Modern statistical machine learning (SML) methods share a major limitation with the early approaches to AI: there is no scalable way to adapt them to new domains. Human learning solves this in part by leveraging a rich, shared, updateable world model. Such scalability requires modularity: updating part of the world model should not impact unrelated parts. We have argued that such modularity will require both correctability (so that errors can be corrected without introducing new errors) and interpretability (so that we can understand what components need correcting). To achieve this, one could attempt to adapt state of the art SML systems to be interpretable and correctable; or one could see how far the simplest possible interpretable, correctable learning methods can take us, and try to control the limitations of SML methods by applying them only where needed. Here we focus on the latter approach and we investigate two main ideas: Teacher Assisted Learning, which leverages crowd sourcing to learn language; and Factored Dialog Learning, which factors the process of application development into roles where the language competencies needed are isolated, enabling non-experts to quickly create new applications. We test these ideas in an Automated Personal Assistant (APA) setting, with two scenarios: that of detecting user intent from a user-APA dialog; and that of creating a class of event reminder applications, where a non-expert teacher can then create specific apps. For the intent detection task, we use a dataset of a thousand labeled utterances from user dialogs with Cortana, and we show that our approach matches state of the art SML methods, but in addition provides full transparency: the whole (editable) model can be summarized on one human-readable page. For the reminder app task, we ran small user studies to verify the efficacy of the approach.

We explore the idea of using a "possibilistic graphical model" as the basis for a world model that drives a dialog system. As a first step we have developed a system that uses text-based dialog to derive a model of the user's family relations. The system leverages its world model to infer relational triples, to learn to recover from upstream coreference resolution errors and ambiguities, and to learn context-dependent paraphrase models. We also explore some theoretical aspects of the underlying graphical model.

We present MCTest, a freely available set of stories and associated questions intended for research on the machine comprehension of text. Previous work on machine comprehension (e.g., semantic modeling) has made great strides, but primarily focuses either on limited-domain datasets, or on solving a more restricted goal (e.g., open-domain relation extraction). In contrast, MCTest requires machines to answer multiple-choice reading comprehension questions about fictional stories, directly tackling the high-level goal of open-domain machine comprehension. Reading comprehension can test advanced abilities such as causal reasoning and understanding the world, yet, by being multiple-choice, still provide a clear metric. By being fictional, the answer typically can be found only in the story itself. The stories and questions are also carefully limited to those a young child would understand, reducing the world knowledge that is required for the task. We present the scalable crowd-sourcing methods that allow us to cheaply construct a dataset of 500 stories and 2000 questions. By screening workers (with grammar tests) and stories (with grading), we have ensured that the data is the same quality as another set that we manually edited, but at one tenth the editing cost. By being open-domain, yet carefully restricted, we hope MCTest will serve to encourage research and provide a clear metric for advancement on the machine comprehension of text. 1 Reading Comprehension A major goal for NLP is for machines to be able to understand text as well as people. Several research disciplines are focused on this problem: for example, information extraction, relation extraction, semantic role labeling, and recognizing textual entailment. Yet these techniques are necessarily evaluated individually, rather than by how much they advance us towards the end goal. On the other hand, the goal of semantic parsing is the machine comprehension of text (MCT), yet its evaluation requires adherence to a specific knowledge representation, and it is currently unclear what the best representation is, for open-domain text. We believe that it is useful to directly tackle the top-level task of MCT. For this, we need a way to measure progress. One common method for evaluating someone’s understanding of text is by giving them a multiple-choice reading comprehension test. This has the advantage that it is objectively gradable (vs. essays) yet may test a range of abilities such as causal or counterfactual reasoning, inference among relations, or just basic understanding of the world in which the passage is set. Therefore, we propose a multiple-choice reading comprehension task as a way to evaluate progress on MCT. We have built a reading comprehension dataset containing 500 fictional stories, with 4 multiple choice questions per story. It was built using methods which can easily scale to at least 5000 stories, since the stories were created, and the curation was done, using crowd sourcing almost entirely, at a total of $4.00 per story. We plan to periodically update the dataset to ensure that methods are not overfitting to the existing data. The dataset is open-domain, yet restricted to concepts and words that a 7 year old is expected to understand. This task is still beyond the capability of today’s computers and algorithms.

Animacy detection is a problem whose solution has been shown to be beneficial for a number of syntactic and semantic tasks. We present a state-of-the-art system for this task which uses a number of simple classifiers with heterogeneous data sources in a voting scheme. We show how this framework can give us direct insight into the behavior of the system, allowing us to more easily diagnose sources of error.

Typically, one approaches a supervised machine learning problem by writing down an objective function and finding a hypothesis that minimizes it. This is equivalent to finding the Maximum A Posteriori (MAP) hypothesis for a Boltzmann distribution. However, MAP is not a robust statistic. We present an alternative approach by defining a median of the distribution, which we show is both more robust, and has good generalization guarantees. We present algorithms to approximate this median. One contribution of this work is an efficient method for approximating the Tukey median. The Tukey median, which is often used for data visualization and outlier detection, is a special case of the family of medians we define: however, computing it exactly is exponentially slow in the dimension. Our algorithm approximates such medians in polynomial time while making weaker assumptions than those required by previous work.

This paper studies the problem of sentence-level semantic coherence by answering SAT-style sentence completion questions. These questions test the ability of algorithms to distinguish sense from nonsense based on a variety of sentence-level phenomena. We tackle the problem with two approaches: methods that use local lexical information, such as the n-grams of a classical language model; and methods that evaluate global coherence, such as latent semantic analysis. We evaluate these methods on a suite of practice SAT questions, and on a recently released sentence completion task based on data taken from five Conan Doyle novels. We find that by fusing local and global information, we can exceed 50% on this task (chance baseline is 20%), and we suggest some avenues for further research.

Many learning algorithms generate complex models that are difficult for a human to interpret, debug, and extend. In this paper, we address this challenge by proposing a new learning paradigm called correctable learning, where the learning algorithm receives external feedback about which data examples are incorrectly learned. We define a set of metrics which measure the correctability of a learning algorithm. We then propose a simple and efficient correctable learning algorithm which learns local models for different regions of the data space. Given an incorrect example, our method samples data in the neighborhood of that example and learns a new, more correct local model over that region. Experiments over multiple classification and ranking datasets show that our correctable learning algorithm offers significant improvements over the state-of-the-art techniques.

In this paper, we describe a new, publicly available corpus intended to stimulate research into language modeling techniques which are sensitive to overall sentence coherence. The task uses the Scholastic Aptitude Test's sentence completion format. The test set consists of 1040 sentences, each of which is missing a content word. The goal is to select the correct replacement from amongst five alternates. In general, all of the options are syntactically valid, and reasonable with respect to local N-gram statistics. The set was generated by using an N-gram language model to generate a long list of likely words, given the immediate context. These options were then hand-groomed, to identify four decoys which are globally incoherent, yet syntactically correct. To ensure the right to public distribution, all the data is derived from out-of-copyright materials from Project Gutenberg. The test sentences were derived from five of Conan Doyle's Sherlock Holmes novels, and we provide a large set of Nineteenth and early Twentieth Century texts as training material.

We investigate the problem of learning to rank with document retrieval from the perspective of learning for multiple objective functions. We present solutions to two open problems in learning to rank: first, we show how multiple measures can be combined into a single graded measure that can be learned. This solves the problem of learning from a 'scorecard' of measures by making such scorecards comparable, and we show results where a standard web relevance measure (NDCG) is used for the top-tier measure, and a relevance measure derived from click data is used for the second-tier measure; the second-tier measure is shown to significantly improve while leaving the top-tier measure largely unchanged. Second, we note that the learning-to-rank problem can itself be viewed as changing as the ranking model learns: for example, early in learning, adjusting the rank of all documents can be advantageous, but later during training, it becomes more desirable to concentrate on correcting the top few documents for each query. We show how an analysis of these problems leads to an improved, iteration-dependent cost function that interpolates between a cost function that is more appropriate for early learning, with one that is more appropriate for late-stage learning. The approach results in a significant improvement in accuracy with the same size models. We investigate these ideas using LambdaMART, a state-of-the-art ranking algorithm.