Please join us for the NLP Seminar on Monday, November 13, at 4:00pm in 202 South Hall. All are welcome!
Speaker: He He (Stanford)
Title: Learning agents that interact with humans
The future of virtual assistants, self-driving cars, and smart homes require intelligent agents that work intimately with users. Instead of passively following orders given by users, an interactive agent must actively collaborate with people through communication, coordination, and user-adaptation. In this talk, I will present our recent work towards building agents that interact with humans. First, we propose a symmetric collaborative dialogue setting in which two agents, each with some private knowledge, must communicate in natural language to achieve a common goal. We present a human-human dialogue dataset that poses new challenges to existing models, and propose a neural model with dynamic knowledge graph embedding. Second, we study the user-adaptation problem in quizbowl – a competitive, incremental question-answering game. We show that explicitly modeling of different human behavior leads to more effective policies that exploits sub-optimal players. I will conclude by discussing opportunities and open questions in learning interactive agents.
Please join us for the NLP Seminar on Monday, October 30, at 4:00pm in 202 South Hall. All are welcome!
Speaker: Christopher Potts (Stanford Linguistics)
Title: Enriching distributional linguistic representations with structured resources
One of the most powerful ideas in natural language processing is that we can represent words and phrases using dense vectors learned from co-occurrence patterns in text. Such representations have proven themselves in many settings, and one might even argue that they make good on a common intuition among linguists: that words tend to be incredibly complex and related to each other in all sorts of subtle ways. However, co-occurrence patterns alone tend to yield only a blurry picture of the rich relationships that exist between concepts, which raises the question of how best to incorporate additional information from more structured resources. This talk will explore methods for achieving this synthesis, with special emphasis on the retrofitting method pioneered by Faruqui et al. (2015), in which existing representations are updated based on their position in a knowledge graph. I’ll describe and motivate a generalization of Faruqui et al.’s framework that explicitly models graph relations as functions (Lengerich et al. 2017), and I’ll discuss some potential pitfalls of retrofitting (Cases et al. 2017). My overall goal is to stimulate discussion about how to obtain semantically nuanced distributed representations that are useful in diverse tasks.
( Slides )
Cases, Ignacio; Minh-Thang Luong; and Christopher Potts. 2017. On the effective use of pretraining for natural language inference. Ms., Stanford University. arxiv.org/abs/1710.02076
Faruqui, Manaal; Jesse Dodge; Sujay K. Jauhar; Chris Dyer; Eduard Hovy; and Noah A. Smith. 2015. Retrofitting word vectors to semantic lexicons. NAACL. www.aclweb.org/anthology/N15-1184
Lengerich, Benjamin J.; Andrew L. Maas; and Christopher Potts. 2017. Retrofitting distributional embeddings to knowledge graphs with functional relations. Ms., Carnegie Mellon University, Stanford University, and Roam Analytics. arxiv.org/abs/1708.00112
Please join us for the next NLP Seminar on Monday, October 9, at 4:00pm in 202 South Hall.
Speaker: Siva Reddy (Stanford)
Title: Linguists-defined vs. Machine-induced Natural Language Structures for Executable Semantic Parsing
Querying a database to retrieve an answer, telling a robot to perform an action, or teaching a computer to play a game are tasks requiring communication with machines in a language interpretable by them. Here we consider the task of converting human languages to a knowledge-base (KB) language for question-answering. While human languages have latent structures, machine interpretable languages have explicit formal structures. The computational linguistics community has created several treebanks to understand the formal structures of human languages, e.g., universal dependencies. But are these useful for deriving machine interpretable formal structures?
In the first part of the talk, I will discuss how to convert universal dependencies in multiple languages to both general-purpose and kb-executable logical forms. In the second part, I will present a neural model on how to induce task-specific natural language structures. I will discuss the similarities and differences between linguists-defined and machine-induced structures, and pros and cons of each.
Siva Reddy is a postdoc at the Stanford NLP group working with Chris Manning. His research focuses on finding fundamental representations of language, mostly interpretable, which are useful for NLP applications, especially machine understanding. In this direction, he is currently exploring whether linguistic representations are necessary or all we need is end-to-end learning. His postdoc is partly funded by a Facebook AI Research grant. Prior to the postdoc, he was a Google PhD Fellow at the University of Edinburgh under the supervision of Mirella Lapata and Mark Steedman. He worked with Google Parsing team as an intern during his PhD, and as a full-time employee for Adam Kilgarriff’s Sketch Engine before his PhD. His team won the first place in SemEval 2011 Compositionality Detection task and a best paper at IJCNLP 2011. Apart from language, he loves nature and badminton.
Please join us for our first NLP Seminar of the Fall semester on Monday, September 25, at 4:00pm in 202 South Hall.
Speaker: David Smith (Northeastern University)
Title: Modeling Text Dependencies: Information Cascades, Translations, and Multi-Input Encoders
Dependencies among texts arise when speakers and writers copy manuscripts, cite the scholarly literature, speak from talking points, repost content on social networking platforms, or in other ways transform earlier texts. While in some cases these dependencies are observable—e.g., by citations or other links—we often need to infer them from the text alone. In our Viral Texts project, for example, we have built models of reprinting for noisily-OCR’d nineteenth-century newspapers to trace the flow of news, literature, jokes, and anecdotes throughout the United States. Our Oceanic Exchanges project is now extending that work to information propagation across language boundaries. Other projects in our group involve inferring and exploiting text dependencies to model the writing of legislation, the impact of scientific press releases, and changes in the syntax of language.
In this talk, I will discuss methods both for inferring these dependency structures and for exploiting them to improve other tasks. First, I will describe a new directed spanning tree model of information cascades and a new contrastive training procedure that exploits partial temporal ordering in lieu of labeled link data. This model outperforms previous approaches to network inference on blog datasets and, unlike those approaches, can evaluate individual links and cascades. Then, I will describe methods for extracting parallel passages from large multilingual, but not parallel, corpora by performing efficient search in the continuous document-topic simplex of a polylingual topic model. These extracted bilingual passages are sufficient to train translation systems with greater accuracy than some standard, smaller clean datasets. Finally, I will describe methods for automatically detecting multiple transcriptions of the same passage in a large corpus of noisy OCR and for exploiting these multiple witnesses to correct noisy text. These multi-input encoders provide an efficient and effective approximation to the intractable multi-sequence alignment approach to collation and allow us to produce transcripts with more than 75% reductions in error.
The NLP Seminar is back for Fall 2017! We will slightly change our meeting to Mondays from 4:00-5:00pm, in almost the same location, room 210 South Hall. We’ll be meeting approximately once a month this semester.
Here is the speaker for this semester:
Sep 25: David Smith, Northeastern U
Oct 9: Siva Reddy, Stanford U
Oct 30: Christopher Potts, Stanford U
Nov 13: He He: Stanford U
Amber Boydstun, UC Davis: postponed to Spring 2018
For up to the minute notifications, join the email list (UC Berkeley community only).
Please join us for our final NLP Seminar of the spring semester on Monday, May 1, at 3:30pm in 202 South Hall.
Speaker: Pramod Viswanath, University of Illinois
Title: Geometries of Word Embeddings
Real-valued word vectors have transformed NLP applications; popular examples are word2vec and GloVe, recognized for their ability to capture linguistic regularities via simple geometrical operations. In this talk, we demonstrate further striking geometrical properties of the word vectors. First we show that a very simple, and yet counter-intuitive, post-processing technique, which makes the vectors “more isotropic”, renders off-the-shelf vectors even stronger. Second, we show that a sentence containing a target word is well represented by a low rank subspace; subspaces associated with a particular sense of the target word tend to intersect over a line (one-dimensional subspace). We harness this Grassmannian geometry to disambiguate (in an unsupervised way) multiple senses of words, specifically so on the most promiscuously polysemous of all words: prepositions. A surprising finding is that rare senses, including idiomatic/sarcastic/metaphorical usages, are efficiently captured. Our algorithms are all unsupervised and rely on no linguistic resources; we validate them by presenting new state-of-the-art results on a variety of multilingual benchmark datasets.
Please join us for the NLP Seminar Monday, April 24 at 3:30pm in 202 South Hall.
Speaker: Marta Recasens (Google)
There’s Life Beyond Coreference
I’ll give a bird’s eye view of the coreference resolution task, discussing why after more than two decades of research on this task, state-of-the-art systems are still far from performing satisfactorily for real applications. Then, I’ll focus on the long tail of the problem, exemplifying how to cheaply learn common sense of the kind required by the Winograd Schema Challenge, and I’ll finish by undermining the traditional definition of the task, whose attempt at simplifying the problem may be making it even harder.
Please join us for the NLP Seminar Monday, April 10 at 3:30pm in 202 South Hall.
Speaker: Danqi Chen (Stanford)
Title: Towards the Machine Comprehension of Text
Enabling a computer to understand a document so that it can answer comprehension questions is a central, yet unsolved goal of NLP. The task of reading comprehension (i.e., question answering over unstructured text) has received vast attention recently, and a lot of progress has been made thanks to the creation of large-scale datasets and development of attention-based neural networks.
In this talk, I’ll first present how we advance this line of research. I’ll show how simple models can achieve (nearly) state-of-the-art performance on recent benchmarks, including the CNN/Daily Mail datasets and the Stanford Question Answering Dataset. I’ll focus on explaining the logical structure behind these neural architectures and discussing advantages as well as limits of current approaches.
Lastly I’ll talk about how we leverage existing machine comprehension systems and enable them to answer open-domain questions using full Wikipedia. We demonstrate the promise of our system, as well as set up new benchmarks by evaluating on multiple existing QA datasets.
Danqi Chen is a Ph.D. candidate in Computer Science at Stanford University, advised by Prof. Christopher Manning. Her main research interests lie in deep learning for natural language processing and understanding, and she is particularly interested in the intersection between text understanding and knowledge reasoning. She has been working on machine comprehension, question answering, knowledge base population and dependency parsing. She is a recipient of a Facebook fellowship and a Microsoft Research Women’s Fellowship and an outstanding paper award at ACL’16. Prior to Stanford, she received her B.S. from Tsinghua University in 2012.
Please join us for the NLP Seminar Monday, Mar 6 at 4:00pm in 205 South Hall.
Speaker: Joel Tetreault, Grammarly
Title: Analyzing Formality in Online Communication
Full natural language understanding requires comprehending not only the content or meaning of a piece of text or speech, but also the stylistic way in which it is conveyed. To enable real advancements in dialog systems, information extraction, and human-computer interaction, computers need to understand the entirety of what humans say, both the literal and the non-literal. This talk presents an in-depth investigation of one particular stylistic aspect, formality. First, we provide an analysis of humans’ subjective perceptions of formality in four different genres of online communication. We highlight areas of high and low agreement and extract patterns that consistently differentiate formal from informal text. Next, we develop a statistical model for predicting formality at the sentence level, using rich NLP and deep learning features, and then evaluate the model’s performance against human judgments across genres. Finally, we apply our model to analyze language use in online debate forums. Our results provide new evidence in support of theories of linguistic coordination, underlining the importance of formality for language generation systems.
This work was done with Ellie Pavlick (UPenn) during her summer internship at Yahoo Labs.
Please join us for the NLP Seminar on Monday 2/27 at 3:30pm in 202 South Hall. All are welcome!
Speaker: Jayant Krishnamurthy (Allen Institute for AI)
Title: Semantic Parsing to Probabilistic Programs for Situated Question Answering
Situated question answering is the problem of answering questions about an environment such as an image or diagram. This problem is challenging because it requires jointly interpreting a question and an environment using background knowledge to select the correct answer. We present Parsing to Probabilistic Programs, a novel situated question answering model that can use background knowledge and global features of the question/environment interpretation while retaining efficient approximate inference. Our key insight is to treat semantic parses as probabilistic programs that execute nondeterministically and whose possible executions represent environmental uncertainty. We evaluate our approach on a new, publicly-released data set of 5000 science diagram questions, outperforming several competitive classical and neural baselines.