Conference papers (DIGI) Forfattere "Hardt, Daniel"
Viser 1-8 af i alt 8
-
Structure and ContentAnand, Pranav; Hardt, Daniel (Frederiksberg, 206)[Flere oplysninger][Færre oplysninger]
Resume: Sluicing is an elliptical process where the majority of a question can go unpronounced as long as there is a salient antecedent in previous discourse. This paper considers the task of antecedent selection: finding the correct antecedent for a given case of sluicing. We argue that both syntactic and discourse relationships are important in antecedent selection, and we construct linguistically sophisticated features that describe the relevant relationships. We also define features that describe the relation of the content of the antecedent and the sluice type. We develop a linear model which achieves accuracy of 72.4%, a substantial improvement over a strong manually constructed baseline. Feature analysis confirms that both syntactic and discourse features are important in antecedent selection. URI: http://hdl.handle.net/10398/9425 Filer i denne post: 1
Anand_Hardt.pdf (198.5Kb) -
Evidence of Systematic Differences in User PopulationsWulf, Julie; Hardt, Daniel (Atlanta, GA, 2014)[Flere oplysninger][Færre oplysninger]
Resume: Do user populations differ systematically in the way they express and rate sentiment? We use large collections of Danish and U.S. reviews to investigate this question, and we find evidence of important systematic differences: first, positive ratings are far more common in the U.S. data than in the Danish data. Second, Danish reviewers tend to under-rate their own positive reviews compared to U.S. reviewers. This has potentially far-reaching implications for the interpretation of user ratings, the use of which has exploded in recent years. URI: http://hdl.handle.net/10398/8959 Filer i denne post: 1
Wulff og Hardt.pdf (654.6Kb) -
Evidence for a Uniform AccountHardt, Daniel (Frederiksberg, 2017)[Flere oplysninger][Færre oplysninger]
Resume: Same is an anaphoric element that performs a comparison, which can either be external or internal to a sentence. Hardt and Mikkelsen (2015) show that same, unlike other anaphoric expressions, imposes a parallelism constraint, and they present three types of examples showing that same is infelicitous in the absence of parallelism. Hardt and Mikkelsen propose an account that applies uniformly to internal and external readings; however, the evidence they present largely targets external readings – they don’t offer empirical evidence that clearly supports the uniform approach. Furthermore, Barker (2007) argues that internal readings must be treated differently than external readings. In this paper, I show that the parallelism effects observed by Hardt and Mikkelsen in fact apply to internal readings as well. This provides support for a uniform treatment of internal and external readings of same. It also suggests that discourse relations, which typically apply to separate overt predications, also apply to the implicit predications that arise in distributional structures. URI: http://hdl.handle.net/10398/9574 Filer i denne post: 1
Hardt_2017.pdf (137.4Kb) -
Mikkelsen, Line; Hardt, Daniel; Ørsnes, Bjarne (Frederiksberg, 2011)[Flere oplysninger][Færre oplysninger]
Resume: Overt VP anaphors like do so, do it and do the same can host a following PP (Culicover & Jackendoff (2005:285–6), Huddleston & Pullum (2002:1533), Miller (2011:5–6), Sobin (2008:150, 155–157)): (1) The House is set to take up the final version of the funding bill tomorrow. The Senate will do the same on Thursday. [COCA] (2) You have jilted two previous fiances and I expect you would do the same to me. [COCA] Using (1) to fix terminology, the ANAPHOR is do the same, the ANTECEDENT is take up the final version of the funding bill, the ORPHAN is on Tuesday, and the CORRELATE is tomorrow. Examples like (2) are of particular interest because the correlate (two previous fiances) is inside the antecedent and, consequently, the orphan and the antecedent must interact to produce the interpretation of the clause containing the anaphor. In order to arrive at the interpretation ‘you would jilt me’, the me of the orphan must take the place of two previous fiances inside the antecedent VP. A superficially similar situation arises with remnants of ellipsis, including pseudogapping (3), sluicing (4), and fragment answers (5). In each case, the interpretation of the ellipsis clause combines part of the antecedent with all or part of the remnant. (3) I wouldn’t say that to my mother, but I would to you. (4) I know he gave the dresser away, but I don’t know to who. (5) Q: Who did he give the dresser to? A: To me. URI: http://hdl.handle.net/10398/8469 Filer i denne post: 1
mikkelsen_hardt_oersnes_2011.pdf (136.1Kb) -
Hardt, Daniel; Hovy, Dirk; Sotiris, Lamprinidis (, 2018)[Flere oplysninger][Færre oplysninger]
Resume: Newspapers need to attract readers with headlines, anticipating their readers’ preferences. These preferences rely on topical, structural, and lexical factors. We model each of these factors in a multi-task GRU network to predict headline popularity. We find that pre-trained word embeddings provide significant improvements over untrained embeddings, as do the combination of two auxiliary tasks, newssection prediction and part-of-speech tagging. However, we also find that performance is very similar to that of a simple Logistic Regression model over character n-grams. Feature analysis reveals structural patterns of headline popularity, including the use of forward-looking deictic expressions and second person pronouns. URI: http://hdl.handle.net/10398/9683 Filer i denne post: 1
Hardt_Hovy_Lamprinidis.pdf (1.322Mb) -
Hardt, Daniel; Rambow, Owen (Copenhagen, 2017)[Flere oplysninger][Færre oplysninger]
Resume: We analyze user viewing behavior on an online news site. We collect data from 64,000 news articles, and use text features to predict frequency of user views. We compare predictiveness of the headline and “teaser” (viewed before clicking) and the body (viewed after clicking). Both are predictive of clicking behavior, with the full article text being most predictive. URI: http://hdl.handle.net/10398/9580 Filer i denne post: 1
Hardt_Rambow_2017.pdf (90.26Kb) -
Rønning, Ola; Hardt, Daniel; Søgaard, Anders (, 2018)[Flere oplysninger][Færre oplysninger]
Resume: Sluice resolution in English is the problem of finding antecedents of wh-fronted ellipses. Previous work has relied on handcrafted features over syntax trees that scale poorly to other languages and domains; in particular, to dialogue, which is one of the most interesting applications of sluice resolution. Syntactic information is arguably important for sluice resolution, but we show that multi-task learning with partial parsing as auxiliary tasks effectively closes the gap and buys us an additional 9% error reduction over previous work. Since we are not directly relying on features from partial parsers, our system is more robust to domain shifts, giving a 26% error reduction on embedded sluices in dialogue. URI: http://hdl.handle.net/10398/9642 Filer i denne post: 1
N18-2038.pdf (168.6Kb) -
Hardt, Daniel; Asher, Nicholas; Hunter, Julie (Frederiksberg, 2013)[Flere oplysninger][Færre oplysninger]
Resume: This paper compares two views on the status of indices in syntactic and logical representations. On a structural view, indices are syntactic formants on a par with node labels and phrase bracketings, and are thus a part of the logical forms that are derived from syntactic representations. On the process view, an index is not a syntactic object at all, but rather, an indication of the output of a resolution process. In this paper we argue that a recent body of data provides a clear empirical basis for distinguishing between these two views of indices. We argue that cases of sloppy VP ellipsis pose insurmountable problems for the structural view of indices, while these problems do not arise for the process view. Furthermore, we show that this resolution process is constrained by the semantics of various discourse relations. URI: http://hdl.handle.net/10398/8846 Filer i denne post: 1
Hardt.pdf (152.4Kb)
Viser 1-8 af i alt 8