Summer NLP Projects
- When:
- Tuesday, August 28, 2012, 02:00 pm - 4:00 pm
- Where:
- 11th Floor Conf. Room (#1135)
- Speaker:
- 2012 Summer Interns
- Description:
1. Machine Translation with Synchronous Hyperedge Replacement Grammars (Karl Moritz Hermann, Bevan Jones, Jacob Andreas, Daniel Bauer)
We present an approach to semantics-based statistical machine translation that uses synchronous hyperedge replacement grammars. We use unsupervised string-graph alignment algorithms and two different methods for weighted rule extraction to learn grammars from semantically annotated corpora. We present algorithms and results for semantic parsing and generation, as well as for machine translation using the semantic representation as a pivot.
===
2. English Understanding: From Annotations to AMRs (Nathan Schneider)
Abstract Meaning Representations (AMRs) capture several kinds of semantic phenomena, including predicate-argument structures, named entities, coreference, relations between nominals, negation, and modality. This talk will consider the existing annotation schemes, corpora, and tools for such phenomena and examine their relevance to producing AMRs for English sentences. I will present a JSON representation integrating relevant automatic and/or gold-standard linguistic analyses, as well as a highly modular system for exploiting available annotations to produce AMRs.
===
3. Natural Language Generation with Abstract Meaning Representation (Yang Gao, Yaqin Yang)
Abstract Meaning Representation (AMR) is a graph encoding semantic phenomena such as predicate-argument structure, named entity and coreference. We present a generation system that converts AMR into English, based on handwritten rules and automatic rules extracted from several linguistic resources. The system works in two passes, by first translating AMR into a forest and then extracting k-best outputs based on various features.
===
4. An NLP Analysis of Several Historical Ciphers (Taylor Berg-Kirkpatrick, Megha Srivastava, Victoria Knight, Zhang Xuemei, Kevin
Knight)
We use NLP techniques to analyze or break several historical ciphers:
the German Army Enigma cipher, three masonic ciphers, the Kryptos cipher, and the Chinese writing system called Nushu. The task of decryption can often be formulated as a combinatorial optimization problem, or a difficult learning problem in a probabilistic model. In this talk we present the key difficulties behind each cipher, along with the range of methods we used to address these difficulties, and our results so far.
===
