At Cades Cove in the Smokies.

Scott Martin

Research Scientist, Facebook

Public encryption key


About Me

Right now, I'm working to more closely align the Facebook news feed with human values by improving the content choices our ranking algorithms make. Past projects have focused on using language technology to combat violence triggered by online interactions and building the core capabilities of the Portal's conversational assistant using conversational AI and NLP techniques (coreference resolution, dialog management, referring expression generation). Before joining Facebook, I did the research behind Ozlo's dialog management and semantic modeling capabilities. In the past, I have worked on various topics in computational linguistics and NLP during stints at Yahoo Labs (parsing and semantic modeling for search and web analytics) and at Nuance's NLU and AI lab (computational semantics, dialog management for intelligent conversational assistance). Before that, I was a presidential fellow at OSU while finishing my Ph.D. in linguistics there in 2013. My CV has details, and so do my profile pages at Facebook Research, Google Scholar, LinkedIn, GitHub, and ResearchGate.

I'm interested in computational, formal, and mathematical linguistics, especially where they touch on computer science, artificial intelligence, and mathematical logic. My computational work has focused on anaphora and coreference resolution, automatic paraphrase alignment, generation, task selection for dialog management, syntactic parsing, and semantic representation. My work in linguistic theory mostly focuses on developing an interface between syntax and semantics that can model discourse compositionally, especially projective meaning types like anaphora and supplements.

In addition to doing research, I'm also accomplished as a software engineer. I designed and built significant parts of the Portal voice assistant as well as the main dialog management module in Ozlo's intelligent assistant. I contributed to the core parsing components of the SkyPhrase SDK, and I designed and implemented systems for resolving anaphora, dialog task matching, and lambda conversion while at Nuance's NLU/AI lab. I'm one of the main authors of OpenCCG (an open-source parser and realizer for CCG), and I wrote the PEP Java Earley parser. I also built an NLP component in the initial release of Sermo's social network for physicians.

Sometimes I sneak away for a bike ride, usually on or around Mount Tamalpais.

Selected Publications & Talks

On semantics, syntax, and formal and theoretical linguistics

On anaphora and coreference resolution, paraphrasing, generation, and computational linguistics

Code & Data


The Multidomain Coreference (MuDoCo) dataset was released as an open-source project in conjunction with our LREC 2020 paper. It contains almost 8,500 authored human-machine dialogs annotated for coreference links.

Dataset on the Facebook Research GitHub
(Initial release with the paper)

This dataset is encoded in JSON, with named entity and reference types and link annotations. It is broken down by domain and additionally split into an 80%/10%/10% scheme for training, testing, and development data, respectively.


This corpus is an enhanced version of the Edinburgh paraphrase corpus, with both machine- and hand-corrected tokenization, hand-corrected alignments based on retokenization, parses from both the OpenCCG parser and the Stanford dependency parser. It also includes named entity annotations generated by the Stanford parser and Meteor alignments for use as a baseline.

Edinburgh++ Corpus
(03/22/2013 release)

The corpus is encoded in JSON format, but comes with a handy Python script that outputs just the alignments. The training and a test partitions are based on the partitioning scheme in my COLING 2012 paper.


The name PEP stands for PEP is an Earley Parser and is an example of direct left recursion. PEP is an implementation of Earley's chart-parsing algorithm in Java. It includes a thin command-line interface, but is intended to be used as a library. PEP is free software released under the LGPL.

PEP project
on GitHub
PEP source and binaries
Version 0.4
generated using my public key
API Documentation
generated by JavaDoc

The PEP GitHub project, and the tar bundle above, contains PEP's binaries, full source code, generated documentation, and an Ant build file. It also includes several sample grammars for testing and automated JUnit tests.

PEP can parse strings licensed by any CFG (including those that contain recursive rules). PEP's charts use backpointers so that if a grammar allows ambiguity, PEP keeps track of all of the possible parses in a set of traversable parse trees. Version 0.4 is generalized to allow rules with right-hand sides that include a mix of terminals and nonterminals.



Formal Foundations of Linguistic Theory
(Assistant to Carl Pollard.) Foundational course on the mathematical tools used in formal linguistics.
Syntax 1
(Assistant to Bob Levine.) Overview of syntactic theory and description based on HPSG.
  • Fall 2011


Language and Computers
Broad-based overview of topics in computational linguistics.
Language and Formal Reasoning
Truth-conditional meaning in natural language and its interaction with deductive reasoning.
Introduction to Language in the Humanities
Survey course in general linguistics.