Neuron Sandbox is a browser-based tool that helps middle school students grasp basic principles of neural computation. It simulates a linear threshold unit applied to binary decision problems, which students solve by adjusting the unit's threshold and/or weights. Although Neuron Sandbox provides extensive visualization aids, solving these problems is challenging for students who have not yet been exposed to algebra. We collected survey, video, and worksheet data from 21 seventh grade students in two sections of an AI elective, taught by the same teacher, that used Neuron Sandbox. We present a scaffolding strategy that proved effective at guiding these students to achieve mastery of these problems. While the amount of scaffolding required was more than we originally anticipated, by the end of the exercise students understood the computation that linear threshold units perform and were able to generalize their understanding of the worksheet’s "solve for threshold" strategy to also solve for weights.

Today, children of all ages interact with speech recognition systems but are largely unaware of how they work. Teaching K-12 students to investigate how these systems employ phonological, syntactic, semantic, and cultural knowledge to resolve ambiguities in the audio signal can provide them a window on complex AI decision-making and also help them appreciate the richness and complexity of human language. We describe a browser-based tool for exploring the Google Web Speech API and a series of experiments students can engage in to measure what the service knows about language and the types of biases it exhibits. Middle school students taking an introductory AI elective were able to use the tool to explore Google’s knowledge of homophones and its ability to exploit context to disambiguate them. Older students could potentially conduct more comprehensive investigations, which we lay out here. This approach to investigating the power and limitations of speech technology through carefully designed experiments can also be applied to other AI application areas, such as face detection, object recognition, machine translation, or question answering.

This BOF is a networking opportunity for researchers, teachers, resource developers, and state leaders involved or interested in K-12 AI Education. This BOF will serve as an informal opportunity for those who participated in the AI4K12 January 2021 State of AI Education in Your State Workshop and quarterly check-in webinars to reconnect in person. We also want to provide opportunities for newcomers working on K-12 AI Education, either through outreach or their own research, to learn about and join the community. The goal for all attendees is to talk about their work, share successes, form collaborations, and discuss issues with building the framework of implementation. Most importantly, attendees will have opportunities to connect with others with similar interests and challenges, to identify best practices, share resources, and identify next steps to help advance their work. For those who are new to the community, we want to provide opportunities to learn about K-12 AI efforts that are underway in their state and connect with their state leaders. Attendees from states with no efforts underway can learn about resources to start a state team and begin planning. The AI4K12 Initiative and the state workshop were funded by National Science Foundation award# DRL-1846073.

CS education advocates have made substantial progress toward getting states to provide universal K-12 computing education. But as computers and networking continue to drive the third industrial revolution, artificial intelligence and robotics are driving the fourth. What should states be doing now about K-12 AI education? In 2021 AI4K12.org organized The State of AI Education in Your State Workshop and held a series of follow-up meetings for state education officials and other interested parties to help plan for the introduction of AI into their state's K-12 computing education standards. Some are already well along in the work. In this lightning talk we highlight progress in several states and refer attendees to resources where they can form new professional connections and begin contributing to this effort. This work is crucial to prepare students for the economic and social disruptions anticipated to result from the AI-powered fourth industrial revolution that is already under way.

Over the past year, our AI4GA team of university faculty and middle school teachers have co-designed a middle school AI curriculum. In this poster we share how we used co-design both as a tool for collaboratively developing engaging AI activities and as a mechanism for mutual professional development. We explain our co-design process, give examples of curriculum materials provided to teachers, and showcase several teacher-created activities. We believe this approach to curriculum development centers the lived experiences of teachers and leverages the knowledge and expertise of university researchers to create high quality and engaging AI learning experiences for K-12 students.

The AI4GA project is developing a nine-week elective course called Living and Working with Artificial Intelligence and piloting it in several Georgia middle schools. Since we aspire to educate all students about AI, the course addresses a wide range of student abilities, levels of academic preparedness, and prior computing experience, and leaves room for teachers to adapt the material to their own students' needs and interests. The course content is primarily focused on unplugged activities and online demonstration programs. We also provide small programming projects using AI tools as an option for teachers to incorporate. In this poster we describe lessons learned from initial pilot offerings by five teachers who taught 12 sections of the course totaling 299 students. We present evidence that middle school students can successfully engage with substantive technical content about Artificial Intelligence.

Word embeddings, which represent words as dense feature vectors, are widely used in natural language processing. In their seminal paper on word2vec, Mikolov and colleagues showed that a feature space created by training a word prediction network on a large text corpus will encode semantic information that supports analogy by vector arithmetic, e.g., "king" minus "man" plus "woman" equals "queen". To help novices appreciate this idea, people have sought effective graphical representations of word embeddings. We describe a new interactive tool for visually exploring word embeddings. Our tool allows users to define semantic dimensions by specifying opposed word pairs, e.g., gender is defined by pairs such as boy/girl and father/mother, and age by pairs such as father/son and mother/daughter. Words are plotted as points in a zoomable and rotatable 3D space, where the third ”residual” dimension encodes distance from the hyperplane defined by all the opposed word vectors with age and gender subtracted out. Our tool allows users to visualize vector analogies, drawing the vector from “king” to “man” and a parallel vector from “woman” to “king-man+woman”, which is closest to “queen”. Visually browsing the embedding space and experimenting with this tool can make word embeddings more intuitive. We include a series of experiments teachers can use to help K-12 students appreciate the strengths and limitations of this representation.

The ubiquity of AI in society means the time is ripe to consider what educated 21st century digital citizens should know about this subject. In May 2018, the Association for the Advancement of Artificial Intelligence (AAAI) and the Computer Science Teachers Association (CSTA) formed a joint working group to develop national guidelines for teaching AI to K-12 students. Inspired by CSTA's national standards for K-12 computing education, the AI for K-12 guidelines will define what students in each grade band should know about artificial intelligence, machine learning, and robotics. The AI for K-12 working group is also creating an online resource directory where teachers can find AI- related videos, demos, software, and activity descriptions they can incorporate into their lesson plans. This blue sky talk invites the AI research community to reflect on the big ideas in AI that every K-12 student should know, and how we should communicate with the public about advances in AI and their future impact on society. It is a call to action for more AI researchers to become AI educators, creating resources that help teachers and students understand our work.

Previous research in K-5 CS education has focused on improving students' engagement in programming using visual block-based environments like Scratch. However, little is known about how elementary school students' reason about programs. We define computational reasoning as the ability to read, write, trace and debug programs and predict program behavior. Recently, computing education researchers have become interested in exploring how elementary school students build their computational reasoning abilities. This poster presents results from a study which analyzed the role of explicit instruction in the form of 'laws of computation' in cultivating elementary school (4th and 5th graders) students' ability to reason about programs using Microsoft Kodu Game Lab. We used pretests to record students' default models of reasoning about programs and then used posttests to measure the effectiveness of intervention by noting students' reasoning responses on a similar program. Our findings indicate that by default students reason sequentially about program execution which can be incorrect in situations like parallel rule execution. We also found that the use of explicit instruction in the form of 'laws' is helpful for students to refine their understanding of program execution and to improve their reasoning ability.

As CS expands into more K-12 classrooms and children become familiar with computational thinking, advances in AI pose new challenges for CS educators. Children now enjoy conversing with AI-powered agents such as Alexa and Siri, while their parents worry about the imminent arrival of autonomous robots and self-driving cars. As AI technologies become more prominent in our lives, we need to consider what every child should know about AI. This BOF provides a timely opportunity to introduce CS educators and researchers to several AI for K-12 efforts, including available curricula, tools, and resources. Attendees will discuss how AI can best be incorporated into the K-12 CS curriculum, the tools/resources that will be needed to support students and teachers learning about AI, and how AI education might impact their own work. This BOF is complementary to the SIGCSE 2019 Special Session: AI for K-12 Guidelines Initiative that introduces the current draft of our 'Big Ideas in AI." Further information about the initiative and resources is available at http://ai4k12.org.

In May 2018, the Association for the Advancement of Artificial Intelligence (AAAI) and the Computer Science Teachers Association (CSTA) formed a joint working group to develop national guidelines for teaching K-12 students about artificial intelligence. Inspired by CSTA's national standards for K-12 computing education, the - AI for K-12 guidelines (ai4k12.org) will define what students in each grade band should know about artificial intelligence, machine learning, and robotics. The working group is also creating an online resource directory where teachers can find AI-related videos, demo software, and activity descriptions they can incorporate into their lesson plans. The goal of this session is to raise the SIGCSE community's awareness of the initiative, its deliverables, and outcomes, and to foster a community-wide conversation about AI education in K-12. This initiative parallels other recent initiatives in K-12 AI education and community-wide initiatives and discussions around CS For All and CS in K-12. This Special Session is aimed toward K-12 CS educators, researchers, and curriculum and tool designers.

Over the last decade, CS Education researchers have developed different curricula, resources, and strategies to foster computer science learning in K-12 education. However, there is a lack of research about how elementary school students develop the ability to reason about programs. Reasoning about programs consists of a student's ability to read, write, debug, trace, and predict program behavior. This paper presents results from a think-aloud study of fourth and fifth grade students learning to program in Kodu. The goal of this study was to track students' understanding of how Kodu interprets and executes rules of a program. To understand students' reasoning of program execution, we explicitly taught them the Laws of Kodu computation which govern the decision making and execution process of Kodu rules. We collected students' responses on pre- and post-assessments, and we conducted think-aloud interviews with students where students explained their answers to assessment questions. We found that explicitly teaching students how Kodu rules are interpreted significantly improved their ability to understand the execution of programs and to explain program behavior. The results of this study provide insight into how elementary school students reason about simple programs, and how this ability can be scaffolded.

In light of our field/s progress in making programming accessible to novices, we contemplate an even more ambitious goal: make AI accessible to all. The Cozmo robot by Anki is revolutionizing consumer and educational robotics through built-in computer vision and artificial intelligence algorithms. Calypso is a scaffolded robot programming environment for Cozmo inspired by Microsoft/s Kodu Game Lab. Calypso allows novices to program with advanced features such as visual recognition of objects and faces, simultaneous localization and mapping (SLAM), landmark-based navigation, and speech input. Like Kodu, Calypso emphasizes rule-based programming with high-level primitives such as "see", "hear", "move toward", and "grab", and it uses an Xbox game controller as its primary interface. User testing of Calypso has shown that children as young as eight can easily use it to program Cozmo.

We believe teaching elementary school students to reason about programs is as important as teaching them to write programs. To facilitate development of this skill in young children one must choose a developmentally appropriate domain. Microsoft's Kodu Game Lab is a pattern-matching rule-based language whose semantics is significantly different than Scratch or Python. We chose Kodu because one can write non-trivial programs in two to four lines, and analyzing these programs is within the abilities of a typical 8 year old. Reasoning about programs requires students to understand the structure of code. The approach we're advocating is analogous to sentence diagramming, where one starts with a sequence of words and develops a representation of their syntactic and semantic relationships. One can similarly analyze Kodu programs by characterizing rules and recognizing relationships between rules. In this poster we describe "couplets", an analysis technique that reveals the presence within a program of an important Kodu design pattern called Pursue and Consume. Using this technique leads to accurate predictions about program behavior, and uncovers bugs if the pattern is not fully realized. As part of a study of 40 third graders who were learning Kodu, we provided brief instruction in the couplets technique. We found that they were able to apply couplets to 3-4 line programs and answer prediction questions with a roughly 85% success rate. Our results demonstrate that elementary school children can learn to reason abstractly about programs if given the right mental tools.

Researchers and educators have designed curricula and resources for introductory programming environments such as Scratch, App Inventor, and Kodu to foster computational thinking in K-12. This paper is an empirical study of the effectiveness and usefulness of tiles and flashcards developed for Microsoft Kodu Game Lab to support students in learning how to program and develop games. In particular, we investigated the impact of physical manipulatives on 3rd -- 5th grade students' ability to understand, recognize, construct, and use game programming design patterns. We found that the students who used physical manipulatives performed well in rule construction, whereas the students who engaged more with the rule editor of the programming environment had better mental simulation of the rules and understanding of the concepts.

Reading, tracing, and explaining the behavior of code are strongly correlated with the ability to write code effectively. To investigate program understanding in young children, we introduced two groups of third graders to Microsoft's Kodu Game Lab; the second group was also given four semantic "Laws of Kodu" to better scaffold their reasoning and discourage some common misconceptions. Explicitly teaching semantics proved helpful with one type of misconception but not with others. During each session, students were asked to predict the behavior of short Kodu programs. We found different styles of student reasoning (analytical and analogical) that may correspond to distinct neo-Piagetian stages of development as described by Teague and Lister (2014). Kodu reasoning problems appear to be a promising tool for assessing computational thinking in young programmers.

Mental simulation is an important skill for program understanding and prediction of program behavior. Assessing students' ability to mentally simulate program execution can be challenging in graphical programming environments and on paper-based assessments. This poster presents the iterative design and refinement process for assessing students' ability to mentally simulate and predict code behavior using a novel introductory computational thinking curriculum for Microsoft's Kodu Game Lab. We present an analysis of question prompts and student responses from data collected from three rising 3rd - 6th graders where the curriculum was implemented. Analysis of student responses suggest that this type of question can be used to identify misconceptions and misinterpretation of instructions. Finally, we present recommendations for question prompt design to foster better student simulation of program execution.

This paper introduces reasoning about lawful behavior as an important computational thinking skill and provides examples from a novel introductory programming curriculum using Microsoft's Kodu Game Lab. We present an analysis of assessment data showing that rising 5th and 6th graders can understand the lawfulness of Kodu programs. We also discuss some misconceptions students may develop about Kodu, their causes, and potential remedies.

The Pascaline was the first working mechanical calculator, created in 1642 by the French polymath Blaise Pascal. Over the next two decades Pascal built 40 of these machines, of which nine survive today. Several good web resources describe the Pascaline, but to properly appreciate the sautoir, Pascal's kinetic energy solution to jam-free ripple carry, building a working replica is invaluable.

One essential goal in education is to improve understanding of how humans acquire knowledge and how students vary in their abilities to learn. Building an intelligent agent that models student learning would be a significant achievement in the learning sciences. SimStudent is a state-of-the-art intelligent agent that simulates a human's learning process. However, SimStudent has only been living in the world of graphical user interfaces. To construct a more human-like learning agent, we integrate SimStudent with a cognitive robot, Calliope5KP, to create a physical agent that is able to learn skill knowledge by interacting with users in the physical world. We demonstrate the integration in a tic-tac-toe game, and show that the SimStudent robot is able to learn reasonably well with 12 games.

We describe a three-stage model of computing instruction beginning with a simple, highly scaffolded programming environment (Kodu) and progressing to more challenging frameworks (Alice and Lego NXT-G). In moving between frameworks, students explore the similarities and differences in how concepts such as variables, conditionals, and looping are realized. This can potentially lead to a deeper understanding of programming, bringing students closer to true computational thinking. Some novel strategies for teaching with Kodu are outlined. Finally, we briefly report on our methodology and select preliminary results from a pilot study using this curriculum with students ages 10-17, including several with disabilities.

In this half-day tutorial, we will introduce the ARTSI "Robotics Roadshow-in-a-Box (RRIB)", a single point resource for those getting started in robotics outreach. The RRIB is a kit which contains robots, software and prepared materials for providing robotics workshops for middle and high school students that focuses on showing computer scientists as problem solvers and not just programmers through activities with a larger context. The RRIB fills a need for materials that are accessible to those who may have limited knowledge of robotics or limited experience in middle school outreach, whether that is undergraduate students or faculty researchers who might have limited outreach experience or preparation time. Laptop Required.

Robotics is widely recognized as an interdisciplinary mixture of engineering and computer science, but the latter component is not well represented at many undergraduate institutions. The sophisticated technologies that underlie perception, planning, and control mechanisms in modern robots need to be made accessible to more computer science undergraduates. Following the curriculum design principles of Wiggins and McTighe (Understanding by Design, 2nd Ed.), I present seven big ideas in robotics that can fit together in a one semester undergraduate course. Each is introduced with an essential question, such as "How do robots see the world?" The answers expose students to deep concepts in computer science in a context where they can be immediately demonstrated. Hands-on labs using the Tekkotsu open source software framework and robots costing under $1,000 facilitate mastery of these important ideas. Courses based on parts of an early version of this curriculum are being offered at Carnegie Mellon and several other universities.

Tekkotsu has grown from a specialized framework for development on the Sony Aibo to a general purpose robotics development environment with support for a variety of hardware, algorithms for autonomous operation, virtual simulation, and associated curriculum for undergraduate education. This paper describes the implementation of these features, provides examples of their use in research and education, and draws a comparison with other popular open-source robotics frameworks.

Hand-eye systems combine computer vision with kinematics and dynamics calculations to achieve dexterous manipulation. These versatile platforms for teaching robotics principles have not been widely used in undergraduate laboratories due to cost. We describe a new hand-eye system constructed from Robotis Dynamixel servos, a USB interface module, and a webcam, that can be built for under $500 and run by a PC using the Tekkotsu open source software framework. A suggested curriculum is outlined.

A successful collaboration between Spelman College and Carnegie Mellon University led to an NSF-funded Broadening Participation in Computing project to set up robotics education laboratories and introduce undergraduate instruction in cognitive robotics at three other Historically Black Colleges and Universities (HBCUs). We give a brief overview of cognitive robotics and the Tekkotsu software architecture, and describe our experiences teaching computer science students with no previous robotics exposure to program sophisticated mobile robots.

We present a Bayesian statistical theory of context learning in the rodent hippocampus. While context is often defined in an experimental setting in relation to specific background cues or task demands, we advance a single, more general notion of context that suffices for a variety of learning phenomena. Specifically, a context is defined as a statistically stationary distribution of experiences, and context learning is defined as the problem of how to form contexts out of groups of experiences that cluster together in time. The challenge of context learning is solving the model selection problem: How many contexts make up the rodent's world? Solving this problem requires balancing two opposing goals: minimize the variability of the distribution of experiences within a context and minimize the likelihood of transitioning between contexts. The theory provides an understanding of why hippocampal place cell remapping sometimes develops gradually over many days of experience and why even consistent landmark differences may need to be relearned after other environmental changes. The theory provides an explanation for progressive performance improvements in serial reversal learning, based on a clear dissociation between the incremental process of context learning and the relatively abrupt context selection process. The impact of partial reinforcement on reversal learning is also addressed. Finally, the theory explains why alternating sequence learning does not consistently result in unique context-dependent sequence representations in hippocampus.

Although the responses of dopamine neurons in the primate midbrain are well characterized as carrying a temporal difference (TD) error signal for reward prediction, existing theories do not offer a credible account of how the brain keeps track of past sensory events that may be relevant to predicting future reward. Empirically, these shortcomings of previous theories are particularly evident in their account of experiments in which animals were exposed to variation in the timing of events. The original theories mispredicted the results of such experiments due to their use of a representational device called a tapped delay line. Here we propose that a richer understanding of history representation and a better account of these experiments can be given by considering TD algorithms for a formal setting that incorporates two features not originally considered in theories of the dopaminergic response: partial observability (a distinction between the animal's sensory experience and the true underlying state of the world) and semi-Markov dynamics (an explicit account of variation in the intervals between events). The new theory situates the dopaminergic system in a richer functional and anatomical context, since it assumes (in accord with recent computational theories of cortex) that problems of partial observability and stimulus history are solved in sensory cortex using statistical modeling and inference and that the TD system predicts reward using the results of this inference rather than raw sensory data. It also accounts for a range of experimental data, including the experiments involving programmed temporal variability and other previously unmodeled dopaminergic response phenomena, which we suggest are related to subjective noise in animals' interval timing. Finally, it offers new experimental predictions and a rich theoretical framework for designing future experiments.

We propose a probabilistic, generative account of configural learning phenomena in classical conditioning. Configural learning experiments probe how animals discriminate and generalize between patterns of si- multaneously presented stimuli (such as tones and lights) that are dif- ferentially predictive of reinforcement. Previous models of these issues have been successful more on a phenomenological than an explanatory level: they reproduce experimental findings but, lacking formal founda- tions, provide scant basis for understanding why animals behave as they do. We present a theory that clarifies seemingly arbitrary aspects of pre- vious models while also capturing a broader set of data. Key patterns of data, e.g. concerning animals' readiness to distinguish patterns with varying degrees of overlap, are shown to follow from statistical inference.

We develop a framework based on Bayesian model averaging to explain how animals cope with uncertainty about contingencies in classical con- ditioning experiments. Traditional accounts of conditioning fit parame- ters within a fixed generative model of reinforcer delivery; uncertainty over the model structure is not considered. We apply the theory to ex- plain the puzzling relationship between second-order conditioning and conditioned inhibition, two similar conditioning regimes that nonethe- less result in strongly divergent behavioral outcomes. According to the theory, second-order conditioning results when limited experience leads animals to prefer a simpler world model that produces spurious corre- lations; conditioned inhibition results when a more complex model is justified by additional experience.

According to a series of influential models, dopamine (DA) neurons sig- nal reward prediction error using a temporal-difference (TD) algorithm. We address a problem not convincingly solved in these accounts: how to maintain a representation of cues that predict delayed consequences. Our new model uses a TD rule grounded in partially observable semi-Markov processes, a formalism that captures two largely neglected features of DA experiments: hidden state and temporal variability. Previous models pre- dicted rewards using a tapped delay line representation of sensory inputs; we replace this with a more active process of inference about the under- lying state of the world. The DA system can then learn to map these inferred states to reward predictions using TD. The new model can ex- plain previously vexing data on the responses of DA neurons in the face of temporal variability. By combining statistical model-based learning with a physiologically grounded TD theory, it also brings into contact with physiology some insights about behavior that had previously been confined to more abstract psychological models.

This article addresses the relationship between long-term reward predictions and slow-timescale neural activity in temporal difference (TD) models of the dopamine system. Such models attempt to explain how the activity of dopamine (DA) neurons relates to errors in the prediction of future rewards. Previous models have been mostly restricted to short-term predictions of rewards expected during a single, somewhat artificially defined trial. Also, the models focused exclusively on the phasic pause-and-burst activity of primate DA neurons; the neurons' slower, tonic background activity was assumed to be constant. This has led to difficulty in explaining the results of neurochemical experiments that measure indications of DA release on a slow timescale, results that seem at first glance inconsistent with a reward prediction model. In this article, we investigate a TD model of DA activity modified so as to enable it to make longer-term predictions about rewards expected far in the future. We show that these predictions manifest themselves as slow changes in the baseline error signal, which we associate with tonic DA activity. Using this model, we make new predictions about the behavior of the DA system in a number of experimental situations. Some of these predictions suggest new computational explanations for previously puzzling data, such as indications from microdialysis studies of elevated DA activity triggered by aversive events.

The Temporal Coding Hypothesis of Miller and colleagues [7] sug(cid:173) gests that animals integrate related temporal patterns of stimuli into single memory representations. We formalize this concept using quasi-Bayes estimation to update the parameters of a con(cid:173) strained hidden Markov model. This approach allows us to account for some surprising temporal effects in the second order condition(cid:173) ing experiments of Miller et al. [1 , 2, 3], which other models are unable to explain.

We suggest that the hippocampus plays two roles that allow rodents to solve the hidden-platform water maze: self-localization and route replay. When an animal explores an environment such as the water maze, the combination of place fields and correlational (Hebbian) long-term potentiation produces a weight matrix in the CA3 recurrent collaterals such that cells with overlapping place fields are more strongly interconnected than cells with nonoverlapping fields. When combined with global inhibition, this forms an attractor with coherent representations of position as stable states. When biased by local view information, this allows the animal to determine its position relative to the goal when it returns to the environment. We call this self-localization. When an animal traces specific routes within an environment, the weights in the CA3 recurrent collaterals become asymmetric. We show that this stores these routes in the recurrent collaterals. When primed with noise in the absence of sensory input, a coherent representation of position still forms in the CA3 population, but then that representation drifts, retracing a route. We show that these two mechanisms can coexist and form a basis for memory consolidation, explaining the anterograde and limited retrograde amnesia seen following hippocampal lesions.

O'Keefe (1991) has proposed that spatial information in rats might be represented as phasors: phase and amplitude of a sine wave encoding angle and distance to a landmark. We describe computer simulations showing that operations on phasors can be efficiently realized by arrays of spiking neurons that recode the temporal dimension of the sine wave spatially. Some cells in motor and parietal cortex exhibit response properties compatible with this proposal.

The authors present PARSEC-a system for generating connectionist parsing networks from example parses. PARSEC is not based on formal grammar systems and has been geared towards spoken language tasks. PARSEC networks exhibit three strengths important for application to speech processing: they learn to parse, and generalize well compared to hand-coded grammars; they tolerate several types of noise; and they can learn to use multimodal input. The authors also present the PARSEC architecture, its training algorithms, and performance analyses along several dimensions that demonstrate PARSEC's features. They compare PARSEC's performance to that of traditional grammar-based parsing systems.>

We describe a parallel mapping matrix that performs several types of sequence manipulations that are the building blocks of well-known phonological processes. Our results indicate that human phonological behavior can by modeled by a highly constrained connectionist architecture, one that uses purely feedforward circuitry and imposes tight limits on depth of derivations.