The Conference on Computational Natural Language Learning (CoNLL) features a shared task, in which participants train and test their learning systems on the same data sets. In 2017, one of two tasks was devoted to learning dependency parsers for a large number of languages, in a realworld setting without any gold-standard annotation on input. All test sets followed a unified annotation scheme, namely that of Universal Dependencies. In this paper, we define the task and evaluation methodology, describe data preparation, report and analyze the main results, and provide a brief categorization of the different approaches of the participating systems.

We present a series of experiments on automatically identifying the sense of implicit discourse relations, i.e. relations that are not marked with a discourse connective such as "but" or "because". We work with a corpus of implicit relations present in newspaper text and report results on a test set that is representative of the naturally occurring distribution of senses. We use several linguistically informed features, including polarity tags, Levin verb classes, length of verb phrases, modality, context, and lexical features. In addition, we revisit past approaches using lexical pairs from unannotated text as features, explain some of their shortcomings and propose modifications. Our best combination of features outperforms the baseline from data intensive approaches by 4% for comparison and 16% for contingency.

Discourse connectives are words or phrases such as once, since, and on the contrary that explicitly signal the presence of a discourse relation. There are two types of ambiguity that need to be resolved during discourse processing. First, a word can be ambiguous between discourse or non-discourse usage. For example, once can be either a temporal discourse connective or a simply a word meaning "formerly". Secondly, some connectives are ambiguous in terms of the relation they mark. For example since can serve as either a temporal or causal connective. We demonstrate that syntactic features improve performance in both disambiguation tasks. We report state-ofthe-art results for identifying discourse vs. non-discourse usage and human-level performance on sense disambiguation.

We introduce a novel method of generating synthetic question answering corpora by combining models of question generation and answer extraction, and by filtering the results to ensure roundtrip consistency. By pretraining on the resulting corpora we obtain significant improvements on SQuAD2 (Rajpurkar et al., 2018) and NQ (Kwiatkowski et al., 2019), establishing a new state-of-the-art on the latter. Our synthetic data generation models, for both question generation and answer extraction, can be fully reproduced by finetuning a publicly available BERT model (Devlin et al., 2018) on the extractive subsets of SQuAD2 and NQ. We also describe a more powerful variant that does full sequence-to-sequence pretraining for question generation, obtaining exact match and F1 at less than 0.1% and 0.4% from human performance on SQuAD2. Related WorkQuestion generation is a well-studied task in its own right (Heilman and Smith,