Input Models

From CS260 Fall 2011
Jump to: navigation, search

Bjoern's slides

Media:Cs260-slides-02-input.pdf

Extra Materials

Discussant's slides and Materials

Media:Input_Models.pdf

Reading Responses

Steve Rubin - 8/30/2011 14:55:21

The first paper, the excerpt from HCI Models, Theories, and Frameworks mainly asserts the importance of both descriptive and predictive models of motor behavior in HCI. The discussion was primarily focused on Fitts' Law and Guiard's model of bimanual skill. The second paper proposed a new type a pointing device called the bubble cursor. The circular bubble cursor expands such that there is always exactly one object in its range, and the paper showed that it outperformed standard point cursor in experimentation.

The first paper gives a simple explanation of both Fitts' Law and Guiard's model of bimanual skill. It surrounds these models with examples spanning from keyboards and handedness to text messaging, which serve to clarify the models. The main contribution of this paper was to present the necessity of these models of motor behavior in HCI research. Without the models, we would have no way of evaluating proposed systems of motor interaction. It inspires future research only so much as it informs the reader that there are both rigorous and metaphorical considerations that must be considered when designing interfaces.

The second paper, a typical research paper from CHI '05, attempts to convince us that we should not be apathetic about the ubiquitous point cursor: it may not be the best cursor for some (or even all) applications. By showing us a new cursor that builds on previous cursor research, and then using ISO-standardized methods to show that the new cursor is better than our old stand-by, the paper gives us reason to believe that new cursor types can be beneficial. Even if the bubble cursor itself is not widely adapted, this paper implicitly encourages us to look for alternatives to the accepted norms. The bubble cursor itself is a cool idea because there is always an object that is in its range, so the amount of time spent tracking is effectively minimized.

Because the first paper is mostly a review of prior, established research, I assume that their general argument is sound. The examples make sense, and the models are clear. In the paper on bubble cursors, I am mostly convinced by their methodology, but they made a few questionable decisions. For example, the changed colors of various elements between trials, giving the 2D experiment more well-defined "distractions" (this is mostly fine because they aren't comparing the 1D and 2D results, but I still don't know why they bothered changing the colors). They understandably used only square activation boundaries for the sake of simplicity in their calculations, but we don't know if this assumption affected their final results. I would also like to see how people performed when they had more extensive experience with the bubble cursor--most people having been using point cursors for years, and a learning curve for the new cursor was demonstrated in this paper. In general, though, their principled experimentation using Fitts' Law seems good.


Viraj Kulkarni - 8/30/2011 15:46:58

Motor Behavior Models for Human Computer Interaction, MacKenzie' briefly discusses the concept and usefulness of models, their two types - predictive and descriptive, and then goes on to discuss in detail two models that have had significant impact on the world of human computer interaction, namely Fitt's Law and Guiard's Model of Bimanual Skill.

'The bubble cursor, Tovi Grossman and Ravin Balakrishnan' presents a new HCI technique which is claimed to be more efficient than previously used pointing techniques and also presents the results of two experiments which back the above mentioned claim.


'Motor Behavior Models for Human Computer Interaction, MacKenzie'

This article presents two models that have had a significant impact on the way we see human computer interaction. It starts by discussing the concept and usefulness of models and their two types. Predictive models are based on mathematical reasoning and are good at predicting precise values and measurements. Descriptive models, on the other hand, are based more on empirical observations and lack the mathematical rigor of the predictive models but are very good at suggesting and bringing out critical arguments concerning the object or concept that they model.

The article then goes on to discuss Fitt's Law in detail and places the law in the historical context of Shannon's work on the Information Theory. The reason that makes Fitt's Law so important is that it can provide us heuristics and tools to quantify and measure the efficiency of a particular interaction device. This makes it possible to compare one HCI device with another. The Fitt's Law gives us equations to compute the movement time required to select a particular target if we know the (i) width of the target (ii) the distance of the center of the target (iii) two empirically measured constants.

The article then discusses Guiard's Model which deals with how human beings use their two hands differently. The dominant hand and the secondary hand takes up different roles that complement each other. This model is important because it is very good at identifying and specifying the different roles and showing how they complement each other.

There are two case studies at the end. The first case is about the application of Fitt's Law to determine text entry speeds for mobile phone users. The second case deals with the way humans interact with keyboards and the left-right as asymmetrical nature of it.


'The bubble cursor, Tovi Grossman and Ravin Balakrishnan'

The paper presents a new pointing technique, called the 'bubble cursor', which has been built upon previous pointing techniques. It starts with a brief description of existing pointing techniques like 'standard point cursors' and 'area cursors'. It then relies heavily on the Fitt's Law to bring out the factors that affect human usage of the pointing devices. After proposing the new technique of bubble cursors, the paper presents the results of two user experiments that attempt to show that the bubble cursor is a more efficient pointing device than the existing ones.

Through these detailed experiments, it is clear that the Bubble Cursor is more efficient than existing pointing devices. However, I feel there are a couple of problems that would need to be overcome before bubble cursor can be adopted as a default pointing device for most users. The authors have considered the problem of the user getting disturbed by a varying sized cursor. I feel that enough effort has not been spent in considering this problem. I used an online demo of the bubble cursor and I do not feel comfortable using it and would prefer the standard cursor to it. This may, ofcourse, be due to the fact that I am simply used to the standard cursor.

There is no doubt that the bubble cursor is more efficient and would end up saving user time if it is adopted. But, if the bubble cursor has to be adopted, the superiority gap between the bubble cursor and the standard cursor should be large enough to justify this transition. The bubble cursor is only incrementally better than its previous forms. I am not sure this incremental increase in efficiency is large enough to justify its adoption.


Laura Devendorf - 8/30/2011 17:58:42

Motor Behavioral Models for HCI provides an overview of descriptive and predictive models and how they are applied to HCI research. The Bubble Cursor paper provides the reader with an application of the predictive model Fitts' law.

In order to examine both articles I will focus on the role of the predictive model in the Bubble Cursor text. What struck me most about the article was that it was published in 2005 with very convincing evidence for the Bubble Cursors superiority to other cursor implementations, most notably the current single point cursor. With this in mind, I'm wondering why I have never even heard of a Bubble Cursor until now.

It could be that the cursor is multi-functional and used for much more than selecting. I drag the cursor, select multiple items and even use it as a guide while reading papers. The charts in the paper don't show me if the Bubble Cursor be effective or present a distraction in these situations? While the paper does address some of these issues, it doesn't provide solid evidence that the implementation would be beneficial in practical use, when the user is doing much for than just selecting circles.

Some of these concerns are addressed by the final argument that the bubble cursor could be an optional cursor, only to be used at specific times. Assuming that it really is the optimal cursor, you could also attribute its lack of adoption to the fact that large populations simply aren't used to it, similar to the way non-Qwerty keyboards are superior to the current Qwerty configuration but the time to learn a new method presents a barrier to adoption. If this is a case, a lesson learned from the Bubble Cursor is that models will only take you so far in understanding the "Human" in Human-Computer interaction. If nothing of what I have said above is the case, then it could be that I have managed to avoid the interfaces and tasks where the bubble cursor really shines.

These articles show me that while models, both descriptive and predictive, are useful in problem solving it's important to pick the right model and understand the target user physically and psycologically to gain maximum utility. This article relates to my own research developing visualizations and how I've learned, at times the hard way, that no matter how wonderful I think my interface is and how great it performs at specific scenarios, if it doesn't work for my target user, it doesn't work.


Derrick Coetzee - 8/30/2011 20:47:29

The chapter from HCI Models, Theories, and Frameworks summarized several prescriptive and descriptive models of human motor behaviour and their role in human-computer interaction. The bubble cursor applied one of these, Fitts' law, to demonstrate the advantage of a novel mouse pointing technique where the active clicking region of the cursor grows to encompass the nearest target.

Motor behavior models aid in the design of new interfaces by allowing their efficiency to be predicted without performing a slow and expensive user test. This was compellingly demonstrated by the bubble cursor, which was able to use Fitts' Law to successfully predict not only that their technique would be superior to standard mouse pointing, but also to predict the degree of speedup by comparing width and effective width of the target. MacKenzie et al's pressure-sensitive touchpad interface provides another compelling demonstration of the value of a motor behavior model: by duplicating the state machine model of a conventional mouse, they could effectively duplicate its advantages and avoid spurious behavior.

Motor behavior models are however, necessarily limited in a number of ways:

  • Overlooked costs: the bubble cursor authors felt user testing was still necessary due to factors like the visual distraction of the changing cursor size, not a factor considered in Fitts' Law. On phones, the one-key with disambiguation entry method is more efficient in the hands of a trained user, but errors due to incorrect word selection are more likely to lead to miscommunication than with multitap.
  • Long-term effects: a device, however efficient, that led to long-term injury such as repetitive strain injury would be disadvantageous. Such effects are difficult to anticipate or measure in a short-term study using existing models.
  • Training: highly efficient devices may be difficult to deploy due to an irrational reluctance to pay the cost of training, even if their increased efficiency would ultimately justify it. For example, the Dvorak keyboard and Graffiti one-stroke entry system are rarely used despite efficiency advantages compared to the dominant QWERTY solutions.

Bubble cursors are designed to apply to existing systems with relatively small modifications, and this is an advantage; however, as the authors remark upon in their last section, many application have little "void space," without which the method has no benefit. For example, in a typical windowing system, clicking on any inactive window will raise it to the top; arbitrary text can be selected in any web browser or document editor; and in games, clicking on the ground may be used to move a player or unit to a particular precise location. The idea of introducing modality to fix this is replacing one problem with another, and it's not clear which system is superior.

Methodologically, I was also concerned that the authors of the bubble cursor paper studied only circular targets with square Voronoi regions. This is far from the experience a typical application interface would produce, where most targets and Voronoi regions are rectangular. Little evidence is offered that this is unimportant.


Amanda Ren - 8/30/2011 21:11:49

Motor Behavior Models for Human Computer Interaction gives an overview of descriptive models and predictive model and stresses the importance of models for research and development in the HCI field.

The bubble cursor article covers how through experimentation, the bubble cursor is superior to the point cursor and object pointing by modifying its shape to maximize the target's effective width.

The model paper is important because it shows the power of predictive models and experimental models in development of human computer interfaces. With the different predictive models, human performance can be predicted without the time and resource consumming testing process. On the other hand, descriptive models give us an outline when thinking about user interactions. Given the case studies of Fitts' Law and Guiard's model, we can see that can see the importance of these models as research tools in aiding the improving or fixing current interfaces.

In terms of new results, the paper also mentions that Fitts' Law has been previously used only in time minimizing tasks, but is being applied to space minimizing tasks, which shows the flexibility and power of this model. Whereas Guiard's work is mentioned to be continuous inspiration for interactive 3D graphics. It's interesting to note that models that weren't developed specifically for human computer interaction are just as important in the development of the field. It's interesting how when the paper was written, it mentioned the growth of mobile phones and texting, and it used the standard 12 key mobile phone keypad in the case study. Now with today's technologies, instead of multitap and T9 entry methods, there are now touch keyboards and new technologies such as Swype, which could be used to provide new studies. It would be interesting if the paper had mentioned where applying these models presented results that were not expected or failed to explain results of actual user testing.

The bubble cursor paper is important because it provides an example of the signigicance of Fitts' Law (that we read about in the previous paper) in HCI research by showing how we are able to model the case of the bubble cursor even with its constantly changing size. Given the results of the experiments, it leaves consideration with our current selection technique and if it'll gain widespeard following. In summary, new results show that the bubble cursor reduces performance time, and is superior to object pointing and the point cursor. This relates to today's technologies because of the many software programs and user interfaces out there that could possibly have a better experience if users had a different selection technique. Although the experiments only dealt with circles, it would be interesting if they delt with other shapes.


Yoon Jung Jeong - 8/30/2011 21:56:59

I would like to pass on this, I am still not sure whether I will take this class or not. Thanks


Valkyrie Savage - 8/30/2011 22:38:39

Many everyday input devices are lacking in usability; by framing these usability problems in the context of models spanning from descriptive to predictive, we can attempt to correct them. One such usability improvement is the bubble mouse, which improves user performance on target acquisition tasks by dynamic modification of its size and/or shape (and thereby maximization of the effective width of targets) in a predictable way.

The textbook contributes to my understanding, I suppose, a novel intellectual framework, in that it’s really laying the groundwork for the presentation of the other paper. It has neatly tied a fairly long history of research in the area of input mechanisms into a tidy little chapter.

I like the idea of a continuum of models ranging from descriptive to predictive; it fits neatly into my mental model of how other fields somewhat within my experience seem to work (linguists range along a scale from descriptive to prescriptive, psychologists are on a multi-dimensional grid ranging from nature to nurture, developmental to evolutionary, etc.). However, I was vaguely disappointed that despite the fact that it was touted as a continuum, no example was given that seemed to be solidly situated between the two.

The right-handed vs. left-handed question of keyboard and mouse use has come up fairly often in my own life, living as I do with as many left-handed people as right-handed people. I suppose we've all considered the problems with keyboards before, and it was interesting to read a formal discourse on them. The poor interactions for righties between keyboards and mice is one of the reasons I stick to using text editors like vim that don't require mouse manipulation. Someday I’ll teach myself dvorak... in the meantime, I'm actually more interested in their discussion of mice and how they aren't "true 2D devices". That was sort of a new idea for me; I’d never really paused to apply actual mathematical insight to the basic interaction of myself and my mouse (well, now touchpad). Although the paper cited by the textbook seemed to mainly extol the values of such a thing in the context of e.g. photo manipulation and drawing programs, I can't help but imagine its contributions to games: e.g. computer pool, ping pong, or air hockey.

The bubble cursor concept seemed interesting, too, and perhaps useful. The value in studying this area of interaction is a great one; as target acquisition is such a common task, improvements can make a huge impact. The bubble cursor being obviously a natural extension of the prior work mentioned about area cursors, its ideas are certainly improvements over that earlier work. The change from a square cursor to a round one seems like a natural choice, but the decision to have a dynamically resizing interaction tool is strikingly novel. “Hit boxes” for all kinds of targets, from icons on the desktop to bad guys in a video game, have long frustrated me as a user. And while enlarging a Goomba’s hit box based on Mario’s location seems unfair, if my computer can, in some sense, “predict” or “understand” what I’m trying to click on or navigate to in my new, free copy of CS5, my life gets easier. (I wonder what sorts of interactions would pop out of applying dynamic target acquisition and a true 2D mouse.)

I wonder about a couple things related to that study: first, how many of the participants in the study consider themselves "power users" (for some definition of "power user") versus "casual users"? I think I would find my mouse behaving in such a way to be rather disorienting, at the very least until a cap was instituted on how large the targeting circle could be, which, happily, they discuss in the future work section. It certainly sounds like the bubble cursor adheres more closely to the principle of least surprise than the area cursor, but I suppose I'm curious to implement it and see if I could actually live with such a thing. I do consider myself a 'power user', and its given behavior seems, while easily predictable, quite odd. Second, I wondered at the fact that there were no left-handed participants in either the 1D study or the 2D study. Was it intentional? Maybe they just have a left-handed mouse on hand? I doubt it would make much of a difference, given that the targets alternated sides of the screen, but I'm curious. (Cue Valkyrie Googling about hemispheric and ocular dominance; she returns with the result that, since vision is controlled more or less equally by both hemispheres, Right Shift Theory, which indicates different strengths in processing that are correlated to handedness, is sadly irrelevant.)


Hong Wu - 8/30/2011 23:37:45

Main Point:

‘Motor Behavior Models for HCI’ described and explained Guiard’s model (descriptive model) and Fitts’ model (predictive model). ‘The Bubble Cursor’ proposed a new target acquisition technique which can select target faster and more easily.

Interpretation:

Guiard’s model and Fitts’ model, described in ‘Motor Behavior Models for HCI’, is the fundamental model to determine whether an interface design is good or bad. Guiard’s model emphasized on the key characteristics in a problem and it usually applied after the fact. Fitts’ model is a mathematic expression based on probability theory so it is good for quantitative comparison. Both models are easy to use. As we do research in HCI, it’s better to understand both models in depth.

Target selection is one of the basic ideas in HCI. ‘The Bubble Cursor’ further developed its target selection on area cursors. The core concept of ‘The Bubble Cursor’ was to select the nearest target of the cursor. It is simple but it works. The author operated a lot of experiment and statistic comparison to convince the reader bubble cursor is better, which I should also do in my project.

There are two major shortcomings of HCI. First, bubble cursor will select exact one target no matter how much distance between the cursor and the nearest target. Second, bubble cursor will have difficulty to handle irregular polygon.


Pulkit - 8/31/2011 1:20:30

The readings discusses and details the paradigms involved in design and evaluation of user interfaces and classifies them into either descriptive or predictive.

Predictive models (such as KLM, Hick-Hyman Law) on one hand provide mathematically rigorous ways of comparing utility of user interfaces, whereas descriptive models (eg Key action, 3 state model of graphical input) aid in identifying the critical ideas behind a interface subsequently helping in design. Fitts motivated by information theoretic constructions provides a analogy between channel capacity and human rates of information processing. This is further built upon to formulate prediction of movement times (of pointing devices) based on spatial parameters such as target size and spacing between targets. Empirically, the model is found to be really good. Also, such a model provides the developer with parameters to consider in order to make better interfaces.

Bubble cursor is a proposed pointing interface which allows target objects to be of maximum possible size in the motor space by using an adaptive cursor size. Intuitively, this should decrease the movement times in comparison to a regular cursor. Experimentally bubble cursor is shown to follow to the Fitts model and an equivalence is established between using a bubble cursor with normal target sizes with and a regular cursor with effective target sizes. Bubble cursor is shown to perform really well even in high target densities, which has conventionally been an issue in developing various pointing devices. (for eg: object pointing). This illustrates a direct use of a predictive model in comparing different methods of input. Comparison of T9 input in mobile devices with mutli-tap input is also one example.

Descriptive studies, such as of the 3 state graphical model have historically provided insights into development of things like touch pads which are extensively used. Further there are studies which indicate that humans like to use both hands for inputs - something which motivates multi touch user interfaces. Moreover, design of 2 balls mouse to capture rotation on the plane may also be attributed to such studies. Guiard's model of bimanual skill illustrates how humans like to use their hands and thus provides general guidelines of how an ideal input device should be like. It also exposes the asymmetry in the utility of conventional desktop input devices like keyboard and mouse for left and right hand users.

In the era of ever increasing data sharing, the demand of faster and better interfaces is ever increasing. The concepts presented in the reading aid in the same and are thus fundamentally important. The reading talks of 2 paradigms independently, the complementary nature of both could have been stressed a bit more. One more issue is that it would be difficult to generalize standard models such as Fitts when input will be richer and would include more modalities such as speech which is the future.




Hanzhong Ye (Ayden) - 8/31/2011 1:21:54

Reading Response 1: Motor Behavior Models for Human-Computer Interaction

This article gives introduction to varies models used in development and evaluation process of Human-Computer Interaction process. Basically the models can be divided into two categories: predictive ones (such as Fitts’ Law) and descriptive ones (such as Guiard’s model for bimanual skill). Both predictive models and descriptive models have been explained and evaluated using specific examples.

Predictive models are important in a way that they give quantitative method to evaluate the efficiency of a given Human-Computer Interaction strategy. In the Fitts’ Law example, the mathematical formula not only abstracts the inner relationship between movement time and task parameters, but also introduces a way to measure the index of difficulty (ID). In the cell phone input example, the Fitts’ law is well applied to evaluate the efficiency of several different input mechanisms. In my point of view, Fitts’ Law is an important foundation of Human-Computer Interaction research, and more predictive model like this should be developed to establish theoretical foundation for evaluation and strategy analysis.

Descriptive models are equally important, because although they do not provide measurable method, they often depict important rules upon which new interaction system or improvements can be made. Guiard’s models for bimanual skill is a great example of descriptive models and we can clearly see how it is applied in the evaluation process of current keyboard and mouse input system, including the design for scroll devices.

However, I believe a new strategy of Human-Computer Interaction can never be analyzed thoroughly based on either predictive or descriptive models, or both. This is because there are many other aspects within interaction process, and human-body individuals always have much more complicated factors needing to be taken into consideration. Nevertheless, this introductory article gives me a good first glance to the world of Human-Computer Interaction.


Reading Response 2: The Bubble Cursor: Enhancing Target Acquisition by Dynamic Resizing of the Cursor’s Activation Area

This interesting paper introduces a new strategy of target acquisition process, based on Fitts’ Law introduced in the previous article. The paper discusses the inadequate of several existing pointing strategies, innovatively introduces a new approach to enhance acquisition by dynamic resizing of the cursor’s activation area, gives explanation about the implementation process, and shows evidence of its significance in improving interaction efficiency through two different experiments.

One significant feature of this paper is its great emphasis on the application of Fitts’ Law. Fitts’ Law is not only the theoretical foundation upon which the bubble cursor strategy is designed, but also the most important way to evaluate its efficiency after implementation. This method is worth learning because such connection built with classic theory warrants the credibility of the design, and is also beneficial to analyze and improve current strategy in a more scientific approach.

However, I think there still exist some aspects of the problem that have not yet been discussed. For example, is it against most people’s daily habit of mouse using to introduce bubble cursor? It might be not hard to implement bubble cursor in a given context of distributed targets, but will it be equally feasible to implement bubble cursor in our daily used operating system such as Windows, MacOS or Linux? Will there be any problem if bubble cursor is widely used in most of today’s operating system? Such problems might deserve more consideration and discussion.

In short, although there are some blind spots existing, I like the idea of bubble cursor very much and I appreciate the scientific method in analyzing interaction efficiency in this paper. I cannot wait to re-implement the strategy by myself and experience such creative improvements in target acquisition process.


Yun Jin - 8/31/2011 1:38:45

1 Motor Behavior Models for HCI Summary: The chapter used two models which are descriptive model (such as Guiard’s model of bimanual skill) and predictive model (such as Fitts’ law) to motivate the human-computer interaction. Also, it provides several cases and examples to illustrate the importance of the two models to provide a simplification of complex interactions between human and computer.

The predictive model is significant for matching the movement limits, capabilities and potential of humans with input devices and interaction techniques on computer system. We can take Fitts’ law as an example which uses mathematical method to achieve the interaction between human and computer to explain the importance. It models human behavior as an information-processing activity and it has many applications. For instance, we can predict text-entry rates by using Fitts’ law. Bubble cursor also uses this model to enhance target acquisition by dynamically resizing of the activation area.

Another important model the article talked about is the descriptive model, such as Guiard’s model of bimanual skill. Using descriptive model arms designer with a tool to describe problems for thinking use-interaction experience. For example, we can design keyboard and scrolling especially for left-handed users based on the Guiard’s model of bimanual control.

Considering the importance of two models for human-computer interaction, there are also some more applications for both models in the area of human-computer interaction for us to discover. What’s more, some other models should also be taken into consideration for human-computer interaction.

2 The bubble cursor---Enhancing target acquisition by dynamic resizing of the cursor’s activation area

Summary: The paper describes about the bubble cursor, which is a new target acquisition technique based on area cursors and predicted by using Fitts’ law. In order to make sure only one target could be selected at any time, we should dynamically resize its activation area depending on the surrounding targets.

To compare to the pointing cursor and object pointing technique, bubble cursor significantly outperforms by showing two experiments in 1D and 2D. And the experiments suggest that the bubble cursor could be beneficial to user interfaces cause the interface layout is unchanged and individual software programs do not need not be re-written, instead the mouse driver software could be updated to have the bubble cursor in any program. A web browser is a good example of such interface.

What interests me a lot is the extension of the bubble cursor. For instance, we can explore methods for other selections, such as adding some buttons or gestures to modify the size or shape of the bubble so that it could be seen more functional. Additionally, some other transformations can also be added to the bubble cursor. More importantly, the bubble cursor could be an effective tool in 3D environment for further design.

Even though there are so many advantages of bubble cursor, I still have some questions about it. Could it be used with keyboard instead of mouse? Could it be implemented in Linux, Windows, Mac ox, or some other systems?


Yun Jin - 8/31/2011 1:38:55

1 Motor Behavior Models for HCI Summary: The chapter used two models which are descriptive model (such as Guiard’s model of bimanual skill) and predictive model (such as Fitts’ law) to motivate the human-computer interaction. Also, it provides several cases and examples to illustrate the importance of the two models to provide a simplification of complex interactions between human and computer.

The predictive model is significant for matching the movement limits, capabilities and potential of humans with input devices and interaction techniques on computer system. We can take Fitts’ law as an example which uses mathematical method to achieve the interaction between human and computer to explain the importance. It models human behavior as an information-processing activity and it has many applications. For instance, we can predict text-entry rates by using Fitts’ law. Bubble cursor also uses this model to enhance target acquisition by dynamically resizing of the activation area.

Another important model the article talked about is the descriptive model, such as Guiard’s model of bimanual skill. Using descriptive model arms designer with a tool to describe problems for thinking use-interaction experience. For example, we can design keyboard and scrolling especially for left-handed users based on the Guiard’s model of bimanual control.

Considering the importance of two models for human-computer interaction, there are also some more applications for both models in the area of human-computer interaction for us to discover. What’s more, some other models should also be taken into consideration for human-computer interaction.

2 The bubble cursor---Enhancing target acquisition by dynamic resizing of the cursor’s activation area

Summary: The paper describes about the bubble cursor, which is a new target acquisition technique based on area cursors and predicted by using Fitts’ law. In order to make sure only one target could be selected at any time, we should dynamically resize its activation area depending on the surrounding targets.

To compare to the pointing cursor and object pointing technique, bubble cursor significantly outperforms by showing two experiments in 1D and 2D. And the experiments suggest that the bubble cursor could be beneficial to user interfaces cause the interface layout is unchanged and individual software programs do not need not be re-written, instead the mouse driver software could be updated to have the bubble cursor in any program. A web browser is a good example of such interface.

What interests me a lot is the extension of the bubble cursor. For instance, we can explore methods for other selections, such as adding some buttons or gestures to modify the size or shape of the bubble so that it could be seen more functional. Additionally, some other transformations can also be added to the bubble cursor. More importantly, the bubble cursor could be an effective tool in 3D environment for further design.

Even though there are so many advantages of bubble cursor, I still have some questions about it. Could it be used with keyboard instead of mouse? Could it be implemented in Linux, Windows, Mac ox, or some other systems?


Alex Chung - 8/31/2011 1:52:54

Motor Behavior Models for Human Computer Interaction

Like every other scientific disciplines, any new approach to improve the user interface to computer system is a hypothesis that has to be tested and evaluated with experimental data. The authors discussed two categories of motor behavior models that have provided a basis for studying human behavior around computer system. They are predictive models focusing on quantitative analysis and descriptive models relying on contextual inquiry.

Fitts’ model measures the usability of a user interface by the speed and accuracy of the human’s cognition of the graphical display and one’s operation of muscle movement in response. The types of human behavior that Fitts’ model considers are: a) movement time to the target; b) the distance from the starting location to the targeted center; c) the width of the target. I agree with the case study that the arrival mobile devices with smaller screens and information intensive user interfaces make Fitts’ model an important tool to evaluate the practicality of new user interface designs on mobile devices.

While Fitts’ model is motivated to reduce the amount of user input to achieve the desired effect, Guiard’s model is interested in how the tasks are carried out by human natural behavior. Guiard’s descriptive model is a tool to provide insightful theoretical model for designers of interactive systems. For example, people with different characteristics would react differently to the same computer system. While some grown adults would find the small screens on a smartphone to difficult to operate, children with smaller fingers might be delighted with the proportion. Another example in the paper is the different performance outcomes between left-handed and right-handed users on an asymmetric keyboard.

There are tradeoffs with both models in HCI research. Basically, predictive model pushes the envelope and creates theories when the technology is not yet available. In the article “As We May Think”, Vannevar Bush was theorizing the possibility of using machine to gather, to preserve and to transfer knowledge from one person to another. Without the limit of real world implementation, he was allowed to explore a design scenario hypothetically. On the other hand, the descriptive model is a framework to help designers to understand the real world application of a design and to discover insightful idea and new usage. For example, tech editors reviewed the initial release of Apple iPad to be a coffee table computer with limited usability. Yet early adopters continued to discover new applications such as the first iPad DJ using the multi-touch screens as the turning tables. Such is the genius of the crowd. Both models have strength and weakness but they all contribute to the advance of HCI research.


The Bubble Cursor: Enhancing Target Acquisition by Dynamic Resizing of the Cursor’s Activation Area

The Bubble Cursor introduces a novel approach to maximize the pointing performance of selecting a target on a user interface to computer system. The cursor adjusts its size dynamically to minimize the empty space between the starting location and the targeted center in order to shorten the movement distance and time. To support this conclusion, the authors claim that they have tested the bubble cursor in varied conditions and showed that the Fitts’ law accurately models its performance. As the following discussion shows, the authors’ assessment is not well supported.

First of all, the computation of performance is not as straightforward as direct comparison between the plot lines. The cohort of participants in each experiments are made up differently. Since the experiments are meant to study human behavior when interacting with different user interfaces, it is important to maintain the same group to minimize variability. For that matter, it is possible that the users in the bubble cursor group were exceptionally proficient with the mouse and keyboard while the object group was not.

Secondly, in experiment 1, the authors use the expanding diameter of the bubble cursor as the target width to calculate the movement time to hit the target. While a larger bubble cursor shortens the distance between the center of target and the nearest point on the perimeter of the cursor, the accuracy is still dictated by a single point of the cursor reaching the target. A larger cursor decreases the amount of moment when the cursor has to travel around to multiple areas but the rate of accuracy does not change because the target width remains the same.

However, the bubble cursor is an interesting cross platform solution that does not rely on the operating system to adjust the width of the intended such as the zooming icons on Mac OSX. The experiments are very well thought-out and the statistics are convincing. This paper will be valuable as a template for conducting testing on new user interface systems for human center interactions.


peggychi - 8/31/2011 2:05:23

These two papers introduced one of the key factors of HCI: human movement when interacting with computers. MacKenzie gave an overview of motor behavior models, including predictive and descriptive ones, which simplify and predict human interactions; Grossman and Balakrishnan demonstrated an example of reinvention based on the Fitts' Law, while they did not simply consider distance/amplitude (A) and width (W) but dynamically resized the cursor's area, which was especially effective in a multi-target scenario.

It is important to understand, measure, and even predict user behaviors prior to design interactions. No matter how our interfaces transfer into various forms from computer monitors, phone screens, tablets, to large or shared displays, the models that the authors introduced have shown the basis and brought us insights to improve the efficiency of manipulating digital items. Grossman and Balakrishnan’s work particularly presented how a program could dynamically detect and adjust based on users' behaviors or intensions. In other words, interfaces could be adaptive with a properly designed model.

These works remind me of my previous work on photo selection during a chat session that assisted users visually illustrating their narrated stories by suggesting relevant photos in real-time. Considering chat as a prompt activity, I applied one of the common techniques, drag-and-drop, to help users select a certain photo and quickly enhance a chat message. However, chances are that users fail to drop an item because of either 1) the inconsistent distance between a photo and a message or 2) the changed sizes of message boxes that confuse users [1]. On the contrary, current photo browsing systems such as iPhoto and Picasa/Google+ have nicely included animations to enlarge a photo or to expand a photo stack when mouse hovering, for examples. This avoids additional mouse clicks and helps users target on a certain photo set (e.g. dynamic expansion increases the area of a set of photos that covers other adjacent sets [2]). Such an example shows how similar input techniques may apply in different fields.

Though various UI innovations have demonstrated certain degrees of improvement in efficiency, I wonder how we can carefully design to assist people instead of distracting. For example, the magnifying glass design on iOS for text editing was a bit confusing when it first appeared [3] because it altered the traditional static 2D GUI: It is useful for small touchscreens; however, the cognitive load of long press and enlarged partial moving view may not be suitable for frequent edits. It might be a trade-off that designers need to make.

Reference - Figures: [1] http://goo.gl/qTLzm [2] http://goo.gl/OYBNe [3] http://goo.gl/yFDlg

Ali Sinan Koksal - 8/31/2011 2:35:56

The book chapter presents two classes of models for human movement with respect to interaction with computers, namely, predictive models which are firmly based on mathematical representations, and descriptive models, which provide no quantitative measure but encourage rethinking designs and inventing new ones. The Bubble Cursor paper proposes an enhancement to area cursors, maximizing effective width by considering Voronoi diagrams as activation areas, and presents a thourough evaluation by considering a predictive model, based on Fitts' law.

Descriptive models, which are relatively simple frameworks defined verbally or graphically, set the context for studying existing designs and coming up with new ones that will enhance human-computer interaction. One example is Guiard's model of bimanual control, which draws our attention on the fact that we use our hands differently. It exhibits the bias for left-handed users when combining pointing actions with the use of power keys, which are more easily reached using the right hand; it also suggests rethinking scrolling, by delegating this relatively coarse movement to the left hand for a right-handed user.

Predictive models, on the other hand, provide an analytical model of human performance when interacting with computers. This allows for predicting performance of a design without the need to exhaustively evaluating it. A prominent predictive model is Fitts' Law for target acquisition techniques, which the Bubble Cursor considers as model. It is inspired by concepts in information theory, and aims to measure the difficulty of achieving a movement task (the "ID"), therefore gaining insight on the human rate of processing information using a given interface. Devices are then compared by their throughput. This model, as well as Guiard's model, have proven considerably useful in studying and developing interfaces.

The Bubble Cursor aims to improve on existing techniques, by outperforming the point cursor even in dense layouts, a case in which former techniques do not bring considerable benefit. The technique consists of associating Voronoi diagrams as the activation region for each object, therefore maximizing the "effective width" of objects. Furthermore, the authors show in their evaluation that the technique follows Fitts' Law where the width parameter is considered to be this effective width.

What seems to be missing in the evaluation is a real-world scenario to evaluate its performance. It would of course be harder to conduct such an experiment while controlling the parameters precisely, but it would have been interesting to see if the technique causes distraction in achieving common tasks such as browsing the web. The suggested "switching" mechanism should be approached carefully, as it could represent a considerable overhead.


Manas Mittal - 8/31/2011 2:36:33

The ‘Motor Behavior Models For HCI’ presents the role of models, develops a taxonomy for such models (Predictive vs Descriptive). The paper prominently describes the Fitts model, and the Guiard model and presents sample of each (Fitts model for text entry on cell phones, Guiard model for asymmetric keyboard and intellimouse input). The models are useful in that the inform (and enable us to evaluate) designs for new interfaces and interaction techniques theoretically, i.e., without having building prototypes. There is a clear tension between the models and prototyping, - the intellimouse is a terrible thing by Guiard model, yet, users who use it tend to like it (as evidenced by increased intellimouse clones). This brings to question the value of models (and what is being modelled) as a accurate representation of the objective to be optimized.

The bubble cursor paper [Grossman et al.] illustrates a new type of area cursor, one which is less ambiguous than other area cursors. The paper also models this interaction/pointing technique using the Fitts model. The Fitts model attempts to develop a ‘difficultly’ metric, and this is distilled into predicting and modeling the MT (Movement Time). However, MT may not accurately represent the ‘cognitive load’ associated with performing a task. I would be more curious to see researchers using brain probes to model the actual cognitive load. For example, as the bubble cursor authors acknowledge, the bubble cursor is not as predictable as a standard cursor, i.e., the user can’t easily, accurately, model how moving the cursor to a particular location will select something.


Dhjung - 8/31/2011 3:41:54

Fitt’s Law is basic principle that explains the relation between time and accuracy in terms of human movement in HCI. We can predict the time required to move to a target area by using correlation between the distance to the target and the size of the target. From the equation, there is a speed-accuracy trade off. A good example of this law is that if the link button on a web page is too small it is hard to click.

The bubble cursor is a new target acquisition technique based on area cursors. In order to improve the performance, it dynamically resizing its activation area depending on the proximity of surrounding targets. The bottom line of bubble cursor is that there is always exactly one target inside the point of activation. Todays, most desktop and laptop users use a mouse or touchpad as a pointing device. If this concept of target acquisition is used for the users widely, this technique not only improves the speed and accuracy of target acquisition, but also reduces tiredness of users.

First of all, I think that the bubble cursor can relieve users’ efforts when they try to point very small target with long distance from initial point. For instance, pop-up adds usually have one small target for closing its window, in this case, the bubble cursor enable users to get the target much easier than normal arrow-type cursor. Especially, disabled people who cannot use regular pointing devices such as mice may take advantage of this technique to increase their usability. If a person with cerebral palsy so his or her muscles became paralyzed, the bubble cursor can be an outstanding solution because it is always pointing one target instead of somewhere between targets on the screen. In addition, touch-based mobile devices including smartphones and smart pads would be another great fit for the bubble cursor. Many target objects within tiny screen cause a problem of miss-aligned touch sensing. Lastly the bubble cursor can be used for various applications using GUI without changing layouts or core source codes. It is largely because developers will modify pointing mechanism to implement the bubble cursor or OS might support this additional functionality.

Even though the bubble cursor had lower movement times and lower error rates, the experiments they conducted presume very limited users’ activity such as selecting or pointing tasks. General users’ computing environment is not simple as they assumed. In other words, users always change their movement. Today’s Rich Internet Application requires various movements for interaction (Drag and drop, typing, mouse wheeling, or even multi-touch gestures are very common movement). If they conducted the experiments with above actions, the result might not positive as it is. Furthermore, the algorithm for the bubble cursor requires that every target should be a circle so that it cannot support irregularly shaped objects, which are the most common targets on user interfaces.


Apoorva Sachdev - 8/31/2011 3:52:58

The reading was about the advancement of a new kind of cursor technology that builds and improves on area cursors. It was an experimental quantitative analysis of how efficient bubble cursors with dynamically changing bubbles to select targets are.

I believe this paper is important because it captures the process of improving something and validating it in the field really well. The user participation tests allow the authors to really test their hypothesis about the bubble cursor following Fitts’ Law, studying its ease of use and reducing the “movement time”. I personally haven’t used Bubble cursor yet; however I do believe there is a lot of potential for improvement in this area because a standard keyboard and point-mouse are still very limiting in allowing easy mobility. While I was reading the paper and looking at a video demo of bubble cursor, I was impressed with the selection technique it uses to capture a target but at the same time I found the continuously changing “Bubble” of the cursor very distracting at times and I guess work could be done to improve the transitions between the varying bubble scopes.

I was pretty impressed with the way the timing analysis was done and the experiments were carried out testing different variables of the Fitts’ Law, one at a time. However, I do feel like the idea of selecting one green target and then another, beyond a point becomes a reflex action game more than an ease of use test, and comparing reaction times may not be the best way of judging “movement” time. Also, a lot of things have now migrated to Touch screen technology since then and it would interesting to see how something similar to bubble cursor could be used to facilitate easier movement on smaller devices like smartphones, tablets etc.


Suryaveer Singh Lodha - 8/31/2011 4:05:55

The paper presents a novel variation of area cursors for selection technique. The “bubble cursor” follows Fitts’ law and significantly outperforms point cursor and overcomes the limitations of other variations of area cursors developed till date. Fitts' law suggests that if either the target size(W) increases or the distance of target from cursor (A) decreases, then the time taken to select it will decrease. Prior research in this area was aimed at just decreasing A, which got good results only for relatively sparsely laid out targets. Increasing target densities, which are usually a case in common-day GUI tools, proved to be detrimental to such techniques. The other approach adopted was to increase W, but this approach also has a similar issue when the target is densely distributed. A variation to this approach is to have a hotspot at the centre of area cursor, but then the result is same as a point cursor when the target is densely populated. Another interesting approach is to increase target size when the cursor approaches them. This works well in sparsely distributed targets, but to date it has not been shown if this works well when the target is densely populated. Other approaches which tried to increase W and decrease D simultaneously also failed in dense target distributions. “Bubble cursor” uses a circle instead of tradition square cursor and also dynamically changes size based on proximity to the target. Bubble cursor technique also makes use of “Voronoi” diagrams to break the target space into regions (activation zones) such that there is exactly one target in each region and that target is the closest target to any point in that region. However, in this paper, the author has simplified the case by considering only square activation zones. The author not only tries to prove that the bubble cursor does follow Fitts’ law, but is also concerned about other aspects such as visual distraction due to constantly changing size of cursor and difficulty of a user to plan and execute the required motor movements due to introduction of this new kind of cursor. The experiments conducted were quite detailed and the results did compliment the assumptions made by the author. However, the point where it was a bit ambiguous was the way the bubble cursor was compared against object pointing technique. As the test cases were specific, so as to keep square activation zone, keep the effective width of targets in specific range and also strategic placement of target points, it might not be the best data-set to compare object pointing technique and bubble cursor. No doubt that bubble cursor does perform well (even better) in densely distributed target points, but probably it would be a good idea to compare it against other techniques in real-time simulated test cases with no specific pre-ordering/alignment and no bounds on density/effective width of the target points. Fitts’ law (predictive model) is the paramount model for pointing device research. Fitts’ law appears in the ISO standard: 9241-9. Fitts’ law brings consistency to empirical evaluation of computer pointing devices. Throughput is regarded as the major evaluation parameter when Fitts’ law is applied. It was interesting to learn how Fitts’ law can be applied to predict text entry rates on mobile phones (Definitely the assumptions that all words entered using T9 were in dictionary and that if ambiguity arises the most probable word is the word selected, did make the predictive model simpler than it really is) Guiard’s model (a descriptive model) primarily focuses on the examination of the assignment of tasks to hands (A descriptive model of bimanual skill). The way Guiard’s model can be applied to derive that current desktop systems are biased towards left handed users was insightful. Guiard’s method will be very critical to design recommendations for devices which involve humans to use both their hands for interactions. One interesting area where I think Guiard’s law will be critical is in development of multi-touch user interfaces, which allow humans to interact with the device using both hands.


Galen Panger - 8/31/2011 7:04:49

My responses to the bubble cursor and motor behavior pieces are pretty mild. The bubble cursor I felt represented a very good, but ultimately limited, idea for enhancing the effective width of screen objects given their relative positions. On the one hand, I’m impressed by the movement time reductions achieved, and find it remarkable that there was little observed learning or distraction overhead. In addition to these and other benefits mentioned by the authors, I think the bubble cursor might also improve object discovery, given that the bubble is constantly morphing based on what is nearby, perhaps causing the user to notice when the bubble morphs unexpectedly. Finally, the bubble cursor is a more elegant solution than many of the options presented for expanding effective width because it does not require that the interface elements themselves be modified (with the exception of giving all interface elements a hover-over state, which the bubble cursor experiments required).

The bubble cursor may ultimately be unsatisfactory from an elegance standpoint, however. First, though there does not seem to be evidence that users were distracted in their targeting behavior by the ongoing shifts in the cursor's shape and the constant highlighting of hovered-over interface elements, I do wonder what happens to their train of thought in regard to the overall task they are executing. Second, and most important, is the incompleteness of bubble cursor behavior: activation of void space is not designed into its behavior, which is a major omission and something I think that ultimately confines the cursor to specialized rather than generalized use. If you have to switch cursors manually, you've already lost most users.

Transitioning into the motor behavior piece, it's interesting that the bubble cursor article emphasizes Fitts' Law so heavily, because perhaps including a descriptive model of standard cursor behavior would have lent itself to a more thorough evaluation of the bubble cursor's generalizability. A descriptive model of cursor behavior would no doubt include the activation of screen objects as well as void space, and include behaviors such as clicking and dragging, multiple-object selection, and other tasks only briefly considered by the authors. As a side note, Fitts' analogy to Claude Shannon's work seems like an interesting historical accident—would Fitts' law draw those connections if it were developed today?

Finally, I was struck by Guiard's Model of Bimanual Skill. That our hands are specialized in terms of what we use them for, I think, has implications for the design of many, many interfaces (any that utilize both hands, obviously). That keyboards are optimized for left-handed people was lost on me, however, as I am left-handed and have always operated the mouse with my right hand. What a waste! An unmentioned but important additional detriment to right-handed keyboard/mouse users is the extra time needed to move the right hand from the mouse over the number pad and executive keys, to the home row (and back). That does seem like a lot of unnecessary work.


Vinson Chuong - 8/31/2011 8:12:21

The MacKenzie paper surveys various quantitative and qualitative models for evaluating the effectiveness of a user interface design and arrives at a framework both for guiding the refinement of existing interfaces and for quickly estimating the viability of entirely new interfaces. The Grossman and Balakrishnan paper uses this framework to identify a refinement for target acquisition via mouse and is able to arrive at significant gains in user efficiency.

Models which "provide a simplification of the complex interaction between humans and computers...allows designers to understand and anticipate the impact of a design in a meaningful context". Even more than that, as demonstrated in Grossman and Balakrishnan, models reveal inefficiencies in designs and can guide their refinement. From filtering brainstormed ideas for viability, to iterating a design before prototyping, to explaining results of user testing on a prototype, and to continually refine the product, models enrich the entire process of designing an interface.

For example, we can apply the Keystroke-Level Model, the Three-State Model of Graphical Input, and Fitts' Law, and Guiard's Model to multitouch tablet input and arrive at illuminating comparisons between multitouch input and keyboard-mouse input. From my experience, I'm far more efficient in using my hands to directly provide input (multitouch tablet) than I am in using a mouse; this suggests different timings when using the Keystroke-Level Model. In particular, when "tracking", instead of sliding a mouse around and seeing where it ends up on screen, when using a tablet, there's a direct mapping of where my hand ends up on the screen and what on the screen I'm interacting with. Finally, with multitouch, I can use both my left and right hands; do different hands perform specific actions more efficiently as suggested by Guiard's model?

While out of the defined scope of these two papers, I would have liked to see more discussion of how humans process output/feedback (auditory, visual, haptic, etc.) The MacKenzie paper mentioned the Hick-Hyman Law, and the Grossman and Balakrishnan paper mentioned a possible side-effect caused by the bubble cursor changing size visually. I recall several papers presented in CS160 which address this area in more depth.

Overall, I'm very interested in going hands-on and learning more about the power of these models.


Jason Toy - 8/31/2011 8:29:50

Summary: "Motor Behavior Models for Human-Computer Interaction" describes the importance of input devices and explores the use of models in HCI. It gives two examples of models (Fitts' model and Guiard's model) and uses case-studies to illustrate how particular conclusions can be derived from these models.

How the paper contributes to our understanding of human-computer interaction: This paper is similar to "As We May Think" by Vannevar Bush in that they both stress the importance of input devices and their improvement in order to further HCI. The idea proposed of using models for research and development is important because it can help predict the results of a design. The first case study, through a variation of Fitts' model, offers a new methodology in determining what texting methods are best in terms of wpm. In the second case study, the paper uses Guiard's model to come to the new conclusions that the standard 101 keyboard is better suited for left handed people, and that scrolling devices should be in the non-dominant hand. The paper is important and relevant to real-world systems because it describes ways of judging future designs of input devices such as keyboards and mice through models. This can guide the next generation of input devices because you can come up with alternative designs (Descriptive Models) or performance metrics of your design (Predictive Models). Changes might be made to your input device design or the design might be scrapped before an expensive production and testing phase if its performance is predicted to be worst than current offerings.

Strengths and Weaknesses: The paper does a good job acknowledging some of the limitations of the argument. For example, the predictive texting model assumes that everyone is a perfect typist, which is definitely not true, and thus the true value of the wpm should be lower. Another thing done well is the acknowledgement of alternative solutions to the conclusions they propose: for example, the scroll wheel on a mouse in your dominant hand in contrast to their proposed scroll bar used by your non-dominant hand. However the paper does a poor job acknowledging alternative forms of input devices otherwise. The conclusions they come to might be moot should the mainstream method of input change. In "As We May Think" by Vannevar Bush, Bush discusses the use of voice to text and reading brain impulses as alternative forms of input. Both of these would render discussion about keyboards and dominant hands useless. The advent of the "slider" cellphone (phones with keyboards that can slide outward), blackberries, and other smart phones have begun to render the old methods of texting with a keypad, analyzed in this article, extinct. Another problem with the paper was that it was very theoretical, with little experimental data backing it up. The paper appears to try and prove to the reader that models can be useful: I argue that a model, or a prediction (in the case of predictive models), can only be useful if it correctly approximates real life. Using the model to compute the wpm of texting methods could have been compared to experimental data to illustrate to the reader that analytical data is just as good as data from time-consuming experiments: proving that predictive models work. Instead, I am left only with a numerical prediction generated by a formula. How would I know if this number is remotely correct? What if the authors have made a mistake and this phenomena can not be modeled with Fitts' model?

Summary: "The Bubble Cursor" presents an idea for a new selection cursor which expands its activation area to encompass one target at a time, enlarging the effective width of targets, and making them easier to select. Two experiments are detailed in which the performance of the bubble cursor is established.

How the paper contributes to our understanding of human-computer interaction: The first three papers we read all stress the importance of input devices in HCI. However, Scott MacKenzie attempts to improve the experience by focusing on the physical nature of pointing devices, while Toni Grossman and Ravin Balakrishan's bubble cursor focuses on the visual aspect, and many of Vannevar Bush's ideas bypass pointing devices altogether. "The Bubble Cursor" presents a new technique with a larger hotspot for selecting targets on a screen, in comparison to the standard point cursor. It can be used in real-world systems to reduce the amount of time a user spends selecting targets in a gui based operating system, or programs such as Adobe Photoshop, and improve his or her experience. Current hardware products does not need to adapt to the bubble cursor, however future software products can be designed around using the bubble cursor to take advantage of the effective width of targets rather than width the pointer cursor's performance depends on. As stated in the article, future research could be done in seeing how the bubble cursor's performance can hold up in an actual gui environment. Research could also be done in determining the pitfalls and difficulties of making this a feasible offering on standard operating systems. Given the current software offerings today, would a bubble cursor make sense? Would it become clunky and obtrusive? How long would it take for an average person to adopt this system?

Strengths and Weaknesses: One thing that impressed me about this paper was the attention paid to details when recording procedures and data of experiments. The other is a seemingly strict following of scientific methodology i.e. control and experimental groups. "Participants were randomly divided into 2 groups of 5 each, with one group using the point cursor first, and the other group using the bubble cursor first". The very thorough job done by the experimenters allows the experiments to be repeated by the authors or others, and provides greater respectability to their results. One weakness of the experimentation is that the there was no experiment done in a gui-like environment. It its very likely the results would be different, though you might still see improvement with a bubble cursor over a point cursor. Most gui operating systems and programs do not have targets that resemble circles. Even worse, real world programs usually have dashboards which have bunched up targets (buttons, menus, etc) which are right next to each other. The paper acknowledges that "tiled" targets means the bubble cursor has no advantage since effective width is equivalent to width. Another weakness in experimentation was possibly in the use of volunteers. The use of volunteers led to a very biased group: everyone was right handed and between the age of 18-25. We don't know if either the hand you use to control the input device, or the age you are affects the results. Maybe younger people are able to grasp the situation more quickly than other people of other age groups. Although I do acknowledge that the limited number of participants and the use of volunteers might have been due to some unavoidable constraint.

Opinion: I wonder if there was an inadvertent bias towards a point cursor in the experiments. If you get people who have been have been using point cursors all their lives, to try out a bubble cursor that they had 1 trial to get familiar with, it may be possible that point cursors have an advantage due to peoples' familiarity with them. There isn't much I would think the experimenters could do about this, other than give the participants a little more time to get acquainted with the bubble cursor.


Rohan Nagesh - 8/31/2011 8:44:21

The first reading, "Motor Behavior Models for Human Computer Interaction," describes what motor behavior models are, why they're useful for HCI, and discusses in particular differences between predictive and descriptive models. The second reading, "The Bubble Cursor: Enhancing Target Acquisition by Dynamic Resizing of the Cursor’s Activation Area," discusses a novel pointing technique to decrease movement times and potentially increase accuracy.

I found the first reading important and useful as it gave me a very high-level overview without bogging me down in mathematical details regarding the value of modeling in HCI. Models can be useful to designers in engineering new interaction techniques and running experiments to test these techniques' efficacy. The article also provided specific instances of each model, including deep-dives on Fitts' Law (a predictive model) and Guiard's Model of Bimanual Skill (descriptive model).

I particularly found the case studies regarding text entry rates on mobile phones and bimanual control and desktop affordances to be interesting as well as relevant to our times today. Clearly, mobile is blowing up faster than ever, and one significant limitation still remains today--the speed at which we can communicate physically on the device lags our speed in thinking tremendously. While the article only examined multi-tap and T9 (technologies which have existed for a long time now), it did establish a framework for using Fitts' Law to model interaction times. With new text entry methods today such as Swype, it would be interesting to determine whether Fitts' Law can actually be applied and measured.

The second reading introduces a novel target acquisition technique known as the Bubble cursor. Having played with the bubble cursor before in CS160, I found that it improves performance and accuracy in low-density UI's, but leads to more errors and slower times in high-density UI's. As the article mentions, today's web UI's are becoming increasingly dense, and I do not believe a bubble cursor interaction technique as described in the papaer will hold up.

This paper is important in that it recognizes the need for improving on the current interaction techniques. Indeed, current pointing techniques are not perfect, and it'll be interesting to see if enhanced multi-touch trackpads such as those from Apple will solve some of the problems the Bubble cursor set out to solve. It was also important for me to once again see that Fitts' Law can be used to compare acquisition times among various techniques, and the authors did indeed use Fitts' Law to calculate movement times for both the Bubble Cursor and status-quo pointers.


Yin-Chia Yeh - 8/31/2011 8:49:34

<Motor Behaviour Models for Human-Computer Interaction> This paper is an introduction of two main categories of modeling method in HCI research, predictive modeling and descriptive modeling. The author first introduces what these two modeling categories are and why HCI researchers find them useful. He then gives one example for each category, which are Fitts’ model for predictive modeling and Guiard’s model of bimanual control for descriptive modeling.

This paper is a very good first paper in HCI. It reveals the diversity of HCI research by introducing two very different modeling approaches. At one end, predictive models help researchers to evaluate the performance of their researches without carrying out too many experiments. At the other end, descriptive model helps researchers examining if certain interaction behaviors fit human body or not. As can be seen this two models represent two very different research directions in HCI. Your research can be very engineering oriented or focus on user experience. The paper also helps people to catch some common terminologies in HCI field so they can read other HCI papers easier later.

The two case studies given in this paper, text entry rate and desktop computer affordance, are particularly helpful since they are in our daily life. In the text entry rate case, the text entry task is modeled as 27^2 different target acquisition task. I wonder if it is best modeling because in my experience our fingers memorize common words instead of common alphabet transitions. In the desktop computer affordance case, I think a question worth of investigating is how often do we use those duplicated function keys on right hand side. I myself merely use ctrl and shift key at right hand side, but I do use page up/down keys a lot and it’s really a trouble when I want to use mouse at the same time.

<The Bubble Cursor: Enhancing Target Acquisition by Dynamic Resizing of the Cursor’s Activation Area>

This paper presents a new target acquisition technique – bubble cursor. This method guarantees there is always one and only one target captured by the cursor, which resolves the multiple targets problem in previous works on area cursor. The concept of bubble cursor can be best represented by a Voronoi diagram where the display window is segmented according to nearest neighboring targets. The experiments shows that this method outperforms traditional pointing method in various settings which previous works do not work well, especially in high target density environments.

The first thing I like about this method is it is optimal on enlarging targets’ effective width, which is its most difference to previous works. The adaptive effective target width is determined automatically by the spatial layout so we don’t need to decide the size of activation area heuristically. I also like the double bubbles GUI drawn in Fig. 4b. The algorithm itself does not really operate according to the double bubbles but it is much easier for user to catch which object is targeted.

The experiment procedures are well designed so that even ordering of different cursor type is considered. The authors also apply statistical significance test to verify the experiments are meaningful in statistics. I think the experiment results are pretty convincing except the experimental results show that the effective width of bubble cursor works exactly the same as the actual width of traditional cursor. I can understand this result but it is a little mysterious how it could happen.

There are several things I would like to verify. First, the paper states that this method can be easily inserted into mouse driver. I thought mouse driver only taking care of low level stuffs and it should not know the layout of the screen. I think this method should be implemented in somewhere higher than mouse driver, maybe in OS, so that every different mouse can share the same algorithm. Secondly, I would want to know how this method works on common object shape such as rectangular. It’s easy to compute the Voronoi diagram if the shape of targets are limited to circle. I would want to know the computational cost. The last thing to mention is that I would like to know why the authors choose to compare with object pointing method in experiment 2. The experiment design forces that there is always a target on the path from the cursor to the target, which is particularly undesirable to object pointing method. It’s not surprising that object pointing method works even worse than traditional pointing method in such experimental setting.


Sally Ahn - 8/31/2011 8:51:49

In "Motor Behavior Models," MacKenzie presents two ways of modeling HCI interactions (predictive and descriptive) and analyzes two exemplary models: Fitts' Law and Guiard's model of bimanual skill. The "Bubble Cursor" paper improves existing target selection methods with dynamic resizing of target areas that remain effective in environments with densely populated targets; thus, it exemplifies an application of Fitts' Law, a model MacKenzie describes in detail.

As MacKenzie points out, the benefits of modeling are obvious: it allows us to evaluate and compare the efficacy of different interaction designs without having to test each functionality of the device manually. The detailed analyses of the two models presented in this chapter made me wonder about their possible shortcomings. One aspect I saw missing in these evaluations was the ergonomics factor. Fitts' Law, for example, measures throughput, and thus focuses on the time it takes the user to complete tasks. Granted, one could argue that shorter task completion times generally coincide with reduced probability of RSI (and this would be the case for the one-key with disambiguation in the mobile phone example), but it seems possible that some interactions may promote better posture at the cost of time efficiency. In addition, although Fitts' Law successfully models the point-and-click paradigm, which has led to innovations like the Bubble Cursor, this paradigm may become obsolete as touch screen devices and multi-touch gestures gain popularity. With multi-touch gestures, we now have the option of one-finger taps, two-finger taps, pinching, swiping, etc. anywhere on the screen, which renders the distance to a "target" to virtually zero. Nevertheless, touch screen interactions also present new problems such as finger obstruction and lack of precision. Perhaps a more complex model is needed for these interactions.

It seems that the descriptive nature of Guiard's model provides a more general model that can inspire new interactions for modern devices. For example, most gestures today assume one-hand interactions, which makes sense for handheld mobile devices. However, can we lessen the burden of the preferred hand from constant dragging, tapping, and pinching by allowing the nonpreferred hand to do a little more than idly holding the mobile device? The scrolling example from the reading made me wonder if a similar idea can be applied to mobile phones: a scrolling wheel on the left side of the phone that can be scrolled with the thumb of the hand holding the device. Of course, such devices would create more difficulty for left-handed people.

Since there is yet to be a standard layout for the relatively new touch screen devices, we no longer have the problem with the classic keyboard, which we analyzed too late. Many soft keyboards today still reflect the old QWERTY layout which were designed for sight-independent interaction; clearly, the parameters for interaction change completely with the touch pad keyboards, and I wonder if Guiard's model may help us invent a new layout that is better suited for modern devices like tablets.


Jason Toy - 8/31/2011 8:53:45

Summary: "Motor Behavior Models for Human-Computer Interaction" describes the importance of input devices and explores the use of models in HCI. It gives two examples of models (Fitts' model and Guiard's model) and uses case-studies to illustrate how particular conclusions can be derived from these models.

How the paper contributes to our understanding of human-computer interaction: This paper is similar to "As We May Think" by Vannevar Bush in that they both stress the importance of input devices and their improvement in order to further HCI. The idea proposed of using models for research and development is important because it can help predict the results of a design. The first case study, through a variation of Fitts' model, offers a new methodology in determining what texting methods are best in terms of wpm. In the second case study, the paper uses Guiard's model to come to the new conclusions that the standard 101 keyboard is better suited for left handed people, and that scrolling devices should be in the non-dominant hand. The paper is important and relevant to real-world systems because it describes ways of judging future designs of input devices such as keyboards and mice through models. This can guide the next generation of input devices because you can come up with alternative designs (Descriptive Models) or performance metrics of your design (Predictive Models). Changes might be made to your input device design or the design might be scrapped before an expensive production and testing phase if its performance is predicted to be worst than current offerings.

Strengths and Weaknesses: The paper does a good job acknowledging some of the limitations of the argument. For example, the predictive texting model assumes that everyone is a perfect typist, which is definitely not true, and thus the true value of the wpm should be lower. Another thing done well is the acknowledgement of alternative solutions to the conclusions they propose: for example, the scroll wheel on a mouse in your dominant hand in contrast to their proposed scroll bar used by your non-dominant hand. However the paper does a poor job acknowledging alternative forms of input devices otherwise. The conclusions they come to might be moot should the mainstream method of input change. In "As We May Think" by Vannevar Bush, Bush discusses the use of voice to text and reading brain impulses as alternative forms of input. Both of these would render discussion about keyboards and dominant hands useless. The advent of the "slider" cellphone (phones with keyboards that can slide outward), blackberries, and other smart phones have begun to render the old methods of texting with a keypad, analyzed in this article, extinct. Another problem with the paper was that it was very theoretical, with little experimental data backing it up. The paper appears to try and prove to the reader that models can be useful: I argue that a model, or a prediction (in the case of predictive models), can only be useful if it correctly approximates real life. Using the model to compute the wpm of texting methods could have been compared to experimental data to illustrate to the reader that analytical data is just as good as data from time-consuming experiments: proving that predictive models work. Instead, I am left only with a numerical prediction generated by a formula. How would I know if this number is remotely correct? What if the authors have made a mistake and this phenomena can not be modeled with Fitts' model?

Summary: "The Bubble Cursor" presents an idea for a new selection cursor which expands its activation area to encompass one target at a time, enlarging the effective width of targets, and making them easier to select. Two experiments are detailed in which the performance of the bubble cursor is established.

How the paper contributes to our understanding of human-computer interaction: The first three papers we read all stress the importance of input devices in HCI. However, Scott MacKenzie attempts to improve the experience by focusing on the physical nature of pointing devices, while Toni Grossman and Ravin Balakrishan's bubble cursor focuses on the visual aspect, and many of Vannevar Bush's ideas bypass pointing devices altogether. "The Bubble Cursor" presents a new technique with a larger hotspot for selecting targets on a screen, in comparison to the standard point cursor. It can be used in real-world systems to reduce the amount of time a user spends selecting targets in a gui based operating system, or programs such as Adobe Photoshop, and improve his or her experience. Current hardware products does not need to adapt to the bubble cursor, however future software products can be designed around using the bubble cursor to take advantage of the effective width of targets rather than width the pointer cursor's performance depends on. As stated in the article, future research could be done in seeing how the bubble cursor's performance can hold up in an actual gui environment. Research could also be done in determining the pitfalls and difficulties of making this a feasible offering on standard operating systems. Given the current software offerings today, would a bubble cursor make sense? Would it become clunky and obtrusive? How long would it take for an average person to adopt this system?

Strengths and Weaknesses: One thing that impressed me about this paper was the attention paid to details when recording procedures and data of experiments. The other is a seemingly strict following of scientific methodology i.e. control and experimental groups. "Participants were randomly divided into 2 groups of 5 each, with one group using the point cursor first, and the other group using the bubble cursor first". The very thorough job done by the experimenters allows the experiments to be repeated by the authors or others, and provides greater respectability to their results. One weakness of the experimentation is that the there was no experiment done in a gui-like environment. It its very likely the results would be different, though you might still see improvement with a bubble cursor over a point cursor. Most gui operating systems and programs do not have targets that resemble circles. Even worse, real world programs usually have dashboards which have bunched up targets (buttons, menus, etc) which are right next to each other. The paper acknowledges that "tiled" targets means the bubble cursor has no advantage since effective width is equivalent to width. Another weakness in experimentation was possibly in the use of volunteers. The use of volunteers led to a very biased group: everyone was right handed and between the age of 18-25. We don't know if either the hand you use to control the input device, or the age you are affects the results. Maybe younger people are able to grasp the situation more quickly than other people of other age groups. Although I do acknowledge that the limited number of participants and the use of volunteers might have been due to some unavoidable constraint.

Opinion: I wonder if there was an inadvertent bias towards a point cursor in the experiments. If you get people who have been have been using point cursors all their lives, to try out a bubble cursor that they had 1 trial to get familiar with, it may be possible that point cursors have an advantage due to peoples' familiarity with them. There isn't much I would think the experimenters could do about this, other than give the participants a little more time to get acquainted with the bubble cursor.


Shiry Ginosar - 8/31/2011 8:57:29

Book Chapter: Motor Behavior Models for HCI:

This book chapter introduces human behavior models and their use in HCI research specifically concentrating on models used for matching human behavior capabilities with interaction techniques.

The chapter starts by defining predictive (mathematical a-priori) and descriptive (a distillation of observations) models. it then presents in detail Fitts' Law (predictive of pointing) and Guiard's model (descriptive of bimanual) skill, and finally it provides examples of usages in HCI research of both with the necessary mentions of the pitfalls and caveats of each model.

Being an overview of useful models this paper inspires future HCI research specifically in the areas of pointing devices and bimanual user interfaces as such research can rely on the models clearly presented here. As the chapter rightly points out, "Watershed moments in multidisciplinary fields... often occur when... researchers... adopt relevant research in other fields". This paper makes it easy for them to do so.


The Bubble Cursor Paper:

It is in our nature to point at objects we would like to select using our index finger, an operation which could be (and has been) modeled virtually using a point cursor. However, in many modern GUIs the limited visual space is cluttered with many small items which makes selecting a target using a point a difficult task. The central vision of this paper's authors is that nevertheless virtual pointing could perform better than physical pointing.

In order to achieve this goal they present the Bubble Cursor, a round area cursor which changes its size and shape in response to the targets around it. This is an interesting idea as it is basically the opposite (as also acknowledged by the authors) of growing a target when the pointing device reaches it, a technique which resembles the old interface on Macs used to choose an application to open (I now notice by the way that this has been done with).

However, it is not clear to me after reading this paper that the Bubble Cursor is an interface that I would like to use. Based on Fitts's Law, the authors define their main measure of goodness as the time it takes to successfully select a target. While this is a necessary and crucial measure of goodness, it seems to be insufficient in this case. As the suggested cursor visually grows and shrinks when nearing items on the screen it may produce a significant amount of visual noise and may distract users from the task at hand. This suspicion is strengthened by the authors' many mentions of visual distraction, difficulty to plan the right distance of moving in an unnatural pointing task and the learning curves the participants experienced when learning to use the cursor. A user interface may be time efficient and yet not get adopted in practice if users do not enjoy a seamless natural experience when using it. In a paper presenting a new interactive interaction technique I would have expected to see a comparative analysis of users' reactions and preferences in addition to the demonstration of efficiency. This would have proven that the basic assumptions behind Fitts' Law hold here as well and that there are no hidden costs to this interaction technique that are not described by the model at hand.


Cheng Lu - 8/31/2011 8:59:34

Reading Response_2011/8/31 Papers: Motor Behavior Models for Human-Computer Interaction, MacKenzie, Chapter 3 from HCI Models, Theories, and Frameworks. The Bubble Cursor: Enhancing Target Acquisistion by Dynamic Resizing of the Cursor’s Activation Area, Tovi Grossman and Ravin Balakrishnan, 2005, In Proceeding of CHI 2005, p.281-190

Motivated by the need to match the movement limits, capabilities, and potential of human with input devices and interaction techniques on computer system, the first paper instructively introduces the foundations, descriptions and functions of models and modeling, among which one of the predictive models, the Fitt’s Law, is of highly importance for predicting movement time in a pointing task. The second paper presents a new target acquisition method-bubble cursor, which can dynamically resize its activation area depending on the proximity of surrounding targets, and the following experiments demonstrate that the bubble cursor significantly reduces target acquisition times in both simple and complex multi-target environment.

Since the first paper derives from part of the book, it is not that “important” in terms of scientific breakthrough, because the information is quite instructive. However, from my perspective, it is important to me, a student who just entered the HCI field, because it informs me an essential foundation of HCI. I got to know that there exist lots of models of human movement relevant to human-computer interaction, and it is this models that simplify and clarify our research process in HCI field. As to the second paper, it is important due to its innovation in target acquisition technique. The idea is pretty intuitive, but there do have some other important issues to deal with, such as the morph technique and the design of experiments to demonstrate the feasibility of the bubble cursor. As we can see from the second paper, the author spends much of its time to constructed two scientific experiments, and it is the direct number of improvement that demonstrate the superiority of bubble cursor among other techniques.

The second paper offers us a new target acquisition technique, which is a fundamental task in graphical user interfaces. Most of the previous pointing method, such as drag-and-pop and object pointing, only improves the performance in situations where targets are fairly sparsely distributed across the display space. Based on area cursors, this paper present another new approach, bubble cursor, which can dynamically updating its size based on the proximity of surrounding targets, such that there is always exactly one target inside the hotspot. In addition, the bubble cursor morphs to encompass a target when the basic circular cursor cannot completely do so without intersecting a neighboring target. Furthermore, this paper demonstrates that the bubble cursor performance can be accurately modeled and predicted using Fitt’s Law. This newly method in effect increases the target’s size in motor space to the maximum possible extent, and according to Fitt’s Law, decreasing the movement time significantly.

Innovative and effective though, I do find some blind spot in the experiment sectors of the second paper. In the second experiment, the author employed all the targets as same size in circle and stable square activation zone for simplification in further investigation. But in our real world interfaces, this assumption is clearly not the case. Using only the setting environment to test whether this technique fits the Fitt’s Law is feasible, but testing between different pointing methods is not that convincing. This paper should include some real world testing to really demonstrate the superiority of this newly brought method, even not in quantitative way will be sufficient to illustrate it better.


Devdatta - 8/31/2011 9:03:20

The first reading discusses models for humans interacting with machines, in particular predictive (or mathematical) models and descriptive models. An example of each, viz. Fitts Law and Guiard's model are presented and case studies discussed. The second reading presents an innovative target acquisition method for pointing devices that increases the target activation area (or effective width) , via resizable bubbles, thus reducing the difficulty of the tasks.


The papers are important in at the least showing the ability of models (and particularly Fitts Law) to characterize the real world and guide design. This is important, in that, once an engineer knows a good model, better design can result (instead of working in the dark). The bubble cursor offers a new design for target acquisition in pointing tasks. Using a bubble to increase the activation area, makes the ID of a task much lower. Additionally, a resizable bubble makes the interaction still work well in a dense target area.

Target acquisition and two handed interactions are fundamental to all Human computer Interaction used by the general populace. From typing on their keyboard, to using their mouse to interacting with their mobile phones: all these daily tasks involve target acquisition and two handed interactions. The papers allow us to easily identify deficiencies of today's interaction techniques, and offer insights into a new design.

One concern I had with the bubble cursor paper was the purely made up scenarios in the experiments. A real world scenario (say Microsoft Word or a web browser) would have been more interesting and convincing. Additionally, the paper contains vague statements about how targets are getting more dense: an actual measurement of how dense targets are would have been good, and this measurement could have guided the experiment. The use of pseudo random distractions is hardly convincing to me: I suspect that users are attuned to the non-targets in the real world and are able to zero in on their intended target a lot more easily.