Explore the vast range of titles available.

INTRODUCTION The Great Depression is generally recognized as the greatest economic calamity in United States history.1 One of the Great Depression's many causes was reckless financial speculation driven in part by financial fraud.2 In response to the crisis, Congress passed the 1934 Securities Exchange Act (\"the Exchange Act\"), which courts have long held creates a private right of action for plaintiffs who experience an economic loss due to reliance on a material misstatement surrounding the purchase or sale of a security.3 A prima facie claim for securities fraud under the Exchange Act requires a showing of scienter, defined as \"a mental state embracing intent to deceive, manipulate, or defraud. \"4 Securities fraud claims under the Exchange Act provide an important remedy for investors who suffer losses due to fraudulent misstatements related to securities, and they serve a central regulatory function of ensuring accurate disclosure of information in financial markets.5 However, they have also provided fertile ground for frivolous lawsuits that lack merit and are intended solely to coerce defendants into settling to avoid costly discovery.6 In response to a proliferation of frivolous claims,7 Congress enacted the Private Securities Litigation Reform Act (\"PSLRA\") in 1995.8 Among other changes, the PSLRA elevated the pleading standards for the scienter element of securities fraud claims, requiring plaintiffs to plead facts giving rise to a \"strong inference\" of scienter to survive a motion to dismiss.9 The need to plead specific facts around this state-of-mind requirement presents an interpretive puzzle when dealing with corporate defendants because corporations lack their own discrete minds and \"think\" only through their employees, officers, and agents. In the 1990s, Congress became increasingly concerned with a proliferation of frivolous securities fraud claims designed to harass defendants into settling simply to avoid costly and protracted litigation.27 - Lawmakers heard testimony suggesting that, between compiling documents and preparing and deposing key witnesses, many of whom may be high-level executives with important responsibilities, \"discovery costs account for roughly 80% of total litigation costs in securities fraud cases.28 Plaintiffs' attorneys would initiate so-called \"strike suits\" and then use familiar and well-worn \"fishing expedition\" discovery tactics to win coercive settlements even where no fraud had truly occurred.29 In 1995, in an effort to limit this trend, Congress enacted the PSLRA over a presidential veto, elevating the pleading standards for plaintiffs bringing securities fraud claims.30 The PSLRA requires plaintiffs in securities fraud actions to \"state with particularity facts giving rise to a strong inference that the defendant acted with the required state of mind. \"32 In creating this heightened degree of factual particularity, Congress cited three central aims: \"(1) to encourage the voluntary disclosure of information by corporate issuers; (2) to empower investors so that they-not their lawyers-exercise primary control over private securities litigation; and (3)to encourage plaintiffs' lawyers to pursue valid claims and defendants to fight abusive claims.

Pre-trained contextual representations have led to dramatic performance improvements on a range of downstream tasks. Such performance improvements have motivated researchers to quantify and understand the linguistic information encoded in these representations. In general, researchers quantify the amount of linguistic information through probing, an endeavor which consists of training a supervised model to predict a linguistic property directly from the contextual representations. Unfortunately, this definition of probing has been subject to extensive criticism in the literature, and has been observed to lead to paradoxical and counter-intuitive results. In the theoretical portion of this paper, we take the position that the goal of probing ought to be measuring the amount of inductive bias that the representations encode on a specific task. We further describe a Bayesian framework that operationalizes this goal and allows us to quantify the representations' inductive bias. In the empirical portion of the paper, we apply our framework to a variety of NLP tasks. Our results suggest that our proposed framework alleviates many previous problems found in probing. Moreover, we are able to offer concrete evidence that -- for some tasks -- fastText can offer a better inductive bias than BERT.