Natural language programming has proven to be a very challenging task. We present a novel idea which suggests using crowdsourcing to do natural language programming. Our approach asks non-expert workers to provide input/output examples for a task defined in natural language form. We then use a Programming by Example system to induce the intended program from the input/output examples. Our early results are promising, encouraging further research in this area.

To ensure quality results from unreliable crowdsourced workers, task designers often construct complex workflows and aggregate worker responses from redundant runs. Frequently, they create several alternative workflows to accomplish the task, and choose a single workflow to deploy (perhaps the one that achieves the best performance during early experiments). However, this seemingly natural design paradigm does not achieve the full potential of crowdsourcing. In particular, using a single workflow (even the best) to accomplish a task is suboptimal. We show that alternative workflows can compose synergistically to yield a much higher quality output. We formalize the insight with a novel probabilistic graphical model, design and implement AgentHunt, a POMDP-based controller that dynamically switches between these workflows to achieve higher returns on investment, and design offline and online methods for learning model parameters. Live experiments on Amazon Mechanical Turk demonstrate the superiority of AgentHunt for the practical task of generating NLP training data, yielding up to 50% error reduction and greater net utility compared to previous methods.

People complete tasks more quickly when they have concrete plans, especially for open-ended, creative tasks. However, people often fail to create such action plans. (How) can systems provide people with these concrete steps automatically? To scalably provide personalized action plans, this paper introduces and evaluates crowdsourcing and peer approaches for creating plans, and NLP techniques for reusing them. We evaluated the effects of action plans on different types of tasks. A between-subjects experiment found that people who received crowd-created plans completed more tasks than people asked to self-create plans and than a control group without action plans. We found that crowd-created action plans are especially effective for lingering and high-level tasks. A second experiment found that peer-provided plans led to more completed tasks than no plans. A third experiment found that participants who received reused action plans also completed more tasks than a control group without action plans. We have incorporated these principles into TaskGenies: a crowd-powered task management system.

Websites that encourage consumers to research, rate, and review products online have become an increasingly important factor in purchase decisions. This increased importance has been accompanied by a growth in deceptive opinion spam - fraudulent reviews written with the intent to sound authentic and mislead consumers. In this study, we pool deceptive reviews solicited through crowdsourcing with actual reviews obtained from product review websites. We then explore several human- and machine-based assessment methods to spot deceptive opinion spam in our pooled review set. We find that the combination of human-based assessment methods with easily-obtained statistical information generated from the review text outperforms detection methods using human assessors alone.

Design tasks are notoriously difficult, because success is defined by the perception of the target audience, whose feedback is usually not available during design stages. Commonly, design is performed by professionals who have specific domain knowledge (i.e., an intuitive understanding of the implicit requirements of the task) and do not need the feedback of the perception of the viewers during the process. In this paper, we present a novel design methodology for creating minimal sketches of objects that uses an iterative optimization scheme. We define minimality for a sketch via the minimal number of straight line segments required for correct recognition by 75% of naiive viewers. Crowd-sourcing techniques allow us to directly include the perception of the audience in the design process. By joining designers and crowds, we are able to create a human computation system that can efficiently optimize sketches without requiring high levels of domain knowledge (i.e., design skills) from any worker.

Crowdsourcing utilizes human ability by distributing tasks to a large number of workers. It is especially suitable for solving data clustering problems because it provides a way to obtain a similarity measure between objects based on manual annotations, which capture the human perception of similarity among objects.This is in contrast to most clustering algorithms that face the challenge of finding an appropriate similarity measure for the given dataset. Several algorithms have been developed for crowdclustering that combine partial clustering results, each obtained by annotations provided by a different worker, into a single data partition. However, existing crowd-clustering approaches require a large number of annotations, due to the noisy nature of human annotations, leading to a high computational cost in addition to the large cost associated with annotation. We address this problem by developing a novel approach for crowclustering that exploits the technique of matrix completion. Instead of using all the annotations, the proposed algorithm constructs a partially observed similarity matrix based on a subset of pairwise annotation labels that are agreed upon by most annotators. It then deploys the matrix completion algorithm to complete the similarity matrix and obtains the final data partition by applying a spectral clustering algorithm to the completed similarity matrix. We show, both theoretically and empirically, that the proposed approach needs only a small number of manual annotations to obtain an accurate data partition. In effect, we highlight the trade-off between a large number of noisy crowdsourced labels and a small number of high quality labels.

A large number of images with ground truth object bounding boxes are critical for learning object detectors, which is a fundamental task in compute vision. In this paper, we study strategies to crowd-source bounding box annotations. The core challenge of building such a system is to effectively control the data quality with minimal cost. Our key observation is that drawing a bounding box is significantly more difficult and time consuming than giving answers to multiple choice questions. Thus quality control through additional verification tasks is more cost effective than consensus based algorithms. In particular, we present a system that consists of three simple sub-tasks --- a drawing task, a quality verification task and a coverage verification task. Experimental results demonstrate that our system is scalable, accurate, and cost-effective.

With the advent of the increasingly participatory Internet and the growing power of the crowd, “Serious Games” have proven to be a fertile approach for gathering task-specific natural language data at very low cost. In this paper we outline a game we call Doodling, based on the sketch-and-convey metaphor used in the popular board game Pictionary(R), with the goal of generating useful natural language data. We explore whether such a paradigm can be successfully extended for conveying more complex syntactic and semantic constructs than the words or short phrases typically used in the board game. Through a series of user experiments, we show that this is indeed the case, and that valuable parallel language data may be produced as a byproduct. In addition, we explore extensions to this paradigm along two axes – going online (vs. face-to-face) and going cross-lingual. The results in each of the sets of experiments confirm the potential of Doodling game to generate data in large quantities and across languages, and thus provide a new means of developing data sets and technologies for resource-poor languages.

Online labor marketplaces offer the potential to automate a variety of tasks too difficult for computers, but present requesters with significant difficulties in obtaining accurate results. We share experiences from building MobileWorks, a crowd platform that departs from the marketplace model to provide robust, high-quality results. Three architectural contributions yield measurably improved accuracy on input tasks.  A dynamic work routing system identifies expertise in the crowd and ensures that all work posted into the system is completed with bounded completion times and at fair worker prices. A peer management system ensures that incorrect answers are prevented by experienced members of the crowd. Last, social interaction techniques give the best workers the ability and incentives to manage, teach & supervise other members of the crowd, as well as to clarify tasks. This process filters worker error and allows the crowd to collectively learn how to solve unfamiliar tasks. (extended abstract)

The most critical moves that determine the success of a manipulation task are performed within the close vicinities of the object prior to grasping, and the goal prior to the final placement. Memorizing these state-action sequences and reusing them can dramatically improve the task efficiency, whereas even the state-of-the-art planning algorithms may require significant amount of time and computational resources to generate a solution from scratch depending on the complexity and the constraints of the task. In this paper, we propose a hybrid approach that combines the reliability of the past experiences gained through demonstration and the flexibility of a generative motion planning algorithm, namely RRT*, to improve task execution efficiency. As a side benefit of reusing these final moves, we can dramatically reduce the number of nodes used by the generative planner, hence the planning time, by either early-terminating the planner when the generated plan reaches a "recalled state", or deliberately biasing it towards the memorized state-action sequences that are feasible at the moment. This complementary combination of the already available partial plans and the generated ones yield to fast, reliable, and repeatable solutions.