Programming by Demonstration (PbD) is an approach to programming robots by demonstrating the desired behavior. Speech is a natural, hands-free way to augment demonstrations with control commands that guide the PbD process. However, existing speech interfaces for PbD systems rely on ad-hoc, predefined command sets that are rigid and require user training. Instead, we aim to develop flexible speech interfaces to accommodate user variations and ambiguous utterances. To that end, we propose to use a situated semantic parser that jointly reasons about the user's speech and the robot's state to resolve ambiguities. In this paper, we describe this approach and compare its utility to a rigid speech command interface.

We describe a novel system for discovering and characterizing emerging events. We define event emergence to be a developing situation comprised of a series of sub-events. To detect sub-events from a very large, continuous textual input stream, we use two techniques: (1) frequency-based detection of sub-events that are potentially entailed by an emerging event; and (2) anomaly-based detection of other sub-events that are potentially indicative of an emerging event. Identifying emerging events from detected sub-events involves connecting sub-events to each other and to the relevant emerging events within the event models and estimating the likelihood of possible emerging events. Each sub-event can be part of a number of emerging events and supports various event models to varying degrees. We adopt a coherent and compact model that probabilistically identifies emerging events. The innovative aspect of our work is a well-defined framework where statistical Big Data techniques are informed by event semantics and inference techniques (and vice versa). Our work is strongly grounded in semantics and knowledge representation, which enables us to produce more reliable results than would otherwise be possible with a purely statistical approach.

Part-of-speech (POS) tagging is a fundamental component for performing natural language tasks such as parsing, information extraction, and question answering. When POS taggers are trained in one domain and applied in significantly different domains, their performance can degrade dramatically. We present a methodology for rapid adaptation of POS taggers to new domains. Our technique is unsupervised in that a manually annotated corpus for the new domain is not necessary. We use suffix information gathered from large amounts of raw text as well as orthographic information to increase the lexical coverage. We present an experiment in the Biological domain where our POS tagger achieves results comparable to POS taggers specifically trained to this domain.

Large-scale annotated corpora are a prerequisite to developing high-performance NLP systems. Such corpora are expensive to produce, limited in size, often demanding linguistic expertise. In this paper we use text rewriting as a means of increasing the amount of labeled data available for model training. Our method uses automatically extracted rewrite rules from comparable corpora and bitexts to generate multiple versions of sentences annotated with gold standard labels. We apply this idea to semantic role labeling and show that a model trained on rewritten data outperforms the state of the art on the CoNLL-2009 benchmark dataset.

In this paper, we describe a system that jointly extracts entities appearing in images and mentioned in their accompanying captions. As input, the entity linking program takes a segmented image together with its caption. It consists of a sequence of processing steps: part-of-speech tagging, dependency parsing, and coreference resolution that enables us to identify the entities as well as possible textual relations from the captions. The program uses the image regions labelled with a set of predefined categories and computes WordNet similarities between these labels and the entity names. Finally, the program links the entities it detected across the text and the images. We applied our system on the Segmented and Annotated IAPR TC-12 dataset that we enriched with entity annotations and we obtained a correct assignment rate of 55.48%

There has been growing interest in stochastic methods to natural language generation (NLG). While most NLG pipelines separate morphological generation and syntactic linearization, the two tasks are closely related. In this paper, we study joint morphological generation and linearization, making use of word order and inflections information for both tasks and reducing error propagation. Experiments show that the joint method significantly outperforms a strong pipelined baseline (by 1.1 BLEU points). It also achieves the best reported result on the Generation Challenge 2011 shared task.

We propose an unsupervised learning algorithm for automatically inferring the mappings between English nouns and corresponding video objects. Given a sequence of natural language instructions and an unaligned video recording, we simultaneously align each instruction to its corresponding video segment, and also align nouns in each instruction to their corresponding objects in video. While existing grounded language acquisition algorithms rely on pre-aligned supervised data (each sentence paired with corresponding image frame or video segment), our algorithm aims to automatically infer the alignment from the temporal structure of the video and parallel text instructions. We propose two generative models that are closely related to the HMM and IBM 1 word alignment models used in statistical machine translation. We evaluate our algorithm on videos of biological experiments performed in wetlabs, and demonstrate its capability of aligning video segments to text instructions and matching video objects to nouns in the absence of any direct supervision.

In this paper, we present a hybrid grammar formalism designed to learn structured models of natural iconic gesture performances that allow for compressed representation and robust recognition. We analyze a dataset of iconic gestures and show how the proposed Feature-based Stochastic Context-Free Grammar (FSCFG) can generalize over both structural and feature-based variations among different gesture performances.

We investigate weakly-supervised image parsing, i.e., assigning class labels to image regions by using image-level labels only. Existing studies pay main attention to the formulation of the weakly-supervised learning problem, i.e., how to propagate class labels from images to regions given an affinity graph of regions. Notably, however, the affinity graph of regions, which is generally constructed in relatively simpler settings in existing methods, is of crucial importance to the parsing performance due to the fact that the weakly-supervised parsing problem cannot be solved within a single image, and that the affinity graph enables label propagation among multiple images. In order to embed more semantics into the affinity graph, we propose novel criteria by exploiting the weak supervision information carefully, and develop two graphs: L1 semantic graph and k-NN semantic graph. Experimental results demonstrate that the proposed semantic graphs not only capture more semantic relevance, but also perform significantly better than conventional graphs in image parsing.

Online discussions are growing as a popular, effective and reliable source of information for users because of their liveliness, flexibility and up-to-date information. Online discussions are usually developed and advanced by groups of users with various backgrounds and intents. However because of their diversities in topics and issues discussed by the users, supervised methods are not able to accurately model such dynamic conditions. In this paper, we propose a novel unsupervised generative model to derive aspect-action pairs from online discussions. The proposed method simultaneously captures and models these two features with their relationships that exist in each thread. We assume that each user post is generated by a mixture of aspect and action topics. Therefore, we design a model that captures the latent factors that incorporates the aspect types and intended actions, which describe how users develop a topic in a discussion. In order to demonstrate the effectiveness of our approach, we empirically compare our model against the state of the art methods on large-scale discussion dataset, crawled from apple discussions with over 3.3 million user posts from 340k discussion threads.