{ "paper_id": "2021", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T07:24:34.707463Z" }, "title": "Variation in framing as a function of temporal reporting distance", "authors": [ { "first": "Levi", "middle": [], "last": "Remijnse", "suffix": "", "affiliation": { "laboratory": "", "institution": "Vrije Universiteit Amsterdam De", "location": { "addrLine": "Boelelaan 1105", "postCode": "1081 HV", "settlement": "Amsterdam", "country": "The Netherlands" } }, "email": "l.remijnse@vu.nl" }, { "first": "Marten", "middle": [], "last": "Postma", "suffix": "", "affiliation": { "laboratory": "", "institution": "Vrije Universiteit Amsterdam De", "location": { "addrLine": "Boelelaan 1105", "postCode": "1081 HV", "settlement": "Amsterdam", "country": "The Netherlands" } }, "email": "m.c.postma@vu.nl" }, { "first": "Piek", "middle": [], "last": "Vossen", "suffix": "", "affiliation": { "laboratory": "", "institution": "Vrije Universiteit Amsterdam De", "location": { "addrLine": "Boelelaan 1105", "postCode": "1081 HV", "settlement": "Amsterdam", "country": "The Netherlands" } }, "email": "piek.vossen@vu.nl" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "In this paper, we measure variation in framing as a function of foregrounding and backgrounding in a co-referential corpus with a range of temporal distance. In one type of experiment, frame-annotated corpora grouped under event types were contrasted, resulting in a ranking of frames with typicality rates. In contrasting between publication dates, a different ranking of frames emerged for documents that are close to or far from the event instance. In the second type of analysis, we trained a diagnostic classifier with frame occurrences in order to let it differentiate documents based on their temporal distance class (close to or far from the event instance). The classifier performs above chance and outperforms models with words.", "pdf_parse": { "paper_id": "2021", "_pdf_hash": "", "abstract": [ { "text": "In this paper, we measure variation in framing as a function of foregrounding and backgrounding in a co-referential corpus with a range of temporal distance. In one type of experiment, frame-annotated corpora grouped under event types were contrasted, resulting in a ranking of frames with typicality rates. In contrasting between publication dates, a different ranking of frames emerged for documents that are close to or far from the event instance. In the second type of analysis, we trained a diagnostic classifier with frame occurrences in order to let it differentiate documents based on their temporal distance class (close to or far from the event instance). The classifier performs above chance and outperforms models with words.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Abstract", "sec_num": null } ], "body_text": [ { "text": "To understand streams of news and blogs in terms of the ways in which events can be framed, we need to model how these streams develop over time in relation to the common ground that is created. The common ground between interlocutors plays an essential role in how they refer to real-world event instances. 1 Following pragmatic theory (Grice, 1975; Horn, 1998; Clark et al., 1977) , when this common ground is low, the speaker, in an attempt to be cooperative, needs to be as informative as possible, using detailed and marked descriptions of the main event instance. When the common ground is high, the speaker can optimally use less marked expressions and hence background the main event instance in order to foreground related events with a higher informative value (see also the grounding principles of Grimes (2015). The less marked expression then implicates prior knowledge of the event instance, which has become unnecessary to explicate. This is shown in the next two examples that report on instances of the same event type at different points in time (the-day-before versus aweek-ago). In example (1a), reference to the event instance is marked by using multiple indefinite expressions of different syntactic categories in reference to subevents: a shooting in which a man died. In example (1b), reference to the event instance is restricted to one definite expression last week's murder, which presupposes the event instance as shared knowledge and implicates its details. The rest of the text in the example focuses on other events.", "cite_spans": [ { "start": 308, "end": 309, "text": "1", "ref_id": null }, { "start": 337, "end": 350, "text": "(Grice, 1975;", "ref_id": "BIBREF13" }, { "start": 351, "end": 362, "text": "Horn, 1998;", "ref_id": "BIBREF18" }, { "start": 363, "end": 382, "text": "Clark et al., 1977)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "(1) a. One man died in a shooting early Thursday morning in southwest Houston. 2 b. One of the four suspects wanted in last week's murder of Keith Thompson was arrested Wednesday morning at a home in Springfield, according to the Jacksonville Sheriff's Office. 3 Given this theory about variation in referential expressions, we can expect that, from the onset of an unexpected real-world event instance (e.g., a shootout), the constantly developing narrative of related events (e.g. pursuits, arrests, trials) will enforce these mechanics of foregrounding and backgrounding based on growing mutual knowledge. In other words, the common ground determines the extent to which the speaker is able to background (i.e., use minimal expressions or implicatures) the main event in order to foreground related subjects.", "cite_spans": [ { "start": 253, "end": 262, "text": "Office. 3", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "Suppose we want to test these principles empirically by examining references in a large dataset, e.g., a referentially grounded corpus. This requires a large collection of documents all referencing single event instances, with a large spread of temporal distance between the publication dates and the event instance date. However, most of the available co-referential corpora hardly contain multiple reference texts for the same event instance, let alone with a strong range of publication dates .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "In this paper, we propose to overcome the data sparsity by merging data of event instances of the same event type to study foregrounding and backgrounding phenomena. We assume that it takes approximately the same amount of time for information, on for instance shooting events, to become common ground between members of a society. Furthermore, such a specific event type activates a coherent set of conceptual properties typically used in reference (Vossen et al., 2020; Morris and Murphy, 1990 ). Yet, this use of reference might depend on mutual knowledge. Based on the discussed pragmatic principles, we claim that both referential expressions and their meanings vary across documents with different temporal distances to event instances: over time, relevant information about the event instance is left implicit as a means to background reference to the event instance and foreground reference to novel information.", "cite_spans": [ { "start": 450, "end": 471, "text": "(Vossen et al., 2020;", "ref_id": "BIBREF36" }, { "start": 472, "end": 495, "text": "Morris and Murphy, 1990", "ref_id": "BIBREF23" } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "In order to find evidence for our claim, we use FrameNet as a proxy to characterize event semantics. Our prediction is that those frames typically associated with an event type, called typical frames, will also show a different foregrounding and backgrounding distribution as a function of the increased common ground. We expect that subevents of the event instance are foregrounded in texts with little temporal distance, whereas related disjoint events are expected to be foregrounded in texts with large temporal distance. This difference should be reflected by their frames.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "To test this hypothesis, we applied a method based on Grootendorst (2020) to learn frame typicality rates for event types from a large collection of news reports that were processed with an automatic frame-labeler (Swayamdipta et al., 2017) . Furthermore, we trained a Linear Support Vector Machine classifier to distinguish between referential texts with close temporal distance and further temporal distance on the basis of the typical frames evoked by the texts. We contrast this classifier against models trained on words. We provide evidence that frame distributions are learned by the classifier to perform the task, whereas this is lesser the case for word based models. Our analysis of the results shows that the typical frames evoked in texts with a short temporal distance are backgrounded in texts of larger temporal distance by means of implicature.", "cite_spans": [ { "start": 54, "end": 73, "text": "Grootendorst (2020)", "ref_id": "BIBREF15" }, { "start": 214, "end": 240, "text": "(Swayamdipta et al., 2017)", "ref_id": "BIBREF34" } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "The main contributions of our work are:", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "\u2022 We present HDD (Historical Distance Data), an extensive corpus of reference texts for event instances grouped under event types, with a large spread of temporal distance to the event instance;", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "\u2022 We derive a ranking of typical frames crossevent types;", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "\u2022 We show that frames are more informative than their predicates in training a Linear Support Vector to predict the temporal distance class given a document;", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "\u2022 We show that when contrasting frames for an event type between temporal distance classes, the top ranked frames reflect foregrounded topics.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "Our results will help future systems in detecting events in texts and their framing but also help the computational modeling of pragmatics and implicatures.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "This paper is structured as follows. We first describe relevant past work in Section 2. We then introduce our methodology in Section 3. Section 4 provides the results of our experiments, which we discuss in Section 5. We conclude in Section 6.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "In this section, we discuss previous work that has been done with respect to event foregrounding, (2.1 FrameNet (2.2), temporal distance (2.3) and event corpora (2.4).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Background", "sec_num": "2" }, { "text": "Different studies have focused on the recognition and characterization of foregrounded events. On the sentence level, foregrounded events show high probability of appearing in main clauses, being actively voiced and having a high transitivity (Kay and Aylett, 1996; Decker, 1985) . These observations are applied by Upadhyay et al. (2016) to identify the most significant event in a news article.", "cite_spans": [ { "start": 243, "end": 265, "text": "(Kay and Aylett, 1996;", "ref_id": "BIBREF20" }, { "start": 266, "end": 279, "text": "Decker, 1985)", "ref_id": "BIBREF7" }, { "start": 316, "end": 338, "text": "Upadhyay et al. (2016)", "ref_id": "BIBREF35" } ], "ref_spans": [], "eq_spans": [], "section": "Event foregrounding", "sec_num": "2.1" }, { "text": "On the discourse level, it has been observed that normalized frequencies of co-referential event mentions play an important role in detecting the central event of a document (Filatova and Hatzivassiloglou, 2004a,b) . According to Choubey et al. 2018, another crucial factor is the scope of the chain of co-referential mentions throughout the document. These mentions foreground subevents in reference to the central event. The discussed examples in Choubey et al. 2018show that backgrounded events scarcely occur throughout the document, supporting the reader in grounding the foregrounded central event in a commonly known prior event. In line with their proposal, both died and shooting in (1a) form a chain of foregrounded subevents that make reference to the central event of the document. In (1b), arrested is the foregrounded central event, but murder is a backgrounded event.", "cite_spans": [ { "start": 174, "end": 214, "text": "(Filatova and Hatzivassiloglou, 2004a,b)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Event foregrounding", "sec_num": "2.1" }, { "text": "In this paper, we propose that the mentions that foreground the central event instance also activate a coherent set of FrameNet frames typically used in reference to the event type. In analyzing HDD, we find that this set of typical frames is different for documents written long after the event instance, as an effect of backgrounding that event instance and foregrounding related disjoint events.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Event foregrounding", "sec_num": "2.1" }, { "text": "We use FrameNet as a proxy to characterize event semantics in this paper. 4 FrameNet is a lexicographic project anchored in the paradigm of frame semantics Fillmore and Baker, 2010; Baker et al., 2003) . Its lexical database consists of over 1200 semantic frames. Each frame is considered a schematic representation of a situation involving semantic roles, and is assumed to be evoked by a lexical unit, i.e., a lemma in one of its senses. Each frame exhibits an inventory of lexical units. Below, (1) is extended with FrameNet annotations.", "cite_spans": [ { "start": 74, "end": 75, "text": "4", "ref_id": null }, { "start": 156, "end": 181, "text": "Fillmore and Baker, 2010;", "ref_id": "BIBREF11" }, { "start": 182, "end": 201, "text": "Baker et al., 2003)", "ref_id": "BIBREF0" } ], "ref_spans": [], "eq_spans": [], "section": "Frames as implicatures", "sec_num": "2.2" }, { "text": "(2) a. With respect to inferential relations between frames, literature largely focuses on different types of frame-to-frame relations, i.e., asymmetric relations between two frames. The FrameNet database registers frame-to-frame relations between the frames to form a network. For example, the Precedes relation specifies a sequential order between two frames, e.g., ARREST shows a Precedes relation to ARRAIGNMENT (Ruppenhofer et al., 2010) . Thus, when ARRAIGNMENT is evoked in a document, we can infer ARREST as an implicature. Frames connected through Precedes relations form a coherent set in which any frame implicates the \"preceding\" frames. The output of our experiment can be used as input for FrameNet to form more of these cohesive sets of frames with temporal relations.", "cite_spans": [ { "start": 416, "end": 442, "text": "(Ruppenhofer et al., 2010)", "ref_id": "BIBREF30" } ], "ref_spans": [], "eq_spans": [], "section": "Frames as implicatures", "sec_num": "2.2" }, { "text": "The effect of the temporal distance between a reference text's publication date and the event date on variation in reference has been explored in a few studies. 5 Stali\u016bnait\u0117 et al. 2018focus on co-reference to entities in the New York Times Annotated Corpus (Sandhaus, 2008) , which contains articles spanning 20 years. They show that as a function of common knowledge, references to the same entity become definite, of shorter length, i.e., less marked, and with less use of appositives.", "cite_spans": [ { "start": 161, "end": 162, "text": "5", "ref_id": null }, { "start": 259, "end": 275, "text": "(Sandhaus, 2008)", "ref_id": "BIBREF31" } ], "ref_spans": [], "eq_spans": [], "section": "Temporal Distance", "sec_num": "2.3" }, { "text": "Cybulska and Vossen (2010) carried out a statistical analysis on a corpus of reference texts concerning the Srebrenica Massacre. The corpus consisted of 52 news articles (evenly distributed over two news journals) published within a time range of 10 days after the event, and 26 \"historical\" texts published years later. They created a word-based frequency distribution of references. They showed a strong discrepancy in type-token ratio between the two conditions of temporal distance: the subcorpus written close to the event shows a higher number of word types than the sub-corpus written years later. The authors conclude that difference in temporal distance correlates with variation in language use. Short temporal distance leads to more variation in descriptions, due to focus on sub-events, while longer distance leads to less variation in descriptions due to focus on the main event.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Temporal Distance", "sec_num": "2.3" }, { "text": "Our research aims to contribute to Cybulska and Vossen (2010) in the following ways. Our HDD is restricted to news articles under the assumption that variation in reference can also be observed within genres. Hence, the potential confounding variable of variation between text genres in their study is eliminated. Second, HDD covers reference texts of multiple event instances of a single event type. Third, we use FrameNet to measure variation in typically evoked frames on top of expressions. Finally, the dimension of temporal distance in our experiments ranges to 30 days after the event instances, instead of years.", "cite_spans": [ { "start": 35, "end": 61, "text": "Cybulska and Vossen (2010)", "ref_id": "BIBREF5" } ], "ref_spans": [], "eq_spans": [], "section": "Temporal Distance", "sec_num": "2.3" }, { "text": "In event co-reference research of the last decade, the corpus datasets show a small number of documents referencing events. Vossen et al. (2018) provide an overview of the nine governing text corpora (e.g., OntoNotes (Pradhan et al., 2007) , ECB (Bejan and Harabagiu, 2010), ACE2005 (Peng et al., 2016) ) and observed that their sum consists of less than four thousand documents. The number of mentions of events is small within documents (10 mentions per document on average) and only a subset of the corpora contains cross-document event co-reference. Also more recent attempts to manually create annotations for all sentences in articles did not cover a high number of documents (Cybulska and Vossen, 2014; Song et al., 2015; O'Gorman et al., 2016 ).", "cite_spans": [ { "start": 124, "end": 144, "text": "Vossen et al. (2018)", "ref_id": "BIBREF37" }, { "start": 217, "end": 239, "text": "(Pradhan et al., 2007)", "ref_id": "BIBREF28" }, { "start": 283, "end": 302, "text": "(Peng et al., 2016)", "ref_id": "BIBREF26" }, { "start": 696, "end": 709, "text": "Vossen, 2014;", "ref_id": "BIBREF6" }, { "start": 710, "end": 728, "text": "Song et al., 2015;", "ref_id": "BIBREF32" }, { "start": 729, "end": 750, "text": "O'Gorman et al., 2016", "ref_id": "BIBREF24" } ], "ref_spans": [], "eq_spans": [], "section": "Event corpora", "sec_num": "2.4" }, { "text": "Since we need a substantial amount of event reports of the same event type for our experiment, we used the Multilingual Wiki Extraction Platform (MWEP) (Vossen et al., 2020) to obtain a large corpus of referentially grounded news texts. MWEP follows the data-to-text method and takes event types as input to query Wikidata (Vrande\u010di\u0107 and Kr\u00f6tzsch, 2014) for event instances. For the obtained event instances, MWEP crawls the corresponding Wikipedia pages and their primary reference texts. These pages are processed by NLP systems, resulting in a corpus of multilayered linguistic annotation files.", "cite_spans": [ { "start": 152, "end": 173, "text": "(Vossen et al., 2020)", "ref_id": "BIBREF36" }, { "start": 323, "end": 353, "text": "(Vrande\u010di\u0107 and Kr\u00f6tzsch, 2014)", "ref_id": "BIBREF38" } ], "ref_spans": [], "eq_spans": [], "section": "Event corpora", "sec_num": "2.4" }, { "text": "In this section, we describe the methodology used for both the between-event type and within-event type experiments. 6 This includes the resources, (3.1), data processing (3.2), contrastive analysis (3.3, hypotheses (3.4) and evaluation (3.5).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Methodology", "sec_num": "3" }, { "text": "The model used to describe our data relies on three main concepts: event type, incident, and reference text. Let E be a set of event types, let I be a set of real-world event instances, and let R denote a registry of reference texts. Each real-world instance L i \u2208 I is an instance of one or more event types. Also, there can be reference texts that refer to a particular real-world instance L i . For example, the reference text Significance of Orlando gunman calling 911 during standoff 7 refers to the real-world event instance Orlando nightclub shooting 8 , which is an instance of several event types according to Wikidata, including mass shooting 9 and mass murder. 10 Based on Wikidata, the incident date can be obtained.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Resources", "sec_num": "3.1" }, { "text": "Commonly, our pointer to a reference text is an URL. We apply the following steps to locate, retrieve, and process the reference text. First, we make use of the Internet archive Wayback Machine 11 . Please note that this step is not successful for all URLs. Second, we apply news-please (Hamborg et al., 2017) to crawl the reference text as well as the publication date. Third, we process the text using spaCy (Honnibal et al., 2020) for sentence splitting, tokenization, lemmatization, and dependency parsing. Finally, we apply Open-SESAME (Swayamdipta et al., 2017) , which was retrained in order to be used. The collection process results in a document with annotations for various NLP tasks, including frame identification, and the publishing date of the document is typically known.", "cite_spans": [ { "start": 194, "end": 196, "text": "11", "ref_id": null }, { "start": 541, "end": 567, "text": "(Swayamdipta et al., 2017)", "ref_id": "BIBREF34" } ], "ref_spans": [], "eq_spans": [], "section": "Resources", "sec_num": "3.1" }, { "text": "We make use of two routes to obtain data for HDD according to our model. We apply MWEP on three Wikidata event types: presidential election (Q858439), storm (Q81054), and music festival (Q868557). The second source is a Kaggle dataset called Gun Violence Data (Ko, 2018) , which con-tains approximately 260,000 real-world instances regarding the event type gun violence, containing links between reference text URLs and the realworld instances. The four event types are selected due to their differentiation of conceptual properties, which makes them suitable for a contrastive analysis. The descriptive statistics of applying our retrieval and processing software are shown in Table 1.", "cite_spans": [ { "start": 260, "end": 270, "text": "(Ko, 2018)", "ref_id": "BIBREF21" } ], "ref_spans": [], "eq_spans": [], "section": "Resources", "sec_num": "3.1" }, { "text": "For each of the four selected event types, Table 1 presents the descriptive statistics. MWEP is capable of generating data for various different event types. However, the number of incidents and reference texts are limited, while the number of reference texts per incident is relatively high. The gun violence dataset, on the other hand, provides a high number of incidents for one specific event type, i.e., gun violence, but the number of texts per incident is relatively low.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Resources", "sec_num": "3.1" }, { "text": "Finally, we compute the temporal distance, which we define as the number of days between the incident date and the publishing date of a reference texts that makes reference to it. We visualize the distribution of temporal distance for the event type gun violence (Q5618454) for those reference texts for which we were able to obtain a publishing date, see Figure 1 . Figure 1 : The distribution of the temporal distance is shown for the reference texts that are published within 25 days of the incident, which holds for approximately 90% of the reference texts for the event type gun violence (Q5618454) Figure 1 visualizes the distribution of temporal distance for the event type gun violence (Q5618454). Most texts are published at the day of the incident. As time passes, the number of documents written about an incident decreases. Still, more than 10,000 are written after 25 days have passed.", "cite_spans": [], "ref_spans": [ { "start": 356, "end": 364, "text": "Figure 1", "ref_id": null }, { "start": 367, "end": 375, "text": "Figure 1", "ref_id": null }, { "start": 604, "end": 612, "text": "Figure 1", "ref_id": null } ], "eq_spans": [], "section": "Resources", "sec_num": "3.1" }, { "text": "We chose to train our diagnostic classifier on the gun violence data, since this subcorpus of HDD shares the largest volume of texts. The following steps were taken to preprocess the data for training. First, a subset of 6,290 documents containing less than 10 annotated frames were removed. These are most likely documents whose URLs were not successfully retrieved by the Wayback Machine, resulting in raw text of error messages, cookies etc. We also removed a subset of 16,237 documents for which news-please was not able to retrieve the publication date.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Processing the corpus", "sec_num": "3.2" }, { "text": "Next, we specified two temporal distance classes: \"day 0\" and \"day 8-30\". The remainder of documents were categorized into those classes according their publication date. After this step, day 0 covers 38,930 documents and day 8-30 covers 6,291 documents. We chose to train a Linear Support Vector model with both this unbalanced variant and a balanced variant in which the documents of day 0 are reduced to a randomized set of equal size as day 8-30.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Processing the corpus", "sec_num": "3.2" }, { "text": "Per document, both the frequencies of the frames and of their predicates were extracted and separately implemented as features in a data frame. A column was added with the temporal distance classes as labels. Each data frame was split into a training set (80%), a development set (10%) and a test set (10%). We ended up with data frames for both predicates and frames in a balanced and unbalanced corpus condition (4 experiments).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Processing the corpus", "sec_num": "3.2" }, { "text": "For each experiment, LinearSVC from Scikit Learn (Pedregosa et al., 2011) was used to train a Linear Support Vector with both the features of the experiment and the temporal distance classes as labels. This diagnostic classifier was applied to the test set and evaluated as a multi-class task per experiment.", "cite_spans": [ { "start": 49, "end": 73, "text": "(Pedregosa et al., 2011)", "ref_id": "BIBREF25" } ], "ref_spans": [], "eq_spans": [], "section": "Processing the corpus", "sec_num": "3.2" }, { "text": "The HDD corpus was first used for a contrastive analysis between event types and between temporal distance classes of gun violence. The aim was to derive typical frames, i.e., frames that are typically evoked in reference to a certain event type. The following steps were taken to process the data. We selected the data for the event types presidential election, storm and music festival, from which a total set of 57 documents containing less than 10 annotated frames were removed. Table 1 : Descriptive statistics regarding the key data concepts of the data forming HDD, used for the experiments. The first three rows originate from using MWEP to obtain data, whereas Gun Violence Data (Ko, 2018) is used for the data of the last row. The first column indicates the event types and the Wikidata identifier of the event type. The second column, L i , indicates the number of real-world incidents that belong to the event type. The third column, R i , presents the total number of reference texts, each referring to one of the real-world incidents. Finally, the average number of reference texts per real-world event instance are shown.", "cite_spans": [ { "start": 688, "end": 698, "text": "(Ko, 2018)", "ref_id": "BIBREF21" } ], "ref_spans": [ { "start": 483, "end": 490, "text": "Table 1", "ref_id": null } ], "eq_spans": [], "section": "Typical frame detection", "sec_num": "3.3" }, { "text": "gun violence, we used the documents for which the publication date could not be retrieved. Next, the corpus was randomly sampled by equalizing the volume of texts to the smallest collection (N=191), resulting in an equal amount of reference texts per event type. For the analysis between event types, all frame annotations were extracted from the documents and compiled per event type. Next, we apply an FFICF metric (a derivative of C-TFIDF), where FF stands for the frame frequency in a subcorpus, and ICF is the inverse collection frequency (Vossen et al., 2020) . This results in an FFICF score (henceforth typicality score) per frame per event type.", "cite_spans": [ { "start": 544, "end": 565, "text": "(Vossen et al., 2020)", "ref_id": "BIBREF36" } ], "ref_spans": [], "eq_spans": [], "section": "Typical frame detection", "sec_num": "3.3" }, { "text": "C-TFIDF was designed by Grootendorst (2020) with the purpose of determining the topic of a word cluster based on the set of highest scoring words. We have the advantage that, due to the data-to-text approach, the documents in HDD are already clustered based on predefined topics, i.e., event types. It follows that if we apply C-TFIDF to our corpus, we merely have to validate the highest scoring frames. Adapted to collections of frames, the mathematical model reads as follows:", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Typical frame detection", "sec_num": "3.3" }, { "text": "EQUATION", "cite_spans": [], "ref_spans": [], "eq_spans": [ { "start": 0, "end": 8, "text": "EQUATION", "ref_id": "EQREF", "raw_str": "F F \u2212 ICF i = t i f i \u00d7 log m n j t j", "eq_num": "(1)" } ], "section": "Typical frame detection", "sec_num": "3.3" }, { "text": "where the frequency of each frame t is extracted for each event type i and divided by the total number of frames of that event type. Then, the total number of documents m across event types is divided by the total frequency of the frame t across event types n.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Typical frame detection", "sec_num": "3.3" }, { "text": "We applied this metric to our sampled subset of HDD and ranked the typicality scores per event type. Furthermore, we performed a similar FFICF procedure between the temporal distance classes of gun violence.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Typical frame detection", "sec_num": "3.3" }, { "text": "We expect the frames with high typicality scores to differ between event types. The frames with the lowest typicality scores may be similar across event types, being a-typical.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Hypotheses 1. FFICF between event types", "sec_num": "3.4" }, { "text": "We expect the frames with high typicality scores to differ between texts from the same event type gun violence but from different temporal distance classes due to foregrounding and backgrounding.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "FFICF between temporal distance classes", "sec_num": "2." }, { "text": "We expect the diagnostic classifier to perform above chance in predicting the temporal distance class given a document, when the texts are represented by the typically evoked frames. With frame frequencies as features, the model will outperform word based models.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Training and testing the Linear SVM", "sec_num": "3." }, { "text": "In order to validate the outcome of the contrastive analysis between temporal distance classes, we presented two annotators with frames from the subcorpus of gun violence for which the publication dates could not be retrieved. Frames with three or less occurrences across this subcorpus were filtered out. For each of the remaining 282 frames, the annotators were asked to provide a binary judgment about whether it is typical in reference to an incident of gun violence at day 0. We utilized the notion narrative container (NC) from Pustejovsky and Stubbs (2011), i.e., the scope between the onset of the event instance and the document creation time, to estimate the possible subevents that have a high chance of being referred to in a document on day 0. The annotators had to judge whether each frame is part of the NC. We used Cohen's kappa (Cohen, 1960) to obtain a measure of inter-annotator-agreement. We expect that the frames annotated as part of the NC of day 0, also occur in the top rank of FFICF scores for this class, whereas frames annotated as falling outside of the NC occur in the top rank of FFICF scores for day 8-30.", "cite_spans": [ { "start": 845, "end": 858, "text": "(Cohen, 1960)", "ref_id": "BIBREF4" } ], "ref_spans": [], "eq_spans": [], "section": "Evaluation", "sec_num": "3.5" }, { "text": "We evaluated the output of the diagnostic classifier in a multi-class classification report with precision, recall and F1-score in addition to accuracy, macro average and weighted average.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Evaluation", "sec_num": "3.5" }, { "text": "For the contrastive analysis between event types, Table 2 , shows the top and bottom ranked FFICF scores for the event types gun violence and music festival. The top ranked frames differentiate between event types and appear to reflect their typical properties. In contrast, the bottom ranked frames are the same for both types and reflect generic event properties.", "cite_spans": [], "ref_spans": [ { "start": 50, "end": 57, "text": "Table 2", "ref_id": "TABREF3" } ], "eq_spans": [], "section": "Results", "sec_num": "4" }, { "text": "For the contrastive analysis between temporal distance classes, Table 3 shows the top and bottom ranked FFICF scores between the two temporal distance classes of the event type gun violence. LAW ENFORCEMENT AGENCY, KILLING and CATASTROPHE, which were in the top ranking in Table 2 , ended in the bottom ranking here. Furthermore, except for two frames, the top ranking of both classes in Table 3 is occupied by different frames.", "cite_spans": [], "ref_spans": [ { "start": 64, "end": 71, "text": "Table 3", "ref_id": "TABREF4" }, { "start": 273, "end": 280, "text": "Table 2", "ref_id": "TABREF3" }, { "start": 388, "end": 395, "text": "Table 3", "ref_id": "TABREF4" } ], "eq_spans": [], "section": "Results", "sec_num": "4" }, { "text": "The annotators show a Cohen's kappa of .48, which is moderate. However, their judgments on the frames in the top and bottom ranking of FFICF ratings in Table 3 show a rather high agreement (20 out of 26 frames, 77%). Half of the top ranked frames in day 0 are annotated as part of the NC of day 0, and almost all top ranked frames at day 8-30 are annotated as not belonging to that same NC. Note that the three frames that are both in the top ranking of scores between event types and at the bottom ranking of scores between temporal distance classes, are also annotated as part of the NC. Table 4 displays the evaluation report of the experiments with the Linear SVM classifier. In the unbalanced conditions, the accuracy is above 0.85, but biased towards the performance for day 0. The model performed below chance in predicting day 8-30. For frames, the performance in this class is higher than for predicates. In the balanced conditions, the performance decreases for day 0, but increases for day 8-30. For predicates, the model performs around and above chance, with higher recall for day 0 and lower recall for day 8-30. For frames, the F1 is above 0.75, with consistent precision and recall.", "cite_spans": [], "ref_spans": [ { "start": 152, "end": 159, "text": "Table 3", "ref_id": "TABREF4" }, { "start": 590, "end": 597, "text": "Table 4", "ref_id": "TABREF5" } ], "eq_spans": [], "section": "Results", "sec_num": "4" }, { "text": "In Table 3 , we find that SHOOT PROJECTILES and JUDGMENT COMMUNICATION remain in the top ranking, each in a different class. All other frames in the top ranking are typically used in reference to the events of their respective class. Many of those frames can be considered typical for gun violence (e.g., EXPERIENCE BODILY HARM, JUDI-CIAL BODY), but their evocation is subjected to the temporal distance class. The frames on day 0 refer to subevents of the central event instance, while the frames on day 8-30 refer to related disjoint events, as is generally validated by the annotators. We interpret this variation as an effect of foregrounding and backgrounding. Most typical frames on day 0 are backgrounded in day 8-30 due to the high common ground. They are pragmatically implicated in order to foreground the frames of day 8-30, which carry the highest informative value, but are not typically used in reference to the central event instance of day 0.", "cite_spans": [], "ref_spans": [ { "start": 3, "end": 10, "text": "Table 3", "ref_id": "TABREF4" } ], "eq_spans": [], "section": "Discussion", "sec_num": "5" }, { "text": "Recall that in order to implicate shared knowledge, one uses minimal or less marked expressions. Thus, if the typical frames of day 0 have become shared knowledge in day 8-30, then the writer optimally uses short and definite expressions to implicate them. Such expressions then evoke a strong typical frame, an anchor frame, that is sufficient to both refer to the event type and implicate the typical frames as shared knowledge. Such an anchor frame should show a high typicality score for the event type, but a low score across temporal distance classes, due to its frequent usage. KILLING and CATASTROPHE in Table 2 and Table 3 meet both requirements and refer to the main event instance. These might behave as anchor frames on day 8-30, backgrounding the main event instance and implicating the typical frames of day 0 as shared knowledge. This is demonstrated in (1b), where KILLING is evoked in the backgrounded noun phrase.", "cite_spans": [], "ref_spans": [ { "start": 612, "end": 631, "text": "Table 2 and Table 3", "ref_id": "TABREF3" } ], "eq_spans": [], "section": "Discussion", "sec_num": "5" }, { "text": "Finally, the results of the diagnostic classifier in Table 4 show that frame occurrences are more informative for the model than predicate occurrences. The above-chance performance of the model in the balanced/frames condition shows that it is capable to learn temporal patterns, just by paying attention to frame occurrences. We performed a model analysis to derive a ranking of the most important frames that the model used as margins to derive the hyperplane. The top 5 reads: TEMPORAL SUBREGION, BE-COMING SILENT, SELF MOTION, STORE and EN-FORCING. None of these frames get a high typicality score in Table 3 . Although the typical frames in the FFICF analysis show strong effects of foregrounding and backgrounding, idiosyncratic generic frames in the data seem more informative for the model in finding the most optimal separating hyperplane. TEMPORAL SUBREGION might be a strong generic contender across event types due to its inherent temporal properties. 12 BECOMING SILENT 13 , SELF MOTION 14 and EN-FORCING 15 might show a significant frequency in a specific class in reference to the main event instance or subevents.", "cite_spans": [], "ref_spans": [ { "start": 53, "end": 60, "text": "Table 4", "ref_id": "TABREF5" }, { "start": 605, "end": 612, "text": "Table 3", "ref_id": "TABREF4" } ], "eq_spans": [], "section": "Discussion", "sec_num": "5" }, { "text": "We assume that the results of our analysis can be generalized over unpredicted event types. From the onset, the common ground increases over time, affecting the pragmatic principles of foregrounding and backgrounding. Thus, if we would be able to obtain enough texts for the event type storm, we would expect the variation in framing between temporal classes to only occur with this event type as well. Since presidential election and music festival are rather anticipated events, the common ground at day 0 is at maximum height and build up from texts in preceding days. Thus, for these event types, temporal distance classes should be determined from preceding days up until the event itself.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Discussion", "sec_num": "5" }, { "text": "In this paper, we measured variation in framing as a function of pragmatic foregrounding and backgrounding. We hypothesized that difference in common ground determine the extent to which the writer is able to background frames typically used in reference to the main event instance. We presented HDD, a corpus consisting of reference texts grouped under event types and enriched with publication dates. HDD was used to both perform FFICF between event types and between temporal distance classes, and train a diagnostic classifier. The former resulted in a ranking of typical frames per event type and between classes. The Linear SVM to a large extent was able to differentiate documents of different temporal distance classes. Frames turned out to be more informative than their predicates in training the model. Yet, The diagnostic classifier prefers idiosynchratic frames for learning the hyperplane.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "6" }, { "text": "In future work, we extend our experiments to more event types and we want to learn the specific frame-to-frame relations from the typical frames for event types. We expect to learn subevent relations from texts with short temporal distance and (causal) sequence relations from typical frames in texts with larger temporal distance.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion", "sec_num": "6" }, { "text": "In this paper we use the term event instance for event instances of a specific event type, e.g., an instance of shooting", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "https://www.chron.com/houston/articl e/Shooting-levaes-man-dead-in-SW-Houst on-6688587.php, published on the same day as the event instance.3 http://www.news4jax.com/news/crime/1arrest-in-westside-murder, published a week after the event instance.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "https://framenet.icsi.berkeley.edu/f ndrupal/", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "On discourse level, referential variation as an effect of common ground has been studied more intensively. SeeYoshida (2011);Markert et al. (2012); Del Tredici and Fern\u00e1ndez (2018).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "the code is available at https://github.com/clt l/HDDanalysis.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Examples of lexical units: later.a, earlier.a, early.a. 13 Examples of lexical units: quiet.v, silence.v 14 Examples of lexical units: walk.v, run.v, rush.v 15 Examples of lexical units: enforcement.n, enforce.v", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null } ], "back_matter": [ { "text": "This research was funded by the Dutch National Science Organisation (NWO) under the project number VC.GW17.083, Framing Situations in the Dutch language.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Acknowledgments", "sec_num": null } ], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "The Structure of the FrameNet Database", "authors": [ { "first": "Collin", "middle": [ "F" ], "last": "Baker", "suffix": "" }, { "first": "Charles", "middle": [ "J" ], "last": "Fillmore", "suffix": "" }, { "first": "Beau", "middle": [], "last": "Cronin", "suffix": "" } ], "year": 2003, "venue": "International Journal of Lexicography", "volume": "16", "issue": "3", "pages": "281--296", "other_ids": {}, "num": null, "urls": [], "raw_text": "Collin F. Baker, Charles J. Fillmore, and Beau Cronin. 2003. The Structure of the FrameNet Database. International Journal of Lexicography, 16(3):281- 296.", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "Unsupervised event coreference resolution with rich linguistic features", "authors": [ { "first": "Cosmin", "middle": [], "last": "Bejan", "suffix": "" }, { "first": "Sanda", "middle": [], "last": "Harabagiu", "suffix": "" } ], "year": 2010, "venue": "Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics", "volume": "", "issue": "", "pages": "1412--1422", "other_ids": {}, "num": null, "urls": [], "raw_text": "Cosmin Bejan and Sanda Harabagiu. 2010. Unsuper- vised event coreference resolution with rich linguis- tic features. In Proceedings of the 48th Annual Meet- ing of the Association for Computational Linguis- tics, pages 1412-1422, Uppsala, Sweden. Associa- tion for Computational Linguistics.", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "Identifying the most dominant event in a news article by mining event coreference relations", "authors": [ { "first": "Prafulla", "middle": [], "last": "Kumar Choubey", "suffix": "" }, { "first": "Kaushik", "middle": [], "last": "Raju", "suffix": "" }, { "first": "Ruihong", "middle": [], "last": "Huang", "suffix": "" } ], "year": 2018, "venue": "Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies", "volume": "2", "issue": "", "pages": "340--345", "other_ids": { "DOI": [ "10.18653/v1/N18-2055" ] }, "num": null, "urls": [], "raw_text": "Prafulla Kumar Choubey, Kaushik Raju, and Ruihong Huang. 2018. Identifying the most dominant event in a news article by mining event coreference rela- tions. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Com- putational Linguistics: Human Language Technolo- gies, Volume 2 (Short Papers), pages 340-345, New Orleans, Louisiana. Association for Computational Linguistics.", "links": null }, "BIBREF4": { "ref_id": "b4", "title": "A coefficient of agreement for nominal scales. Educational and psychological measurement", "authors": [ { "first": "Jacob", "middle": [], "last": "Cohen", "suffix": "" } ], "year": 1960, "venue": "", "volume": "20", "issue": "", "pages": "37--46", "other_ids": {}, "num": null, "urls": [], "raw_text": "Jacob Cohen. 1960. A coefficient of agreement for nominal scales. Educational and psychological mea- surement, 20(1):37-46.", "links": null }, "BIBREF5": { "ref_id": "b5", "title": "Event models for historical perspectives: Determining relations between high and low level events in text, based on the classification of time, location and participants", "authors": [ { "first": "Agata", "middle": [], "last": "Cybulska", "suffix": "" }, { "first": "Piek", "middle": [], "last": "Vossen", "suffix": "" } ], "year": 2010, "venue": "Proceedings of the Seventh International Conference on Language Resources and Evaluation (LREC'10)", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Agata Cybulska and Piek Vossen. 2010. Event mod- els for historical perspectives: Determining relations between high and low level events in text, based on the classification of time, location and partici- pants. In Proceedings of the Seventh International Conference on Language Resources and Evaluation (LREC'10), Valletta, Malta. European Language Re- sources Association (ELRA).", "links": null }, "BIBREF6": { "ref_id": "b6", "title": "Using a sledgehammer to crack a nut? lexical diversity and event coreference resolution", "authors": [ { "first": "Agata", "middle": [], "last": "Cybulska", "suffix": "" }, { "first": "Piek", "middle": [], "last": "Vossen", "suffix": "" } ], "year": 2014, "venue": "Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC'14)", "volume": "", "issue": "", "pages": "4545--4552", "other_ids": {}, "num": null, "urls": [], "raw_text": "Agata Cybulska and Piek Vossen. 2014. Using a sledgehammer to crack a nut? lexical diversity and event coreference resolution. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC'14), pages 4545- 4552, Reykjavik, Iceland. European Language Re- sources Association (ELRA).", "links": null }, "BIBREF7": { "ref_id": "b7", "title": "The use of syntactic clues in discourse processing", "authors": [ { "first": "Nan", "middle": [], "last": "Decker", "suffix": "" } ], "year": 1985, "venue": "23rd Annual Meeting of the Association for Computational Linguistics", "volume": "", "issue": "", "pages": "315--323", "other_ids": { "DOI": [ "10.3115/981210.981249" ] }, "num": null, "urls": [], "raw_text": "Nan Decker. 1985. The use of syntactic clues in dis- course processing. In 23rd Annual Meeting of the Association for Computational Linguistics, pages 315-323, Chicago, Illinois, USA. Association for Computational Linguistics.", "links": null }, "BIBREF8": { "ref_id": "b8", "title": "The road to success: Assessing the fate of linguistic innovations in online communities", "authors": [ { "first": "Marco", "middle": [], "last": "Del Tredici", "suffix": "" }, { "first": "Raquel", "middle": [], "last": "Fern\u00e1ndez", "suffix": "" } ], "year": 2018, "venue": "", "volume": "", "issue": "", "pages": "1591--1603", "other_ids": {}, "num": null, "urls": [], "raw_text": "Marco Del Tredici and Raquel Fern\u00e1ndez. 2018. The road to success: Assessing the fate of linguistic inno- vations in online communities. pages 1591-1603.", "links": null }, "BIBREF9": { "ref_id": "b9", "title": "Event-based extractive summarization", "authors": [ { "first": "Elena", "middle": [], "last": "Filatova", "suffix": "" }, { "first": "Vasileios", "middle": [], "last": "Hatzivassiloglou", "suffix": "" } ], "year": 2004, "venue": "Text Summarization Branches Out", "volume": "", "issue": "", "pages": "104--111", "other_ids": {}, "num": null, "urls": [], "raw_text": "Elena Filatova and Vasileios Hatzivassiloglou. 2004a. Event-based extractive summarization. In Text Summarization Branches Out, pages 104-111, Barcelona, Spain. Association for Computational Linguistics.", "links": null }, "BIBREF10": { "ref_id": "b10", "title": "A formal model for information selection in multisentence text extraction", "authors": [ { "first": "Elena", "middle": [], "last": "Filatova", "suffix": "" }, { "first": "Vasileios", "middle": [], "last": "Hatzivassiloglou", "suffix": "" } ], "year": 2004, "venue": "COLING 2004: Proceedings of the 20th International Conference on Computational Linguistics", "volume": "", "issue": "", "pages": "397--403", "other_ids": {}, "num": null, "urls": [], "raw_text": "Elena Filatova and Vasileios Hatzivassiloglou. 2004b. A formal model for information selection in multi- sentence text extraction. In COLING 2004: Pro- ceedings of the 20th International Conference on Computational Linguistics, pages 397-403, Geneva, Switzerland. COLING.", "links": null }, "BIBREF11": { "ref_id": "b11", "title": "A Frames Approach to Semantic Analysis", "authors": [ { "first": "J", "middle": [], "last": "Charles", "suffix": "" }, { "first": "Collin", "middle": [], "last": "Fillmore", "suffix": "" }, { "first": "", "middle": [], "last": "Baker", "suffix": "" } ], "year": 2010, "venue": "The Oxford handbook of linguistic analysis", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Charles J Fillmore and Collin Baker. 2010. A Frames Approach to Semantic Analysis. In The Oxford handbook of linguistic analysis. Oxford University Press.", "links": null }, "BIBREF12": { "ref_id": "b12", "title": "Background to framenet", "authors": [ { "first": "J", "middle": [], "last": "Charles", "suffix": "" }, { "first": "", "middle": [], "last": "Fillmore", "suffix": "" }, { "first": "Miriam Rl", "middle": [], "last": "Christopher R Johnson", "suffix": "" }, { "first": "", "middle": [], "last": "Petruck", "suffix": "" } ], "year": 2003, "venue": "International journal of lexicography", "volume": "16", "issue": "3", "pages": "235--250", "other_ids": {}, "num": null, "urls": [], "raw_text": "Charles J Fillmore, Christopher R Johnson, and Miriam RL Petruck. 2003. Background to framenet. International journal of lexicography, 16(3):235- 250.", "links": null }, "BIBREF13": { "ref_id": "b13", "title": "Logic and conversation", "authors": [ { "first": "", "middle": [], "last": "H Paul Grice", "suffix": "" } ], "year": 1975, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "H Paul Grice. 1975. Logic and conversation. Cole, P., and Morgan, J.(Eds.), 3.", "links": null }, "BIBREF14": { "ref_id": "b14", "title": "The thread of discourse", "authors": [ { "first": "E", "middle": [], "last": "Joseph", "suffix": "" }, { "first": "", "middle": [], "last": "Grimes", "suffix": "" } ], "year": 2015, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Joseph E Grimes. 2015. The thread of discourse, vol- ume 207. Walter de Gruyter GmbH & Co KG.", "links": null }, "BIBREF15": { "ref_id": "b15", "title": "Bertopic: Leveraging bert and c-tf-idf to create easily interpretable topics", "authors": [ { "first": "Maarten", "middle": [], "last": "Grootendorst", "suffix": "" } ], "year": 2020, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": { "DOI": [ "10.5281/zenodo.4430182" ] }, "num": null, "urls": [], "raw_text": "Maarten Grootendorst. 2020. Bertopic: Leveraging bert and c-tf-idf to create easily interpretable topics.", "links": null }, "BIBREF16": { "ref_id": "b16", "title": "news-please: A generic news crawler and extractor", "authors": [ { "first": "Felix", "middle": [], "last": "Hamborg", "suffix": "" }, { "first": "Norman", "middle": [], "last": "Meuschke", "suffix": "" }, { "first": "Corinna", "middle": [], "last": "Breitinger", "suffix": "" }, { "first": "Bela", "middle": [], "last": "Gipp", "suffix": "" } ], "year": 2017, "venue": "15th International Symposium of Information Science (ISI 2017)", "volume": "", "issue": "", "pages": "218--223", "other_ids": {}, "num": null, "urls": [], "raw_text": "Felix Hamborg, Norman Meuschke, Corinna Bre- itinger, and Bela Gipp. 2017. news-please: A generic news crawler and extractor. In 15th Interna- tional Symposium of Information Science (ISI 2017), pages 218-223.", "links": null }, "BIBREF17": { "ref_id": "b17", "title": "spaCy: Industrial-strength Natural Language Processing in Python", "authors": [], "year": null, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": { "DOI": [ "10.5281/zenodo.1212303" ] }, "num": null, "urls": [], "raw_text": "Matthew Honnibal, Ines Montani, Sofie Van Lan- deghem, and Adriane Boyd. 2020. spaCy: Industrial-strength Natural Language Processing in Python.", "links": null }, "BIBREF18": { "ref_id": "b18", "title": "Toward a new taxonomy for pragmatic inference: Q-based and r-based implicature", "authors": [ { "first": "", "middle": [], "last": "Lawrence R Horn", "suffix": "" } ], "year": 1998, "venue": "Pragmatics: Critical Concepts", "volume": "4", "issue": "", "pages": "383--417", "other_ids": {}, "num": null, "urls": [], "raw_text": "Lawrence R Horn. 1998. Toward a new taxonomy for pragmatic inference: Q-based and r-based implica- ture. Pragmatics: Critical Concepts, 4:383-417.", "links": null }, "BIBREF19": { "ref_id": "b19", "title": "Semantic overfitting: what 'world' do we consider when evaluating disambiguation of text?", "authors": [ { "first": "Filip", "middle": [], "last": "Ilievski", "suffix": "" }, { "first": "Marten", "middle": [], "last": "Postma", "suffix": "" }, { "first": "Piek", "middle": [], "last": "Vossen", "suffix": "" } ], "year": 2016, "venue": "Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers", "volume": "", "issue": "", "pages": "1180--1191", "other_ids": {}, "num": null, "urls": [], "raw_text": "Filip Ilievski, Marten Postma, and Piek Vossen. 2016. Semantic overfitting: what 'world' do we consider when evaluating disambiguation of text? In Pro- ceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Techni- cal Papers, pages 1180-1191, Osaka, Japan. The COLING 2016 Organizing Committee.", "links": null }, "BIBREF20": { "ref_id": "b20", "title": "Transitivity and foregrounding in news articles: Experiments in information retrieval and automatic summarising", "authors": [ { "first": "Roderick", "middle": [], "last": "Kay", "suffix": "" }, { "first": "Ruth", "middle": [], "last": "Aylett", "suffix": "" } ], "year": 1996, "venue": "34th Annual Meeting of the Association for Computational Linguistics", "volume": "", "issue": "", "pages": "369--371", "other_ids": { "DOI": [ "10.3115/981863.981918" ] }, "num": null, "urls": [], "raw_text": "Roderick Kay and Ruth Aylett. 1996. Transitivity and foregrounding in news articles: Experiments in information retrieval and automatic summaris- ing. In 34th Annual Meeting of the Association for Computational Linguistics, pages 369-371, Santa Cruz, California, USA. Association for Computa- tional Linguistics.", "links": null }, "BIBREF21": { "ref_id": "b21", "title": "Gun Violence Data", "authors": [ { "first": "James", "middle": [], "last": "Ko", "suffix": "" } ], "year": 2018, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "James Ko. 2018. Gun Violence Data. https://www. kaggle.com/jameslko/gun-violence-data.", "links": null }, "BIBREF22": { "ref_id": "b22", "title": "Collective classification for fine-grained information status", "authors": [ { "first": "Katja", "middle": [], "last": "Markert", "suffix": "" }, { "first": "Yufang", "middle": [], "last": "Hou", "suffix": "" }, { "first": "Michael", "middle": [], "last": "Strube", "suffix": "" } ], "year": 2012, "venue": "Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics", "volume": "", "issue": "", "pages": "795--804", "other_ids": {}, "num": null, "urls": [], "raw_text": "Katja Markert, Yufang Hou, and Michael Strube. 2012. Collective classification for fine-grained information status. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics (Vol- ume 1: Long Papers), pages 795-804, Jeju Island, Korea. Association for Computational Linguistics.", "links": null }, "BIBREF23": { "ref_id": "b23", "title": "Converging operations on a basic level in event taxonomies", "authors": [ { "first": "W", "middle": [], "last": "Michael", "suffix": "" }, { "first": "Gregory L", "middle": [], "last": "Morris", "suffix": "" }, { "first": "", "middle": [], "last": "Murphy", "suffix": "" } ], "year": 1990, "venue": "Memory & Cognition", "volume": "18", "issue": "4", "pages": "407--418", "other_ids": {}, "num": null, "urls": [], "raw_text": "Michael W Morris and Gregory L Murphy. 1990. Con- verging operations on a basic level in event tax- onomies. Memory & Cognition, 18(4):407-418.", "links": null }, "BIBREF24": { "ref_id": "b24", "title": "Richer event description: Integrating event coreference with temporal, causal and bridging annotation", "authors": [ { "first": "Kristin", "middle": [], "last": "Tim O'gorman", "suffix": "" }, { "first": "Martha", "middle": [], "last": "Wright-Bettner", "suffix": "" }, { "first": "", "middle": [], "last": "Palmer", "suffix": "" } ], "year": 2016, "venue": "Proceedings of the 2nd Workshop on Computing News Storylines (CNS 2016)", "volume": "", "issue": "", "pages": "47--56", "other_ids": { "DOI": [ "10.18653/v1/W16-5706" ] }, "num": null, "urls": [], "raw_text": "Tim O'Gorman, Kristin Wright-Bettner, and Martha Palmer. 2016. Richer event description: Integrating event coreference with temporal, causal and bridg- ing annotation. In Proceedings of the 2nd Work- shop on Computing News Storylines (CNS 2016), pages 47-56, Austin, Texas. Association for Com- putational Linguistics.", "links": null }, "BIBREF25": { "ref_id": "b25", "title": "Scikit-learn: Machine learning in Python", "authors": [ { "first": "F", "middle": [], "last": "Pedregosa", "suffix": "" }, { "first": "G", "middle": [], "last": "Varoquaux", "suffix": "" }, { "first": "A", "middle": [], "last": "Gramfort", "suffix": "" }, { "first": "V", "middle": [], "last": "Michel", "suffix": "" }, { "first": "B", "middle": [], "last": "Thirion", "suffix": "" }, { "first": "O", "middle": [], "last": "Grisel", "suffix": "" }, { "first": "M", "middle": [], "last": "Blondel", "suffix": "" }, { "first": "P", "middle": [], "last": "Prettenhofer", "suffix": "" }, { "first": "R", "middle": [], "last": "Weiss", "suffix": "" }, { "first": "V", "middle": [], "last": "Dubourg", "suffix": "" }, { "first": "J", "middle": [], "last": "Vanderplas", "suffix": "" }, { "first": "A", "middle": [], "last": "Passos", "suffix": "" }, { "first": "D", "middle": [], "last": "Cournapeau", "suffix": "" }, { "first": "M", "middle": [], "last": "Brucher", "suffix": "" }, { "first": "M", "middle": [], "last": "Perrot", "suffix": "" }, { "first": "E", "middle": [], "last": "Duchesnay", "suffix": "" } ], "year": 2011, "venue": "Journal of Machine Learning Research", "volume": "12", "issue": "", "pages": "2825--2830", "other_ids": {}, "num": null, "urls": [], "raw_text": "F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duch- esnay. 2011. Scikit-learn: Machine learning in Python. Journal of Machine Learning Research, 12:2825-2830.", "links": null }, "BIBREF26": { "ref_id": "b26", "title": "Event detection and co-reference with minimal supervision", "authors": [ { "first": "Haoruo", "middle": [], "last": "Peng", "suffix": "" }, { "first": "Yangqiu", "middle": [], "last": "Song", "suffix": "" }, { "first": "Dan", "middle": [], "last": "Roth", "suffix": "" } ], "year": 2016, "venue": "Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing", "volume": "", "issue": "", "pages": "392--402", "other_ids": { "DOI": [ "10.18653/v1/D16-1038" ] }, "num": null, "urls": [], "raw_text": "Haoruo Peng, Yangqiu Song, and Dan Roth. 2016. Event detection and co-reference with minimal su- pervision. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Process- ing, pages 392-402, Austin, Texas. Association for Computational Linguistics.", "links": null }, "BIBREF27": { "ref_id": "b27", "title": "Moving away from semantic overfitting in disambiguation datasets", "authors": [ { "first": "Marten", "middle": [], "last": "Postma", "suffix": "" }, { "first": "Filip", "middle": [], "last": "Ilievski", "suffix": "" }, { "first": "Piek", "middle": [], "last": "Vossen", "suffix": "" }, { "first": "Marieke", "middle": [], "last": "Van Erp", "suffix": "" } ], "year": 2016, "venue": "Proceedings of the Workshop on Uphill Battles in Language Processing: Scaling Early Achievements to Robust Methods", "volume": "", "issue": "", "pages": "17--21", "other_ids": { "DOI": [ "10.18653/v1/W16-6004" ] }, "num": null, "urls": [], "raw_text": "Marten Postma, Filip Ilievski, Piek Vossen, and Marieke van Erp. 2016. Moving away from seman- tic overfitting in disambiguation datasets. In Pro- ceedings of the Workshop on Uphill Battles in Lan- guage Processing: Scaling Early Achievements to Robust Methods, pages 17-21, Austin, TX. Associa- tion for Computational Linguistics.", "links": null }, "BIBREF28": { "ref_id": "b28", "title": "SemEval-2007 task-17: English lexical sample, SRL and all words", "authors": [ { "first": "Sameer", "middle": [], "last": "Pradhan", "suffix": "" }, { "first": "Edward", "middle": [], "last": "Loper", "suffix": "" }, { "first": "Dmitriy", "middle": [], "last": "Dligach", "suffix": "" }, { "first": "Martha", "middle": [], "last": "Palmer", "suffix": "" } ], "year": 2007, "venue": "Proceedings of the Fourth International Workshop on Semantic Evaluations (SemEval-2007)", "volume": "", "issue": "", "pages": "87--92", "other_ids": {}, "num": null, "urls": [], "raw_text": "Sameer Pradhan, Edward Loper, Dmitriy Dligach, and Martha Palmer. 2007. SemEval-2007 task-17: En- glish lexical sample, SRL and all words. In Proceed- ings of the Fourth International Workshop on Se- mantic Evaluations (SemEval-2007), pages 87-92, Prague, Czech Republic. Association for Computa- tional Linguistics.", "links": null }, "BIBREF29": { "ref_id": "b29", "title": "Increasing informativeness in temporal annotation", "authors": [ { "first": "James", "middle": [], "last": "Pustejovsky", "suffix": "" }, { "first": "Amber", "middle": [], "last": "Stubbs", "suffix": "" } ], "year": 2011, "venue": "Proceedings of the 5th Linguistic Annotation Workshop", "volume": "", "issue": "", "pages": "152--160", "other_ids": {}, "num": null, "urls": [], "raw_text": "James Pustejovsky and Amber Stubbs. 2011. Increas- ing informativeness in temporal annotation. In Pro- ceedings of the 5th Linguistic Annotation Workshop, pages 152-160, Portland, Oregon, USA. Associa- tion for Computational Linguistics.", "links": null }, "BIBREF30": { "ref_id": "b30", "title": "FrameNet II: Extended theory and practice", "authors": [ { "first": "Josef", "middle": [], "last": "Ruppenhofer", "suffix": "" }, { "first": "Michael", "middle": [], "last": "Ellsworth", "suffix": "" }, { "first": "R L", "middle": [], "last": "Miriam", "suffix": "" }, { "first": "", "middle": [], "last": "Petruck", "suffix": "" }, { "first": "R", "middle": [], "last": "Christopher", "suffix": "" }, { "first": "", "middle": [], "last": "Johnson", "suffix": "" } ], "year": 2010, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Josef Ruppenhofer, Michael Ellsworth, Miriam R L Petruck, Christopher R Johnson, and Jan Schefczyk. 2010. FrameNet II: Extended theory and practice.", "links": null }, "BIBREF31": { "ref_id": "b31", "title": "The new york times annotated corpus. Linguistic Data Consortium", "authors": [ { "first": "Evan", "middle": [], "last": "Sandhaus", "suffix": "" } ], "year": 2008, "venue": "", "volume": "6", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Evan Sandhaus. 2008. The new york times annotated corpus. Linguistic Data Consortium, Philadelphia, 6(12):e26752.", "links": null }, "BIBREF32": { "ref_id": "b32", "title": "From light to rich ERE: Annotation of entities, relations, and events", "authors": [ { "first": "Zhiyi", "middle": [], "last": "Song", "suffix": "" }, { "first": "Ann", "middle": [], "last": "Bies", "suffix": "" }, { "first": "Stephanie", "middle": [], "last": "Strassel", "suffix": "" }, { "first": "Tom", "middle": [], "last": "Riese", "suffix": "" }, { "first": "Justin", "middle": [], "last": "Mott", "suffix": "" }, { "first": "Joe", "middle": [], "last": "Ellis", "suffix": "" }, { "first": "Jonathan", "middle": [], "last": "Wright", "suffix": "" }, { "first": "Seth", "middle": [], "last": "Kulick", "suffix": "" }, { "first": "Neville", "middle": [], "last": "Ryant", "suffix": "" }, { "first": "Xiaoyi", "middle": [], "last": "Ma", "suffix": "" } ], "year": 2015, "venue": "Proceedings of the The 3rd Workshop on EVENTS: Definition, Detection, Coreference, and Representation", "volume": "", "issue": "", "pages": "89--98", "other_ids": { "DOI": [ "10.3115/v1/W15-0812" ] }, "num": null, "urls": [], "raw_text": "Zhiyi Song, Ann Bies, Stephanie Strassel, Tom Riese, Justin Mott, Joe Ellis, Jonathan Wright, Seth Kulick, Neville Ryant, and Xiaoyi Ma. 2015. From light to rich ERE: Annotation of entities, relations, and events. In Proceedings of the The 3rd Workshop on EVENTS: Definition, Detection, Coreference, and Representation, pages 89-98, Denver, Colorado. As- sociation for Computational Linguistics.", "links": null }, "BIBREF33": { "ref_id": "b33", "title": "Getting to \"hearer-old\": Charting referring expressions across time", "authors": [ { "first": "Ieva", "middle": [], "last": "Stali\u016bnait\u0117", "suffix": "" }, { "first": "Hannah", "middle": [], "last": "Rohde", "suffix": "" }, { "first": "Bonnie", "middle": [], "last": "Webber", "suffix": "" }, { "first": "Annie", "middle": [], "last": "Louis", "suffix": "" } ], "year": 2018, "venue": "Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing", "volume": "", "issue": "", "pages": "4350--4359", "other_ids": { "DOI": [ "10.18653/v1/D18-1466" ] }, "num": null, "urls": [], "raw_text": "Ieva Stali\u016bnait\u0117, Hannah Rohde, Bonnie Webber, and Annie Louis. 2018. Getting to \"hearer-old\": Chart- ing referring expressions across time. In Proceed- ings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 4350-4359, Brussels, Belgium. Association for Computational Linguistics.", "links": null }, "BIBREF34": { "ref_id": "b34", "title": "Semantic Parsing with Softmax-Margin Segmental RNNs and a Syntactic Scaffold", "authors": [ { "first": "Swabha", "middle": [], "last": "Swayamdipta", "suffix": "" }, { "first": "Sam", "middle": [], "last": "Thomson", "suffix": "" }, { "first": "Chris", "middle": [], "last": "Dyer", "suffix": "" }, { "first": "Noah", "middle": [ "A" ], "last": "Smith", "suffix": "" } ], "year": 2017, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": { "arXiv": [ "arXiv:1706.09528" ] }, "num": null, "urls": [], "raw_text": "Swabha Swayamdipta, Sam Thomson, Chris Dyer, and Noah A. Smith. 2017. Frame-Semantic Parsing with Softmax-Margin Segmental RNNs and a Syntactic Scaffold. arXiv preprint arXiv:1706.09528.", "links": null }, "BIBREF35": { "ref_id": "b35", "title": "making the news\": Identifying noteworthy events in news articles", "authors": [ { "first": "Shyam", "middle": [], "last": "Upadhyay", "suffix": "" }, { "first": "Christos", "middle": [], "last": "Christodoulopoulos", "suffix": "" }, { "first": "Dan", "middle": [], "last": "Roth", "suffix": "" } ], "year": 2016, "venue": "Proceedings of the Fourth Workshop on Events", "volume": "", "issue": "", "pages": "1--7", "other_ids": { "DOI": [ "10.18653/v1/W16-1001" ] }, "num": null, "urls": [], "raw_text": "Shyam Upadhyay, Christos Christodoulopoulos, and Dan Roth. 2016. \"making the news\": Identifying noteworthy events in news articles. In Proceedings of the Fourth Workshop on Events, pages 1-7, San Diego, California. Association for Computational Linguistics.", "links": null }, "BIBREF36": { "ref_id": "b36", "title": "Large-scale cross-lingual language resources for referencing and framing", "authors": [ { "first": "Piek", "middle": [], "last": "Vossen", "suffix": "" }, { "first": "Filip", "middle": [], "last": "Ilievski", "suffix": "" }, { "first": "Marten", "middle": [], "last": "Postma", "suffix": "" }, { "first": "Antske", "middle": [], "last": "Fokkens", "suffix": "" }, { "first": "Gosse", "middle": [], "last": "Minnema", "suffix": "" }, { "first": "Levi", "middle": [], "last": "Remijnse", "suffix": "" } ], "year": 2020, "venue": "Proceedings of the 12th Language Resources and Evaluation Conference", "volume": "", "issue": "", "pages": "3162--3171", "other_ids": {}, "num": null, "urls": [], "raw_text": "Piek Vossen, Filip Ilievski, Marten Postma, Antske Fokkens, Gosse Minnema, and Levi Remijnse. 2020. Large-scale cross-lingual language resources for ref- erencing and framing. In Proceedings of the 12th Language Resources and Evaluation Conference, pages 3162-3171, Marseille, France. European Lan- guage Resources Association.", "links": null }, "BIBREF37": { "ref_id": "b37", "title": "Don't annotate, but validate: a datato-text method for capturing event data", "authors": [ { "first": "Piek", "middle": [], "last": "Vossen", "suffix": "" }, { "first": "Filip", "middle": [], "last": "Ilievski", "suffix": "" }, { "first": "Marten", "middle": [], "last": "Postma", "suffix": "" }, { "first": "Roxane", "middle": [], "last": "Segers", "suffix": "" } ], "year": 2018, "venue": "Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Piek Vossen, Filip Ilievski, Marten Postma, and Roxane Segers. 2018. Don't annotate, but validate: a data- to-text method for capturing event data. In Proceed- ings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan. European Language Resources As- sociation (ELRA).", "links": null }, "BIBREF38": { "ref_id": "b38", "title": "Wikidata: a free collaborative knowledgebase", "authors": [ { "first": "Denny", "middle": [], "last": "Vrande\u010di\u0107", "suffix": "" }, { "first": "Markus", "middle": [], "last": "Kr\u00f6tzsch", "suffix": "" } ], "year": 2014, "venue": "Communications of the ACM", "volume": "57", "issue": "10", "pages": "78--85", "other_ids": {}, "num": null, "urls": [], "raw_text": "Denny Vrande\u010di\u0107 and Markus Kr\u00f6tzsch. 2014. Wiki- data: a free collaborative knowledgebase. Commu- nications of the ACM, 57(10):78-85.", "links": null }, "BIBREF39": { "ref_id": "b39", "title": "Referring expressions in English and Japanese: patterns of use in dialogue processing", "authors": [ { "first": "Etsuko", "middle": [], "last": "Yoshida", "suffix": "" } ], "year": 2011, "venue": "", "volume": "208", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Etsuko Yoshida. 2011. Referring expressions in En- glish and Japanese: patterns of use in dialogue pro- cessing, volume 208. John Benjamins Publishing.", "links": null } }, "ref_entries": { "TABREF0": { "num": null, "content": "
week's
b. OneofthefourSUSPI-
CION suspects wanted in last
", "text": "One man DEATH died in a KILLING shooting [...] KILLING murder of Keith Thompson was ARREST arrested [...]", "type_str": "table", "html": null }, "TABREF3": { "num": null, "content": "
The scores were remodeled from (-1,1) to (0,1)
", "text": "The top 10 highest ranked frames (FFICF score) and the 7 bottom ranked frames for the event types music festival (Q868557) and gun violence (Q5618454).", "type_str": "table", "html": null }, "TABREF4": { "num": null, "content": "
precision recallF1 support
1. predicates/unbalanced
day 00.861 0.998 0.9253896
day 8-300.357 0.008 0.016630
Accuracy0.8604526
macro avg0.609 0.503 0.4704526
weighted avg0.791 0.860 0.7984526
2. frames/unbalanced
day 00.891 0.974 0.9313896
day 8-300.627 0.267 0.374630
Accuracy0.8804526
macro avg0.759 0.620 0.6534526
weighted avg0.855 0.876 0.8544526
3. predicates/balanced
day 00.562 0.676 0.614630
day 8-300.594 0.473 0.527630
Accuracy0.5751260
macro avg0.578 0.575 0.5701260
weighted avg0.578 0.575 0.5701260
4. frames/balanced
day 00.746 0.789 0.767630
day 8-300.776 0.732 0.753630
Accuracy0.7601260
macro avg0.761 0.760 0.7601260
weighted avg0.761 0.760 0.7601260
", "text": "The top 10 highest ranked frames (FFICF score)[annotators' score: Y = yes, N = no, D = disagreement] and the 11 bottom ranked frames for the classes \"day 0\" and \"day 8-30\" within the event type gun violence. The scores range between -1 and 1.", "type_str": "table", "html": null }, "TABREF5": { "num": null, "content": "", "text": "Classification reports providing, precision, recall, F1 and support for the performance of the Linear SVM on the test sets of four different experiments: 1. predicate frequencies/unbalanced corpus; 2. predicate frequencies/balanced corpus; 3. frame frequencies/unbalanced corpus; 4. frame frequencies/balanced corpus. Accuracy, macro average and weighted average are also provided per condition.", "type_str": "table", "html": null } } } }