CMUML Micro-Reader System for KBP 2016 Cold Start Slot Filling, Event Nugget Detection, and Event Argument Linking Bishan Yang, Ndapandula Nakashole, Bryan Kisiel, Emmanouil A. Platanios, Abulhair Saparov, Shashank Srivastava, Derry Wijaya, Tom Mitchell School of Computer Science Carnegie Mellon University, Pittsburgh, PA {bishan,bkisiel}@cs.cmu.edu Abstract In this paper, we present an overview of the CMUML’s Micro-Reader system for three TAC KBP tasks: Cold Start Slot Filling (SF); Event Nugget Detection and Corefer- ence (EN); and Event Argument Extraction and Linking (EAL). The Micro-Reader system is a result of the CMU NELL team’s research efforts on single-document understanding us- ing background knowledge. It is a general- purpose machine reading system that takes as input a text document and outputs span- level semantic annotations for document un- derstanding. There are several reading com- ponents in the Micro-Reader, each having a different reading capability. For the SF task, we aggregate the outputs of different read- ing components to propose KB assertions that can be used to answer the SF queries. For the EN and EAL tasks, we mainly utilize the event reading component of the Micro- Reader, which performs joint inference of en- tities and events within a document context. 1 Introduction In this paper, we describe the CMUML’s Micro- Reader system for three TAC KBP 2016 tasks: Cold Start Slot Filling (SF); Event Nugget Detection and Coreference (EN); and Event Argument Extraction and Linking (EAL). Our Micro-Reader system con- tains several reading components, each performing span-level or sentence-level semantic analysis of the document. The goal is to achieve a deeper under- standing of what each sentence means in the context of a document. The Micro-Reader system we de- veloped this year has a similar architecture to the Text Document Stanford CoreNLP tokenization, lemmatization sentence splitting part-of-speech tagging dependency parsing, NER, coreference Micro-Reader Noun phrase categorizer Verb relation analyzer Nominals analyzer Semantic parser Event Extractor Background Knowledge NELL KB, WordNet, FrameNet, … Query Relation Extractor Aggregation & Filtering KB assertions RegExp extractor Figure 1: Overview of the NELL Micro-Reader system for the TAC KBP 2016 tasks. one that was used in our 2015 submission, with the difference that it includes several newly-developed reading components: a LSTM-based relation extrac- tor, a joint entity and event extractor, and a high- coverage verb relation analyzer. Figure 1 shows the system architecture of our Micro-reader system for TAC KBP 2016. The newly-developed micro- readers are highlighted in red. In the following, we first briefly describe the micro-reader components. Then, we discuss the experiment setup and process- ing pipeline for each KBP task. 2 NELL Micro-Readers Before running our own micro-readers, we process each KBP document using the StanfordCoreNLP pipeline. 1 The tokenize, ssplit, pos, lemma, ner, 1 http://nlp.stanford.edu/software/ corenlp.shtml.