynamic Query Filters with Starfield Displays. Proceedings ACM CHI'94: Human Factors in Comp. Systems, pages 313-317.

Ahlberg, C., Shneiderman, B. (1994b), The Alphaslider: A Compact and Rapid Selector. Proceedings ACM CHI'94: Human Factors in Comp. Systems, pages 365-371.

Ahlberg, C., Truvé, S. (1995), Exploring Terra Incognita in the Design Space of Query Devices. Proceedings EHCI `95, North-Holland.

Ahlberg, C., Wistrand, E. (1995), IVEE: An Environment for Automatic Creation of Dynamic Queries Applications, Proceedings ACM CHI'95:Human Factors in Comp. Systems.

Bates, M. J. (1989), The design of browsing and berrypicking techniques for the online search interface, Online Review 13(5), pages 407-423.

Bederson, B. B., Hollan, J. D. 1994, Pad++: A Zooming Graphical Interface for Exploring Alternate Interface Physics, ACM UIST'94 proceedings.

Borgman, C. L. (1986), Why are online catalogs hard to use? Lessons learned from information-retrieval studies, Journal of the American Society for Information Science 37 (6), pages 387-400.

Card, S.K., Pirolli, P., Mackinlay, J. D. (1994), The Cost-of-Knowledge Characteristic Function: Display Evaluation for Direct-Walk Dynamic Information Visualizations, Proceedings ACM CHI'94: Human Factors in Comp. Systems, pages 238-244.

Eick, S. (1994), Data Visualization Sliders, ACM UIST'94 proceedings.

Godin, R., Gecsei, J., Pichet, C. (1989), Design of a Browsing Interface for Information Retrieval, Proceedings ACM SIGIR'89, pages 32-39.

Jog, N and Shneiderman, B. (1994), Interactive Smooth Zooming of an Information Visualization. Technical report CAR-TR-714, CS-TR-3286, ISR-TR-94-94, University of Maryland.

Norman, D. A., Cognitive engineering. In D. A. Norman & S. W. Draper (Ed.) (1986), User centered system design. Hillsdale, NJ: Lawrence Erlbaum Associates.

Rao, R., Card, S.K., Johnson, W., Klotz, L., Trigg, R. H. (1994), Protofoil: Storing and Finding the Information Worker's Paper Documents in an Electronic File Cabinet, Proceedings ACM CHI'94: Human Factors in Comp. Systems, pages 180-185.

Salton, G. (1989), Automatic Text Processing: the transformation, analysis, and retrieval of information by computer, Addison-Wesley.

Sanderson, M., Van Rijsbergen, C. J. (1991), NRT: news retrieval tool, Electronic Publishing 4(4), pages 205-217.

Shneiderman, B. (1994), Dynamic Queries for Visual Information Seeking, IEEE Software November, pages 70-77.

Williams, M. (1984), What makes RABBIT run? International Journal of Man-Machine Studies 21, pages 333-352.

Williamson, C. and Shneiderman, B. (1992), The Dynamic HomeFinder: Evaluating dynamic queries in a real-estate information exploration system, Proceedings ACM SIGIR'92 Conference, pages 339-346.

Appendix: A lattice structure describing tight coupling

The intuitive model of tight coupling given in (Figure 4) is more formally described below (Figure 14). A lattice structure describes the possible selections and their results for a very small database. This approach has some similarities to the lattice structure demonstrated by Godin (1989), although an important difference is that using our approach empty query results are avoided.

The nodes in the lattice represent tuples of:

Because of space limitations only a few of the nodes are provided. The arcs represent widget manipulations and the text adjacent to the arc denotates which manipulation it represents (T=Title, A=Actor, D=Director, and C=Category). A query is regarded as a series of traversals in the lattice. Horizontal traversal corresponds to selections that are not yet fixated. Downwards traversal fixates the last traversal (this corresponds to downwards traversal in (Figure 3)). Upwards traversal corresponds to relaxing the query.

Below a very small database consisting of four films holding three attributes is used as example. Only some of the possible queries (traversals) are given. The query starts with all films (1,2,3,4) selected. Following the right path, the category SciFi is first tried, followed by selection of Drama instead. This category is fixated by the selection of Bergman as director, which is fixated by the selection of Sydow as actor.

At this stage the titles Passion of Anna and Virgin Spring, the director Bergman, and the category Drama are selectable. However, for the actor attribute, both Sydow and Björnstrand are selectable as the Sydow selection has not yet been fixated.

Figure 14. Lattice structure describing a tightly coupled dynamic queries environment - in this case a film database with four films 1 = {Passion of Anna,Sydow,Bergman,Drama}, 2= {Fanny & Alexander, Björnstrand, Bergman, Drama}, 3 = {Dune,Sydow,Lynch,SciFI, and 4 = {Virgin Spring, Sydow, Bergman, Drama}. Only a few of the possible traversals are included.