Base_JSON/prefixC/json/C16/C16-1007.json

{
    "paper_id": "C16-1007",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T13:04:43.368962Z"
    },
    "title": "CATENA: CAusal and TEmporal relation extraction from NAtural language texts",
    "authors": [
        {
            "first": "Paramita",
            "middle": [],
            "last": "Mirza",
            "suffix": "",
            "affiliation": {},
            "email": "paramita@mpi-inf.mpg.de"
        },
        {
            "first": "Sara",
            "middle": [],
            "last": "Tonelli",
            "suffix": "",
            "affiliation": {},
            "email": "satonelli@fbk.eu"
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "We present CATENA, a sieve-based system to perform temporal and causal relation extraction and classification from English texts, exploiting the interaction between the temporal and the causal model. We evaluate the performance of each sieve, showing that the rule-based, the machinelearned and the reasoning components all contribute to achieving state-of-the-art performance on TempEval-3 and TimeBank-Dense data. Although causal relations are much sparser than temporal ones, the architecture and the selected features are mostly suitable to serve both tasks. The effects of the interaction between the temporal and the causal components, although limited, yield promising results and confirm the tight connection between the temporal and the causal dimension of texts.",
    "pdf_parse": {
        "paper_id": "C16-1007",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "We present CATENA, a sieve-based system to perform temporal and causal relation extraction and classification from English texts, exploiting the interaction between the temporal and the causal model. We evaluate the performance of each sieve, showing that the rule-based, the machinelearned and the reasoning components all contribute to achieving state-of-the-art performance on TempEval-3 and TimeBank-Dense data. Although causal relations are much sparser than temporal ones, the architecture and the selected features are mostly suitable to serve both tasks. The effects of the interaction between the temporal and the causal components, although limited, yield promising results and confirm the tight connection between the temporal and the causal dimension of texts.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "When the Greek government missed its 1.6 billion euro payment to the IMF as its bailout expired on 30 June 2015, people started to look for information, such as What is going on? Why did it happen and what will happen next? A compact summary that represents the development of a story over time, highlighting not only the temporal connections between events but also cause-effect chains, would be very beneficial for providing information that the readers need. Besides, this kind of knowledge, derived from structured information about events and their temporal-causal relations, could be used in a number of applications, from tools for automated generation of timelines to question answering and decision support systems.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "While temporal relation classification is a well-studied task with a number of systems participating in the TempEval campaigns (Verhagen et al., 2010; UzZaman et al., 2013; Llorens et al., 2015) , less attention has been devoted by the NLP community to the detection of causal links between events. Although recent attempts have tried to settle an annotation standard for causality inspired by TimeML (Mirza et al., 2014) , the interactions between the temporal and the causal dimension of texts have been scarcely explored, especially from an empirical point of view. In this work, we face this challenge by presenting CATENA (CAusal and TEmporal relation extraction from NAtural language texts), 1 a multi-sieve architecture for the extraction and classification of both relation types from English documents, which are pre-annotated with temporal entities, namely events and time expressions.",
                "cite_spans": [
                    {
                        "start": 127,
                        "end": 150,
                        "text": "(Verhagen et al., 2010;",
                        "ref_id": "BIBREF27"
                    },
                    {
                        "start": 151,
                        "end": 172,
                        "text": "UzZaman et al., 2013;",
                        "ref_id": "BIBREF25"
                    },
                    {
                        "start": 173,
                        "end": 194,
                        "text": "Llorens et al., 2015)",
                        "ref_id": "BIBREF15"
                    },
                    {
                        "start": 401,
                        "end": 421,
                        "text": "(Mirza et al., 2014)",
                        "ref_id": "BIBREF18"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Our proposed approach for relation extraction is inspired by recent works on hybrid approaches for temporal relation extraction (D'Souza and Ng, 2013; Chambers et al., 2014) . D'Souza and Ng (2013) introduce 437 hand-coded rules along with supervised classification models using lexical relation, semantic and discourse features. CAEVO, a CAscading EVent Ordering architecture by Chambers et al. (2014) , combines rule-based and data-driven classifiers in a sieve-based architecture for temporal ordering. The classifiers (sieves) are ordered by their individual precision, and transitive closure is applied after each sieve to ensure consistent temporal graph. The problem of detecting causality between events is as challenging as recognizing their temporal order, but less analyzed from an NLP perspective. Besides, previous works have mostly focused on specific types of event pairs and causal expressions in text Do et al., 2011; Riaz and Girju, 2013) . Several works, relying on corpus of parallel temporal and causal relations developed with specific connectives in mind , have presented analyses on the interaction between temporal and causal relations Rink et al., 2010) . Exploiting gold temporal labels as features for the causal relation classifier is shown to be beneficial. Mirza et al. (2014) presented some annotation guidelines to capture explicit causality between event pairs, inspired by TimeML. The resulting corpus, Causal-TimeBank, is then used to build supervised classification models for extracting causal relations (Mirza and Tonelli, 2014a) . None of the above systems presents a hybrid approach in a sieve-based architecture to deal with this task. CATENA is at present the first integrated system available performing temporal and causal relation extraction.",
                "cite_spans": [
                    {
                        "start": 128,
                        "end": 150,
                        "text": "(D'Souza and Ng, 2013;",
                        "ref_id": "BIBREF7"
                    },
                    {
                        "start": 151,
                        "end": 173,
                        "text": "Chambers et al., 2014)",
                        "ref_id": "BIBREF5"
                    },
                    {
                        "start": 176,
                        "end": 197,
                        "text": "D'Souza and Ng (2013)",
                        "ref_id": "BIBREF7"
                    },
                    {
                        "start": 380,
                        "end": 402,
                        "text": "Chambers et al. (2014)",
                        "ref_id": "BIBREF5"
                    },
                    {
                        "start": 918,
                        "end": 934,
                        "text": "Do et al., 2011;",
                        "ref_id": "BIBREF6"
                    },
                    {
                        "start": 935,
                        "end": 956,
                        "text": "Riaz and Girju, 2013)",
                        "ref_id": "BIBREF22"
                    },
                    {
                        "start": 1161,
                        "end": 1179,
                        "text": "Rink et al., 2010)",
                        "ref_id": "BIBREF24"
                    },
                    {
                        "start": 1288,
                        "end": 1307,
                        "text": "Mirza et al. (2014)",
                        "ref_id": "BIBREF18"
                    },
                    {
                        "start": 1542,
                        "end": 1568,
                        "text": "(Mirza and Tonelli, 2014a)",
                        "ref_id": "BIBREF16"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "The CATENA system includes two main classification modules, one for temporal and the other for causal relations between events. As shown in Figure 1 , they both take as input a document annotated with the so-called temporal entities according to TimeML guidelines (Pustejovsky et al., 2003) , including the document creation time (DCT), events and time expressions (timexes). The output is the same document with temporal links (TLINKs) set between pairs of temporal entities, each assigned to one of the TimeML temporal relation types, such as BEFORE, INCLUDES or SIMULTANEOUS, which denotes the temporal ordering. The document is also annotated with causal relations (CLINKs) between event pairs.",
                "cite_spans": [
                    {
                        "start": 264,
                        "end": 290,
                        "text": "(Pustejovsky et al., 2003)",
                        "ref_id": "BIBREF20"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 140,
                        "end": 148,
                        "text": "Figure 1",
                        "ref_id": "FIGREF0"
                    }
                ],
                "eq_spans": [],
                "section": "System architecture",
                "sec_num": "3"
            },
            {
                "text": "The modules for temporal and causal relation classification rely both on a sieve-based architecture, in which the remaining unlabelled pairs -after running a rule-based component and/or a transitive reasonerare fed into a supervised classifier. Although some steps can be run in parallel, the two modules interact, based on the assumption that the notion of causality is tightly connected with the temporal dimension and that information from one module can be used to improve or check the consistency of the other. In particular, (i) TLINK labels for event-event (E-E) pairs, resulting from the rule-based sieve + temporal reasoner modules, are used as features for the CLINK classifier; and (ii) CLINK labels (i.e. CLINK and CLINK-R) are used as a post-editing method for correcting the wrong labelled event pairs by the TLINK classifier. This step relies on a set of rules based on the temporal constraint of causality, i.e. (i) CLINK(e 1 , e 2 ) \u2192 BEFORE(e 1 , e 2 ) and (ii) CLINK-R(e 1 , e 2 ) \u2192 AFTER(e 1 , e 2 ). The modules for temporal and causal relation extraction are detailed in Section 4 and 5 respectively.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "System architecture",
                "sec_num": "3"
            },
            {
                "text": "The module for the extraction of temporal relations contains two main components, one for (i) temporal relation identification, which is based on a set of rules, and the other for (ii) temporal relation type classification, which is a combination of rule-based and supervised classification modules, with a temporal reasoning component in between. The three steps for temporal relation type classification are ordered based on their individual precisions. This mechanism allows the system to first label few links with high precision using rules, then to infer new links through the reasoner, and finally to increase recall through supervised classification, based on the output of the previous steps.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Relation Extraction System",
                "sec_num": "4"
            },
            {
                "text": "All pairs of temporal entities satisfying one of the following rules, inspired by the TempEval-3 task description, are considered as having temporal links (TLINKs): (i) two main events of consecutive sentences, (ii) two events in the same sentence, (iii) an event and a timex in the same sentence, (iv) an event and a document creation time and (v) pairs of all possible timexes (including document creation time) linked with each other. 2 These pairs are then grouped together into four different groups: timex-timex (T-T), event-DCT (E-D), event-timex (E-T) and event-event (E-E).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Relation Identification",
                "sec_num": "4.1"
            },
            {
                "text": "Our sieve-based architecture is inspired by CAEVO (Chambers et al., 2014 ), although we significantly reduce the system complexity as follows:",
                "cite_spans": [
                    {
                        "start": 50,
                        "end": 72,
                        "text": "(Chambers et al., 2014",
                        "ref_id": "BIBREF5"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Relation Type Classification",
                "sec_num": "4.2"
            },
            {
                "text": "\u2022 We merge all rule-based classifiers into one sieve component (rule-based sieve), and all Support Vector Machine (SVM) classifiers in the machine-learned sieve.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Relation Type Classification",
                "sec_num": "4.2"
            },
            {
                "text": "\u2022 Instead of running transitive inference after each classifier, we run our temporal reasoner module on the output of the rule-based sieve, only once.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Relation Type Classification",
                "sec_num": "4.2"
            },
            {
                "text": "Furthermore, we use the output of the rule-based sieve (Section 4.2.1) as features for the machinelearned sieve (Section 4.2.3), specifically: (i) the timex-DCT link label proposed by the timex-timex rules are used as a feature in the event-timex SVM, and (ii) the event-DCT link label proposed by the event-DCT rules are used as a feature in the event-event SVM.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Relation Type Classification",
                "sec_num": "4.2"
            },
            {
                "text": "The temporal rule-based sieve relies on specific hand-crafted rules designed for each type of temporal entity pairs, and takes as input the entity pairs identified in the previous step.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Rule-Based Sieve",
                "sec_num": "4.2.1"
            },
            {
                "text": "Timex-timex Rules For timex-timex relations, we take into account temporal expressions of types DATE and TIME, and determine the relation types based on their normalized values. For example, \"7 PM tonight\" (2015-12-12T19:00) IS INCLUDED in \"today\" (2015-12-12).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Rule-Based Sieve",
                "sec_num": "4.2.1"
            },
            {
                "text": "Event-DCT Rules The rules for labelling E-D pairs are based on the tense and/or aspect of the event word. For example, for the event mention \"(had) fallen\", which is in the past tense with perfective aspect, its relation with the DCT is labelled as BEFORE.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Rule-Based Sieve",
                "sec_num": "4.2.1"
            },
            {
                "text": "As for E-T pairs, we build a set of rules based on the temporal senses of some prepositions (Litkowski and Hargraves, 2006; Litkowski, 2014) . 3 In particular we assign a label whenever a temporal preposition establishes a dependency path between an event (E) and a timex (T), in which T acts as the temporal modifier of E. For example, if T is introduced by a temporal prepositions expressing a STARTTIME sense such as from or since, the relation is labelled as BEGUN BY.",
                "cite_spans": [
                    {
                        "start": 92,
                        "end": 123,
                        "text": "(Litkowski and Hargraves, 2006;",
                        "ref_id": "BIBREF13"
                    },
                    {
                        "start": 124,
                        "end": 140,
                        "text": "Litkowski, 2014)",
                        "ref_id": "BIBREF14"
                    },
                    {
                        "start": 143,
                        "end": 144,
                        "text": "3",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Event-timex Rules",
                "sec_num": null
            },
            {
                "text": "In the absence of a temporal preposition, T might simply be a temporal modifier of E, as exemplified in \"Police [confirmed] E [Friday] T that the body was found...\". In this case, we assume that the E-T label is IS INCLUDED. Moreover, sometimes events are modified by temporal expressions marking the starting time and ending time in a duration pattern such as 'between TBEGIN and TEND' or 'from TBEGIN to/until TEND'. We define additional rules as follows: (i) If T matches TBEGIN then E-T label is BEGUN BY, and (ii) if T matches TEND then E-T label is ENDED BY.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Event-timex Rules",
                "sec_num": null
            },
            {
                "text": "Event-event Rules E-E pairs are finally labelled following two sets of rules. The first set is based on the dependency path possibly existing between the first (e 1 ) and the second event (e 2 ), and the verb information encoded in e 1 . For example, if e 2 is the logical subject of e 1 as in \"...the chain reaction [touched] e 1 off by the [collapse] e 2 of Lehman Brothers\", e 1 and e 2 are connected by an AFTER relation. The other set of rules is taken from CAEVO, including: (i) rules for linking a reporting event and another event syntactically dominated by the first, based on tense and aspect; and (ii) rules based on the role played by various tenses of English verbs in conveying temporal discourse (Reichenbach, 1947) .",
                "cite_spans": [
                    {
                        "start": 711,
                        "end": 730,
                        "text": "(Reichenbach, 1947)",
                        "ref_id": "BIBREF21"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Event-timex Rules",
                "sec_num": null
            },
            {
                "text": "Further details on the implemented rules for the temporal rule-based sieve can be found in Appendix A.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Event-timex Rules",
                "sec_num": null
            },
            {
                "text": "Based on the output of the previous sieve, we run a transitive reasoner layer, similar to CAEVO, in order to infer new temporal links among candidate pairs. This alleviates the issue of high precision and low recall, typical of the rule-based sieve.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Reasoner",
                "sec_num": "4.2.2"
            },
            {
                "text": "An annotated TimeML document can be mapped into a constraint problem according to how TLINKs are mapped into Allen relations (Allen, 1983) . We apply the following mapping: Once the documents are mapped into constraint problems, they are then processed by an automated temporal reasoner for computing their deductive closure, globally reasoning on them. We rely on the Generic Qualitative Reasoner (GQR) (Westphal et al., 2010) , a fast solver for generic qualitative constraint problems, such as Allen constraint problems. The rationale of preferring GQR to other solutions, such as fast Boolean Satisfiability Problem (SAT) solvers, is due to its scalability, simplicity of use and efficient performances (Westphal and W\u00f6lfl, 2009) .",
                "cite_spans": [
                    {
                        "start": 125,
                        "end": 138,
                        "text": "(Allen, 1983)",
                        "ref_id": "BIBREF0"
                    },
                    {
                        "start": 404,
                        "end": 427,
                        "text": "(Westphal et al., 2010)",
                        "ref_id": "BIBREF29"
                    },
                    {
                        "start": 707,
                        "end": 733,
                        "text": "(Westphal and W\u00f6lfl, 2009)",
                        "ref_id": "BIBREF28"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Reasoner",
                "sec_num": "4.2.2"
            },
            {
                "text": "\u2022 <",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Reasoner",
                "sec_num": "4.2.2"
            },
            {
                "text": "We build three supervised classification models, one for event-DCT (E-D), one for event-timex (E-T) and one for event-event (E-E) pairs. We use LIBLINEAR (Fan et al., 2008) L2-loss linear SVM (default parameters), and one-vs-rest strategy for multi-class classification.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Supervised Classifiers",
                "sec_num": "4.2.3"
            },
            {
                "text": "Tools and Resources Several external tools and resources are used to extract features from each temporal entity pair, including:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Supervised Classifiers",
                "sec_num": "4.2.3"
            },
            {
                "text": "\u2022 MorphoPro (Pianta et al., 2008) , to get PoS tags and phrase chunk for each token.",
                "cite_spans": [
                    {
                        "start": 12,
                        "end": 33,
                        "text": "(Pianta et al., 2008)",
                        "ref_id": "BIBREF19"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Supervised Classifiers",
                "sec_num": "4.2.3"
            },
            {
                "text": "\u2022 Mate tools (Bjorkelund et al., 2010) to extract the dependency path between words.",
                "cite_spans": [
                    {
                        "start": 13,
                        "end": 38,
                        "text": "(Bjorkelund et al., 2010)",
                        "ref_id": "BIBREF4"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Supervised Classifiers",
                "sec_num": "4.2.3"
            },
            {
                "text": "\u2022 WordNet similarity module 4 to compute semantic similarity (Lin, 1998) between words.",
                "cite_spans": [
                    {
                        "start": 61,
                        "end": 72,
                        "text": "(Lin, 1998)",
                        "ref_id": "BIBREF12"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Supervised Classifiers",
                "sec_num": "4.2.3"
            },
            {
                "text": "\u2022 Temporal signal lists from Mirza and Tonelli (2014b), further expanded using the Paraphrase Database (Ganitkevitch et al., 2013) , and manually clustered e.g. {before, prior to, in advance of }.",
                "cite_spans": [
                    {
                        "start": 103,
                        "end": 130,
                        "text": "(Ganitkevitch et al., 2013)",
                        "ref_id": "BIBREF9"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Supervised Classifiers",
                "sec_num": "4.2.3"
            },
            {
                "text": "Feature Set We implemented a set of features, listed in Table 1 , largely inspired by the best performing systems in TempEval-2 (Verhagen et al., 2010) and TempEval-3 (UzZaman et al., 2013) campaigns. We simplified the possible values of some features as follows: \u2022 dependencyPath We only consider a dependency path between an event pair if it describes coordination, subordination, subject or object relation. \u2022 signalTokens Given a temporal signal, we do not include in the feature set the token but the clusterID of the cluster containing synonymous signals, e.g. {before, prior to, in advance of }. \u2022 wnSim The value of WordNet similarity measure is discretized as follows: sim \u2264 0.0, 0.0 < sim \u2264 0.5, 0.5 < sim \u2264 1.0 and sim > 1.0.",
                "cite_spans": [
                    {
                        "start": 128,
                        "end": 151,
                        "text": "(Verhagen et al., 2010)",
                        "ref_id": "BIBREF27"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 56,
                        "end": 63,
                        "text": "Table 1",
                        "ref_id": "TABREF2"
                    }
                ],
                "eq_spans": [],
                "section": "Temporal Supervised Classifiers",
                "sec_num": "4.2.3"
            },
            {
                "text": "Feature TLINK CLINK Rep. Description E-D E-T E-E E-E",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Supervised Classifiers",
                "sec_num": "4.2.3"
            },
            {
                "text": "We exclude lexical features such as token/lemma of temporal entities from the feature set in order to increase the classifiers' robustness in dealing with completely new texts with different vocabularies. Instead, we include WordNet similarity in the feature set to capture the semantic relations between event words.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Supervised Classifiers",
                "sec_num": "4.2.3"
            },
            {
                "text": "Label Simplification For training the classification models, we only consider 10 out of the 14 relation types defined in TimeML by collapsing some types, i.e., IBEFORE into BEFORE, IAFTER into AFTER, DURING and DURING INV into SIMULTANEOUS, due to the sparse annotation of such labels in the datasets.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal Supervised Classifiers",
                "sec_num": "4.2.3"
            },
            {
                "text": "We propose the same hybrid approach combining rule-based and supervised classifiers for the identification of causal relations. However, while temporal order has a clear formalization in the NLP community, capturing causal relationships in natural language text is more challenging, for they can be expressed by different syntactic and semantic features and involve both situation-specific information and world knowledge. We adopt the notion of causality proposed in the annotation guidelines of the Causal-TimeBank (Mirza et al., 2014; Mirza and Tonelli, 2014a) , which accounts for CAUSE, ENABLE and PREVENT phenomena (Wolff, 2007; Wolff and Song, 2003) that are overtly expressed in text. In particular, we aim at assigning a causal link to pairs of events when: (i) the causal relation is expressed by affect, link and causative verbs (CAUSE-, ENABLE-and PREVENT-type verbs), hereinafter simply addressed as causal verbs; or (ii) the causal relation is marked by a causal signal (see e.g. footnote 6). The two cases require different algorithms: while causal constructions containing causal verbs are quite straightforward to identify, causal signals are very ambiguous and can appear in different syntactic constructions. 6 Therefore, we tackle the first through a rule-based approach, while the second is best covered via supervision, taking advantage of the freely available Causal-TimeBank.",
                "cite_spans": [
                    {
                        "start": 517,
                        "end": 537,
                        "text": "(Mirza et al., 2014;",
                        "ref_id": "BIBREF18"
                    },
                    {
                        "start": 538,
                        "end": 563,
                        "text": "Mirza and Tonelli, 2014a)",
                        "ref_id": "BIBREF16"
                    },
                    {
                        "start": 621,
                        "end": 634,
                        "text": "(Wolff, 2007;",
                        "ref_id": "BIBREF31"
                    },
                    {
                        "start": 635,
                        "end": 656,
                        "text": "Wolff and Song, 2003)",
                        "ref_id": "BIBREF30"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Causal Relation Extraction System",
                "sec_num": "5"
            },
            {
                "text": "Similar to the temporal processing module, the first step towards causal relation classification is the identification of candidate event pairs. Given a document already annotated with events, we take into account every possible combination of events in a sentence in a forward manner as candidate event pairs. For example, if we have a sentence \"e 1 , triggered by e 2 , cause them to e 3 ,\" the candidate event pairs are (e 1 ,e 2 ), (e 1 ,e 3 ) and (e 2 ,e 3 ). We also include as candidate event pairs the combination of each event in a sentence with events in the following one, to account for inter-sentential causality, under the simplifying assumption that causality may be expressed also between events in two consecutive sentences.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Causal Relation Identification",
                "sec_num": "5.1"
            },
            {
                "text": "In the rule-based sieve, we classify causal constructions containing causal verbs. These show strong regularities: given a causal verb v, the first event e 1 is usually the subject of v and the second event e 2 is either the object or the predicative complement of v. Such relations between events and causal verbs are usually syntactically expressed, therefore our rules aim at identifying pairs of events being related to a causal verb in a causal construction by looking at their dependency paths.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Causal Rule-Based Sieve",
                "sec_num": "5.2"
            },
            {
                "text": "We take the list of 56 affect, link and causative verbs presented in Mirza et al. (2014) as the causal verb list. We further expand the list using the Paraphrase Database (Ganitkevitch et al., 2013) and original verbs as seeds, resulting in a total of 97 verbs. We then manually cluster the causal verbs sharing the same syntactic behaviour in groups and define a set of rules for each verb group, taking into account the possible existing dependency paths between v and e 1 /e 2 , as well as the causal direction sense 7 conveyed in v. Further details on the implemented rules for the causal rule-based sieve can be found in Appendix B.",
                "cite_spans": [
                    {
                        "start": 69,
                        "end": 88,
                        "text": "Mirza et al. (2014)",
                        "ref_id": "BIBREF18"
                    },
                    {
                        "start": 171,
                        "end": 198,
                        "text": "(Ganitkevitch et al., 2013)",
                        "ref_id": "BIBREF9"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Causal Rule-Based Sieve",
                "sec_num": "5.2"
            },
            {
                "text": "In order to recognize and determine the causal direction of CLINKs that are signalled by a causal signal, we adopt a supervised approach. We build a classification model using LIBLINEAR (Fan et al., 2008) L2-loss linear SVM (default parameters), and one-vs-rest strategy for multi-class classification. The classifier has to label an event pair (e 1 , e 2 ) with CLINK, CLINK-R or O for others.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Causal Supervised Classifier",
                "sec_num": "5.3"
            },
            {
                "text": "We take as candidate event pairs only those in which the causal signal is connected via dependency path to either e 1 or e 2 , or both. Besides, we exclude event pairs where the two events are directly connected through relations such as subject, object, coordinating or locative adverbial, because a causal relation usually does not hold in these cases.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Causal Supervised Classifier",
                "sec_num": "5.3"
            },
            {
                "text": "The same external tools and resources mentioned in Section 4.2.3 for building the temporal classifiers are used to extract features from each event pair. Additionally, we take the list of causal signals from the annotation guidelines presented in Mirza et al. (2014) as the causal signal list. Again we expand the list using the Paraphrase Database (Ganitkevitch et al., 2013) , resulting in a total of 200 signals. We also manually cluster some signals together, e.g. {therefore, thereby, hence, consequently}, as we did for temporal signals.",
                "cite_spans": [
                    {
                        "start": 247,
                        "end": 266,
                        "text": "Mirza et al. (2014)",
                        "ref_id": "BIBREF18"
                    },
                    {
                        "start": 349,
                        "end": 376,
                        "text": "(Ganitkevitch et al., 2013)",
                        "ref_id": "BIBREF9"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Tools and Resources",
                "sec_num": null
            },
            {
                "text": "Feature Set The implemented features are listed in Table 1 . As shown in Figure 1 , the event-event labels added by the rule-based sieve and the reasoner in the temporal relation extraction module are also used as features for the CLINK classifier.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 51,
                        "end": 58,
                        "text": "Table 1",
                        "ref_id": "TABREF2"
                    },
                    {
                        "start": 73,
                        "end": 81,
                        "text": "Figure 1",
                        "ref_id": "FIGREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Tools and Resources",
                "sec_num": null
            },
            {
                "text": "The purpose of the evaluation is two-fold: (i) to evaluate the quality of extracted temporal and causal links separately; and (ii) to investigate the interaction between temporal and causal relation extraction systems in the integrated architecture.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation",
                "sec_num": "6"
            },
            {
                "text": "We perform two evaluations, one following TempEval-3 and the other TimeBank-Dense evaluation methodology.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal and Causal Relation evaluation",
                "sec_num": "6.1"
            },
            {
                "text": "Dataset For the evaluation of the temporal relation extraction module following TempEval-3, we use the same training and test data released for the shared task, 8 i.e. TBAQ-cleaned (cleaned and improved version of the TimeBank 1.2 and the AQUAINT corpora) and TempEval-3-platinum, respectively. The TimeBank 1.2 corpus contains 183 documents coming from a variety of news report, specifically from the ACE program and PropBank, while the AQUAINT corpus contains 73 news report documents and often referred to as the Opinion corpus. The TempEval-3-platinum corpus, containing 20 news articles, was annotated/reviewed by the TempEval-3 organizers.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal and Causal Relation evaluation",
                "sec_num": "6.1"
            },
            {
                "text": "The TimeBank-Dense corpus (Chambers et al., 2014) is created to address the sparsity issue in the existing TimeML corpora. The resulting corpus contains 12,715 temporal relations over 36 documents taken from TimeBank 1.2. For the TimeBank-Dense evaluation, we follow the experimental setup in Chambers et al. (2014) , in which the TimeBank-Dense corpus is split into a 22 document training set, a 5 document development set and a 9 document test set. 9 To evaluate the causal relation extraction module, we use the Causal-TimeBank corpus 10 (Mirza and Tonelli, 2014a) for training. For TimeBank-Dense evaluation, the test set is a subset of TimeBank, so we exclude the 9 test documents from Causal-TimeBank during training. For TempEval-3 evaluation, we manually annotated 20 TempEval-3-platinum documents with causal links following the annotation guidelines of the Causal-TimeBank. 11 Causal relations are much sparser than temporal ones, and we found only 26 CLINKs.",
                "cite_spans": [
                    {
                        "start": 26,
                        "end": 49,
                        "text": "(Chambers et al., 2014)",
                        "ref_id": "BIBREF5"
                    },
                    {
                        "start": 293,
                        "end": 315,
                        "text": "Chambers et al. (2014)",
                        "ref_id": "BIBREF5"
                    },
                    {
                        "start": 451,
                        "end": 452,
                        "text": "9",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Temporal and Causal Relation evaluation",
                "sec_num": "6.1"
            },
            {
                "text": "Label Adjustment Since the set of TLINK types used in the TimeBank-Dense corpus is slightly different from the one used in TempEval-3, 12 we map the relation types of TLINKs labelled by the rule-based sieve of CATENA (Section 4.2.1) as follows: (i) BEGINS, ENDED BY \u2192 BEFORE, (ii) BEGUN BY, ENDS \u2192 AFTER, and (iii) DURING, IDENTITY \u2192 SIMULTANEOUS. The set of labels for the TLINK classifiers (Section 4.2.3) is also adjusted accordingly following the labels in the TimeBank-Dense training data. Table 2 , we compare the performance of CATENA with the two best-performing systems participating in the Task C of TempEval-3 (relation annotation given gold entities) and Task C 'relation type only' (relation annotation given gold entities and related pairs). We also compare the results on the second task with the results of Laokulrat et al. (2015) , who recently presented a state-of-the-art system for relation classification based on timegraphs and stacked learning. In CATENA, Task C 'relation type only' is performed by disabling the module for identifying temporal links described in Section 4.1.",
                "cite_spans": [
                    {
                        "start": 823,
                        "end": 846,
                        "text": "Laokulrat et al. (2015)",
                        "ref_id": "BIBREF11"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 495,
                        "end": 502,
                        "text": "Table 2",
                        "ref_id": "TABREF4"
                    }
                ],
                "eq_spans": [],
                "section": "Temporal and Causal Relation evaluation",
                "sec_num": "6.1"
            },
            {
                "text": "The evaluation shows that CATENA is the best performing system in both tasks, even if in Task C best precision and best recall are yielded by Bethard (2013) and Laokulrat et al. (2013) , respectively. The recall drop (from .613 to .595) in Task C is because we remove the timex-timex pairs from the final annotated documents in order to avoid a relevant decrease in precision, since only very few of such pairs are annotated in the gold standard. The significant drop in precision shows the difficulty in matching annotators' decision to set TLINKs between entity pairs, although CATENA implements the instructions they had to follow in the annotation guidelines. We also report in Table 2 the performance of CATENA in the TimeBank-Dense evaluation and compare it with CAEVO. We report only F1-score, since all possible links are labelled, yielding the same P and R values. We achieve a small improvement in the overall F1-score, i.e., .511 vs .507. If we consider the different entity pairs, CATENA performs best on timex-timex and event-timex relations, while CAEVO still achieves the best results on event-DCT and event-event pairs. One of the possible reasons for that is the lack of rules in CATENA to classify VAGUE TLINKs between E-E pairs, a relation type present only in TimeBank-Dense.",
                "cite_spans": [
                    {
                        "start": 142,
                        "end": 156,
                        "text": "Bethard (2013)",
                        "ref_id": "BIBREF3"
                    },
                    {
                        "start": 161,
                        "end": 184,
                        "text": "Laokulrat et al. (2013)",
                        "ref_id": "BIBREF10"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 682,
                        "end": 689,
                        "text": "Table 2",
                        "ref_id": "TABREF4"
                    }
                ],
                "eq_spans": [],
                "section": "Evaluation Results In",
                "sec_num": null
            },
            {
                "text": "In order to measure the contribution of each component to the overall performance of CATENA, we also evaluate the performance of each sieve both in the temporal and in the causal module. Results are reported in Table 3 , evaluated on both TempEval-3 and TimeBank-Dense test data. As expected, running a transitive closure module after the temporal rule-based sieve (RB + TR) results in improving recall, but the overall performance is still lacking (less than .30 F1-score).",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 211,
                        "end": 218,
                        "text": "Table 3",
                        "ref_id": "TABREF5"
                    }
                ],
                "eq_spans": [],
                "section": "Evaluation Results In",
                "sec_num": null
            },
            {
                "text": "Combining rule-based and machine-learned sieves (RB + ML) yields a slight improvement compared with enabling only the machine-learned sieve in the system (ML). Introducing the temporal reasoner module between the two sieves (RB + TR + ML) proves to be even more beneficial. This is especially evident in the TimeBank-Dense evaluation. The same phenomena are also observed by CAEVO; Table 3 (right) shows the related numbers reported in Chambers et al. (2014) . Note that in CAEVO, the machinelearned sieves are not the last sieves, instead, the AllVague sieve is finally activated to label all remaining unlabelled pairs as VAGUE.",
                "cite_spans": [
                    {
                        "start": 436,
                        "end": 458,
                        "text": "Chambers et al. (2014)",
                        "ref_id": "BIBREF5"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 382,
                        "end": 389,
                        "text": "Table 3",
                        "ref_id": "TABREF5"
                    }
                ],
                "eq_spans": [],
                "section": "Evaluation Results In",
                "sec_num": null
            },
            {
                "text": "For causal relation extraction, the combination of rule-based and machine-learned sieves (RB + ML) achieves .622 F1-score in TempEval-3 evaluation, with the ML component contributing to increase ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation Results In",
                "sec_num": null
            },
            {
                "text": "As shown in Figure 1 , E-E labels returned by the temporal reasoner are used by the CLINK classifier as features, whose causal relations are then used to post-edit TLINK labels. We evaluate the impact of the first step through an ablation test, by removing TLINK types from the features used by the CLINK classifier. We only analyse the results of TempEval-3 evaluation, since there are no causal links recognized in the TimeBank-Dense test corpus. Without TLINK types, the F1-score drops from .622 to .571, with a significant recall drop from .538 to .462. This shows that temporal information is beneficial to the classification of causal relations between events, especially in terms of recall.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 12,
                        "end": 20,
                        "text": "Figure 1",
                        "ref_id": "FIGREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Interaction between Temporal and Causal Relations",
                "sec_num": "6.2"
            },
            {
                "text": "As for the evaluation of TLINK post-editing using CLINKs, the system identifies 19 causal links in the test set, which are passed to the temporal module. While 15 of them are already consistent with BEFORE/AFTER labels, 3 would add new correct TLINKs that are currently not annotated in the evaluation corpus, and were wrongly labelled by the temporal module of CATENA, as shown in Table 4 . The fourth would add a BEFORE relation between cloaked and coughing in \"A haze akin to volcanic fumes [cloaked] S the capital, causing convulsive [coughing] T ...\". This relation is labelled as INCLUDES in the gold standard, but we believe that BEFORE would be correct as well.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 382,
                        "end": 389,
                        "text": "Table 4",
                        "ref_id": "TABREF7"
                    }
                ],
                "eq_spans": [],
                "section": "Interaction between Temporal and Causal Relations",
                "sec_num": "6.2"
            },
            {
                "text": "We presented CATENA, a hybrid system for the extraction and classification of temporal and causal relations in text, which we make freely available to the research community. We adopt a sieve-based architecture both for the temporal and the causal module, integrating rule-based and machine learning components. The two modules were evaluated separately, showing that they achieve state-of-the-art performance on different tasks. Furthermore, the interaction between temporal and causal components, especially the benefits of passing information from one module to the other, was also analysed.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusions",
                "sec_num": "7"
            },
            {
                "text": "The system relies on the notion of events as defined in the TimeML standard, making it possible to easily put temporal and causal information in relation. Although the interplay between causality and temporality may seem obvious from a theoretical point of view, CATENA allows a systematic study and a quantification of this phenomenon. The presented approach would probably have more impact if implicit causality was also considered, which we did not take into account because it is not annotated in the Causal-TimeBank corpus. However, we plan to investigate this issue in the near future. Table 7 : E-E label rules based on dependency path (dep) and verb information of e 1 (e 1 verb info). Table 8 : Causal verb rules for E-E pairs based on causal verb (v) category, dependency paths between v and e 1 /e 2 , and causal direction sense (dir). (*) denotes all possible dependency paths listed in Table 9 . : Dependency paths considered for setting a causal link between two events e 1 and e 2 when a causal verb v is present.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 592,
                        "end": 599,
                        "text": "Table 7",
                        "ref_id": null
                    },
                    {
                        "start": 694,
                        "end": 701,
                        "text": "Table 8",
                        "ref_id": null
                    },
                    {
                        "start": 899,
                        "end": 906,
                        "text": "Table 9",
                        "ref_id": "TABREF12"
                    }
                ],
                "eq_spans": [],
                "section": "Conclusions",
                "sec_num": "7"
            },
            {
                "text": "Note that this is not included in the enumerated possible TLINKs in the TempEval-3 task description.3 We took the list of temporal prepositions from http://www.clres.com/db/classes/ClassTemporal.php.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "http://ws4jdemo.appspot.com/",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "The order of e1 and e2 in E-E pairs is always according to the appearance order in the text, while in E-T pairs, e2 is always a timex regardless of the appearance order.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "\"The building[collapsed] T because of the [earthquake] S \" vs \"Because of the [earthquake] S the building [collapsed] T \". S and T denote the source (cause) and target (effect) of the causal relation.7 For example, result in and result from have different senses affecting the causal direction, i.e. the causing event is the subject of result in and the object of result from.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "Available at https://www.cs.york.ac.uk/semeval-2013/task1/index.php\\%3Fid=data.html. 9 Available at http://www.usna.edu/Users/cs/nchamber/caevo/. 10 Available at http://hlt-nlp.fbk.eu/technologies/causal-timebank. 11 Available at https://github.com/paramitamirza/CATENA/data/.12 Some relation types are not used, and the VAGUE relation introduced in the first TempEval task(Verhagen et al., 2007) is adopted to cope with ambiguous temporal relations, or to indicate pairs for which no clear temporal relation exists. The final set of TLINK types in TimeBank-Dense includes: BEFORE, AFTER, INCLUDES, IS INCLUDED, SIMULTANEOUS and VAGUE.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            }
        ],
        "back_matter": [
            {
                "text": "The research leading to this paper was partially supported by the European Union's 7th Framework Programme via the NewsReader Project (ICT-316404).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Acknowledgments",
                "sec_num": null
            }
        ],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "Maintaining knowledge about temporal intervals",
                "authors": [
                    {
                        "first": "James",
                        "middle": [
                            "F"
                        ],
                        "last": "Allen",
                        "suffix": ""
                    }
                ],
                "year": 1983,
                "venue": "Commun. ACM",
                "volume": "26",
                "issue": "11",
                "pages": "832--843",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "James F. Allen. 1983. Maintaining knowledge about temporal intervals. Commun. ACM, 26(11):832-843, Novem- ber.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "Learning semantic links from a corpus of parallel temporal and causal relations",
                "authors": [
                    {
                        "first": "Steven",
                        "middle": [],
                        "last": "Bethard",
                        "suffix": ""
                    },
                    {
                        "first": "James",
                        "middle": [
                            "H"
                        ],
                        "last": "Martin",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "Proceedings of ACL-08: HLT, Short Papers",
                "volume": "",
                "issue": "",
                "pages": "177--180",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Steven Bethard and James H. Martin. 2008. Learning semantic links from a corpus of parallel temporal and causal relations. In Proceedings of ACL-08: HLT, Short Papers, pages 177-180, Columbus, Ohio, June. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "Building a corpus of temporalcausal structure",
                "authors": [
                    {
                        "first": "Steven",
                        "middle": [],
                        "last": "Bethard",
                        "suffix": ""
                    },
                    {
                        "first": "William",
                        "middle": [],
                        "last": "Corvey",
                        "suffix": ""
                    },
                    {
                        "first": "Sara",
                        "middle": [],
                        "last": "Klingenstein",
                        "suffix": ""
                    },
                    {
                        "first": "James",
                        "middle": [
                            "H"
                        ],
                        "last": "Martin",
                        "suffix": ""
                    },
                    {
                        "first": ";",
                        "middle": [],
                        "last": "",
                        "suffix": ""
                    },
                    {
                        "first": "Khalid",
                        "middle": [],
                        "last": "Choukri",
                        "suffix": ""
                    },
                    {
                        "first": "Bente",
                        "middle": [],
                        "last": "Maegaard",
                        "suffix": ""
                    },
                    {
                        "first": "Joseph",
                        "middle": [],
                        "last": "Mariani",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "Proceedings of the Sixth International Conference on Language Resources and Evaluation (LREC'08)",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Steven Bethard, William Corvey, Sara Klingenstein, and James H. Martin. 2008. Building a corpus of temporal- causal structure. In Nicoletta Calzolari (Conference Chair), Khalid Choukri, Bente Maegaard, Joseph Mariani, Jan Odijk, Stelios Piperidis, and Daniel Tapias, editors, Proceedings of the Sixth International Conference on Language Resources and Evaluation (LREC'08), Marrakech, Morocco, may. European Language Resources Association (ELRA). http://www.lrec-conf.org/proceedings/lrec2008/.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Cleartk-timeml: A minimalist approach to tempeval 2013",
                "authors": [
                    {
                        "first": "Steven",
                        "middle": [],
                        "last": "Bethard",
                        "suffix": ""
                    }
                ],
                "year": 2013,
                "venue": "Proceedings of the Seventh International Workshop on Semantic Evaluation, SemEval '13",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Steven Bethard. 2013. Cleartk-timeml: A minimalist approach to tempeval 2013. In Proceedings of the Seventh International Workshop on Semantic Evaluation, SemEval '13, Atlanta, Georgia, USA, June. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "A high-performance syntactic and semantic dependency parser",
                "authors": [
                    {
                        "first": "Anders",
                        "middle": [],
                        "last": "Bjorkelund",
                        "suffix": ""
                    },
                    {
                        "first": "Bernd",
                        "middle": [],
                        "last": "Bohnet",
                        "suffix": ""
                    },
                    {
                        "first": "Love",
                        "middle": [],
                        "last": "Hafdell",
                        "suffix": ""
                    },
                    {
                        "first": "Pierre",
                        "middle": [],
                        "last": "Nugues",
                        "suffix": ""
                    }
                ],
                "year": 2010,
                "venue": "Coling 2010: Demonstrations",
                "volume": "",
                "issue": "",
                "pages": "33--36",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Anders Bjorkelund, Bernd Bohnet, Love Hafdell, and Pierre Nugues. 2010. A high-performance syntactic and semantic dependency parser. In Coling 2010: Demonstrations, pages 33-36, Beijing, China, August. Coling 2010 Organizing Committee.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "Dense event ordering with a multi-pass architecture",
                "authors": [
                    {
                        "first": "Nathanael",
                        "middle": [],
                        "last": "Chambers",
                        "suffix": ""
                    },
                    {
                        "first": "Taylor",
                        "middle": [],
                        "last": "Cassidy",
                        "suffix": ""
                    },
                    {
                        "first": "Bill",
                        "middle": [],
                        "last": "Mcdowell",
                        "suffix": ""
                    },
                    {
                        "first": "Steven",
                        "middle": [],
                        "last": "Bethard",
                        "suffix": ""
                    }
                ],
                "year": 2014,
                "venue": "Transactions of the Association for Computational Linguistics",
                "volume": "2",
                "issue": "",
                "pages": "273--284",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Nathanael Chambers, Taylor Cassidy, Bill McDowell, and Steven Bethard. 2014. Dense event ordering with a multi-pass architecture. Transactions of the Association for Computational Linguistics, 2:273-284.",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "Minimally supervised event causality identification",
                "authors": [
                    {
                        "first": "Quang",
                        "middle": [],
                        "last": "Do",
                        "suffix": ""
                    },
                    {
                        "first": "Yee",
                        "middle": [],
                        "last": "Seng Chan",
                        "suffix": ""
                    },
                    {
                        "first": "Dan",
                        "middle": [],
                        "last": "Roth",
                        "suffix": ""
                    }
                ],
                "year": 2011,
                "venue": "Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing",
                "volume": "",
                "issue": "",
                "pages": "294--303",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Quang Do, Yee Seng Chan, and Dan Roth. 2011. Minimally supervised event causality identification. In Proceed- ings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 294-303, Edinburgh, Scotland, UK., July. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Classifying temporal relations with rich linguistic knowledge",
                "authors": [
                    {
                        "first": "D'",
                        "middle": [],
                        "last": "Jennifer",
                        "suffix": ""
                    },
                    {
                        "first": "Vincent",
                        "middle": [],
                        "last": "Souza",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Ng",
                        "suffix": ""
                    }
                ],
                "year": 2013,
                "venue": "Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
                "volume": "",
                "issue": "",
                "pages": "918--927",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Jennifer D'Souza and Vincent Ng. 2013. Classifying temporal relations with rich linguistic knowledge. In Pro- ceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguis- tics: Human Language Technologies, pages 918-927, Atlanta, Georgia, June. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "LIBLINEAR: A library for large linear classification",
                "authors": [
                    {
                        "first": "Kai-Wei",
                        "middle": [],
                        "last": "Rong-En Fan",
                        "suffix": ""
                    },
                    {
                        "first": "Cho-Jui",
                        "middle": [],
                        "last": "Chang",
                        "suffix": ""
                    },
                    {
                        "first": "Xiang-Rui",
                        "middle": [],
                        "last": "Hsieh",
                        "suffix": ""
                    },
                    {
                        "first": "Chih-Jen",
                        "middle": [],
                        "last": "Wang",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Lin",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "Journal of Machine Learning Research",
                "volume": "9",
                "issue": "",
                "pages": "1871--1874",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Rong-En Fan, Kai-Wei Chang, Cho-Jui Hsieh, Xiang-Rui Wang, and Chih-Jen Lin. 2008. LIBLINEAR: A library for large linear classification. Journal of Machine Learning Research, 9:1871-1874.",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "PPDB: The paraphrase database",
                "authors": [
                    {
                        "first": "Juri",
                        "middle": [],
                        "last": "Ganitkevitch",
                        "suffix": ""
                    },
                    {
                        "first": "Benjamin",
                        "middle": [],
                        "last": "Van Durme",
                        "suffix": ""
                    },
                    {
                        "first": "Chris",
                        "middle": [],
                        "last": "Callison-Burch",
                        "suffix": ""
                    }
                ],
                "year": 2013,
                "venue": "Proceedings of NAACL-HLT 2013",
                "volume": "",
                "issue": "",
                "pages": "758--764",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Juri Ganitkevitch, Benjamin Van Durme, and Chris Callison-Burch. 2013. PPDB: The paraphrase database. In Proceedings of NAACL-HLT 2013, pages 758-764, Atlanta, Georgia, June. ACL.",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "Uttime: Temporal relation classification using deep syntactic features",
                "authors": [
                    {
                        "first": "Natsuda",
                        "middle": [],
                        "last": "Laokulrat",
                        "suffix": ""
                    },
                    {
                        "first": "Makoto",
                        "middle": [],
                        "last": "Miwa",
                        "suffix": ""
                    },
                    {
                        "first": "Yoshimasa",
                        "middle": [],
                        "last": "Tsuruoka",
                        "suffix": ""
                    },
                    {
                        "first": "Takashi",
                        "middle": [],
                        "last": "Chikayama",
                        "suffix": ""
                    }
                ],
                "year": 2013,
                "venue": "Proceedings of the Seventh International Workshop on Semantic Evaluation (Se-mEval 2013)",
                "volume": "2",
                "issue": "",
                "pages": "88--92",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Natsuda Laokulrat, Makoto Miwa, Yoshimasa Tsuruoka, and Takashi Chikayama. 2013. Uttime: Temporal re- lation classification using deep syntactic features. In Second Joint Conference on Lexical and Computational Semantics (*SEM), Volume 2: Proceedings of the Seventh International Workshop on Semantic Evaluation (Se- mEval 2013), pages 88-92, Atlanta, Georgia, USA, June. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "Stacking approach to temporal relation classification with temporal inference",
                "authors": [
                    {
                        "first": "Natsuda",
                        "middle": [],
                        "last": "Laokulrat",
                        "suffix": ""
                    },
                    {
                        "first": "Makoto",
                        "middle": [],
                        "last": "Miwa",
                        "suffix": ""
                    },
                    {
                        "first": "Yoshimasa",
                        "middle": [],
                        "last": "Tsuruoka",
                        "suffix": ""
                    }
                ],
                "year": 2015,
                "venue": "Journal of Natural Language Processing",
                "volume": "22",
                "issue": "3",
                "pages": "171--196",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Natsuda Laokulrat, Makoto Miwa, and Yoshimasa Tsuruoka. 2015. Stacking approach to temporal relation classi- fication with temporal inference. Journal of Natural Language Processing, 22(3):171-196.",
                "links": null
            },
            "BIBREF12": {
                "ref_id": "b12",
                "title": "An information-theoretic definition of similarity",
                "authors": [
                    {
                        "first": "Dekang",
                        "middle": [],
                        "last": "Lin",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "Proceedings of the 15th International Conference on Machine Learning, ICML '98",
                "volume": "",
                "issue": "",
                "pages": "296--304",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Dekang Lin. 1998. An information-theoretic definition of similarity. In Proceedings of the 15th International Conference on Machine Learning, ICML '98, pages 296-304, San Francisco, CA, USA. Morgan Kaufmann Publishers Inc.",
                "links": null
            },
            "BIBREF13": {
                "ref_id": "b13",
                "title": "Coverage and inheritance in the preposition project",
                "authors": [
                    {
                        "first": "C",
                        "middle": [],
                        "last": "Kenneth",
                        "suffix": ""
                    },
                    {
                        "first": "Orin",
                        "middle": [],
                        "last": "Litkowski",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Hargraves",
                        "suffix": ""
                    }
                ],
                "year": 2006,
                "venue": "3rd ACL-SIGSEM Workshop on Prepositions",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Kenneth C. Litkowski and Orin Hargraves. 2006. Coverage and inheritance in the preposition project. In 3rd ACL-SIGSEM Workshop on Prepositions.",
                "links": null
            },
            "BIBREF14": {
                "ref_id": "b14",
                "title": "Pattern dictionary of english prepositions",
                "authors": [
                    {
                        "first": "Ken",
                        "middle": [],
                        "last": "Litkowski",
                        "suffix": ""
                    }
                ],
                "year": 2014,
                "venue": "Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics",
                "volume": "1",
                "issue": "",
                "pages": "1274--1283",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Ken Litkowski. 2014. Pattern dictionary of english prepositions. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1274-1283, Baltimore, Maryland, June. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF15": {
                "ref_id": "b15",
                "title": "Semeval-2015 task 5: Qa tempeval -evaluating temporal information understanding with question answering",
                "authors": [
                    {
                        "first": "Hector",
                        "middle": [],
                        "last": "Llorens",
                        "suffix": ""
                    },
                    {
                        "first": "Nathanael",
                        "middle": [],
                        "last": "Chambers",
                        "suffix": ""
                    },
                    {
                        "first": "Naushad",
                        "middle": [],
                        "last": "Uzzaman",
                        "suffix": ""
                    },
                    {
                        "first": "Nasrin",
                        "middle": [],
                        "last": "Mostafazadeh",
                        "suffix": ""
                    },
                    {
                        "first": "James",
                        "middle": [],
                        "last": "Allen",
                        "suffix": ""
                    },
                    {
                        "first": "James",
                        "middle": [],
                        "last": "Pustejovsky",
                        "suffix": ""
                    }
                ],
                "year": 2015,
                "venue": "Proceedings of the 9th International Workshop on Semantic Evaluation",
                "volume": "",
                "issue": "",
                "pages": "792--800",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Hector Llorens, Nathanael Chambers, Naushad UzZaman, Nasrin Mostafazadeh, James Allen, and James Puste- jovsky. 2015. Semeval-2015 task 5: Qa tempeval -evaluating temporal information understanding with ques- tion answering. In Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 792-800, Denver, Colorado, June. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF16": {
                "ref_id": "b16",
                "title": "An analysis of causality between events and its relation to temporal information",
                "authors": [
                    {
                        "first": "Paramita",
                        "middle": [],
                        "last": "Mirza",
                        "suffix": ""
                    },
                    {
                        "first": "Sara",
                        "middle": [],
                        "last": "Tonelli",
                        "suffix": ""
                    }
                ],
                "year": 2014,
                "venue": "Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers",
                "volume": "",
                "issue": "",
                "pages": "2097--2106",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Paramita Mirza and Sara Tonelli. 2014a. An analysis of causality between events and its relation to temporal information. In Proceedings of COLING 2014, the 25th International Conference on Computational Linguis- tics: Technical Papers, pages 2097-2106, Dublin, Ireland, August. Dublin City University and Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF17": {
                "ref_id": "b17",
                "title": "Classifying temporal relations with simple features",
                "authors": [
                    {
                        "first": "Paramita",
                        "middle": [],
                        "last": "Mirza",
                        "suffix": ""
                    },
                    {
                        "first": "Sara",
                        "middle": [],
                        "last": "Tonelli",
                        "suffix": ""
                    }
                ],
                "year": 2014,
                "venue": "Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "308--317",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Paramita Mirza and Sara Tonelli. 2014b. Classifying temporal relations with simple features. In Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, pages 308-317, Gothenburg, Sweden, April. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF18": {
                "ref_id": "b18",
                "title": "Annotating causality in the tempeval-3 corpus",
                "authors": [
                    {
                        "first": "Paramita",
                        "middle": [],
                        "last": "Mirza",
                        "suffix": ""
                    },
                    {
                        "first": "Rachele",
                        "middle": [],
                        "last": "Sprugnoli",
                        "suffix": ""
                    },
                    {
                        "first": "Sara",
                        "middle": [],
                        "last": "Tonelli",
                        "suffix": ""
                    },
                    {
                        "first": "Manuela",
                        "middle": [],
                        "last": "Speranza",
                        "suffix": ""
                    }
                ],
                "year": 2014,
                "venue": "Proceedings of the EACL 2014 Workshop on Computational Approaches to Causality in Language (CAtoCL)",
                "volume": "",
                "issue": "",
                "pages": "10--19",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Paramita Mirza, Rachele Sprugnoli, Sara Tonelli, and Manuela Speranza. 2014. Annotating causality in the tempeval-3 corpus. In Proceedings of the EACL 2014 Workshop on Computational Approaches to Causality in Language (CAtoCL), pages 10-19, Gothenburg, Sweden, April. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF19": {
                "ref_id": "b19",
                "title": "The textpro tool suite",
                "authors": [
                    {
                        "first": "Emanuele",
                        "middle": [],
                        "last": "Pianta",
                        "suffix": ""
                    },
                    {
                        "first": "Christian",
                        "middle": [],
                        "last": "Girardi",
                        "suffix": ""
                    },
                    {
                        "first": "Roberto",
                        "middle": [],
                        "last": "Zanoli",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "Proceedings of the Sixth International Conference on Language Resources and Evaluation (LREC'08)",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Emanuele Pianta, Christian Girardi, and Roberto Zanoli. 2008. The textpro tool suite. In Proceedings of the Sixth International Conference on Language Resources and Evaluation (LREC'08), Marrakech, Morocco, may. European Language Resources Association (ELRA). http://www.lrec-conf.org/proceedings/lrec2008/.",
                "links": null
            },
            "BIBREF20": {
                "ref_id": "b20",
                "title": "Timeml: Robust specification of event and temporal expressions in text",
                "authors": [
                    {
                        "first": "James",
                        "middle": [],
                        "last": "Pustejovsky",
                        "suffix": ""
                    },
                    {
                        "first": "Jos\u00e9",
                        "middle": [],
                        "last": "Casta\u00f1o",
                        "suffix": ""
                    },
                    {
                        "first": "Robert",
                        "middle": [],
                        "last": "Ingria",
                        "suffix": ""
                    },
                    {
                        "first": "Roser",
                        "middle": [],
                        "last": "Saur\u00ed",
                        "suffix": ""
                    },
                    {
                        "first": "Robert",
                        "middle": [],
                        "last": "Gaizauskas",
                        "suffix": ""
                    },
                    {
                        "first": "Andrea",
                        "middle": [],
                        "last": "Setzer",
                        "suffix": ""
                    },
                    {
                        "first": "Graham",
                        "middle": [],
                        "last": "Katz",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "Proceedings of the Fifth International Workshop on Computational Semantics (IWCS-5)",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "James Pustejovsky, Jos\u00e9 Casta\u00f1o, Robert Ingria, Roser Saur\u00ed, Robert Gaizauskas, Andrea Setzer, and Graham Katz. 2003. Timeml: Robust specification of event and temporal expressions in text. In Proceedings of the Fifth International Workshop on Computational Semantics (IWCS-5).",
                "links": null
            },
            "BIBREF21": {
                "ref_id": "b21",
                "title": "Elements of symbolic logic",
                "authors": [
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Reichenbach",
                        "suffix": ""
                    }
                ],
                "year": 1947,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "H Reichenbach. 1947. Elements of symbolic logic. University of California Press, Berkeley, CA.",
                "links": null
            },
            "BIBREF22": {
                "ref_id": "b22",
                "title": "Toward a better understanding of causality between verbal events: Extraction and analysis of the causal power of verb-verb associations",
                "authors": [
                    {
                        "first": "Mehwish",
                        "middle": [],
                        "last": "Riaz",
                        "suffix": ""
                    },
                    {
                        "first": "Roxana",
                        "middle": [],
                        "last": "Girju",
                        "suffix": ""
                    }
                ],
                "year": 2013,
                "venue": "Proceedings of the SIGDIAL 2013",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Mehwish Riaz and Roxana Girju. 2013. Toward a better understanding of causality between verbal events: Ex- traction and analysis of the causal power of verb-verb associations. In Proceedings of the SIGDIAL 2013",
                "links": null
            },
            "BIBREF24": {
                "ref_id": "b24",
                "title": "Learning textual graph patterns to detect causal event relations",
                "authors": [
                    {
                        "first": "Bryan",
                        "middle": [],
                        "last": "Rink",
                        "suffix": ""
                    },
                    {
                        "first": "Sanda",
                        "middle": [
                            "M"
                        ],
                        "last": "Cosmin Adrian Bejan",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Harabagiu",
                        "suffix": ""
                    }
                ],
                "year": 2010,
                "venue": "FLAIRS Conference",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Bryan Rink, Cosmin Adrian Bejan, and Sanda M. Harabagiu. 2010. Learning textual graph patterns to detect causal event relations. In FLAIRS Conference.",
                "links": null
            },
            "BIBREF25": {
                "ref_id": "b25",
                "title": "Semeval-2013 task 1: Tempeval-3: Evaluating time expressions, events, and temporal relations",
                "authors": [
                    {
                        "first": "Naushad",
                        "middle": [],
                        "last": "Uzzaman",
                        "suffix": ""
                    },
                    {
                        "first": "Hector",
                        "middle": [],
                        "last": "Llorens",
                        "suffix": ""
                    },
                    {
                        "first": "Leon",
                        "middle": [],
                        "last": "Derczynski",
                        "suffix": ""
                    },
                    {
                        "first": "James",
                        "middle": [],
                        "last": "Allen",
                        "suffix": ""
                    },
                    {
                        "first": "Marc",
                        "middle": [],
                        "last": "Verhagen",
                        "suffix": ""
                    },
                    {
                        "first": "James",
                        "middle": [],
                        "last": "Pustejovsky",
                        "suffix": ""
                    }
                ],
                "year": 2013,
                "venue": "Proceedings of the Seventh International Workshop on Semantic Evaluation",
                "volume": "2",
                "issue": "",
                "pages": "1--9",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Naushad UzZaman, Hector Llorens, Leon Derczynski, James Allen, Marc Verhagen, and James Pustejovsky. 2013. Semeval-2013 task 1: Tempeval-3: Evaluating time expressions, events, and temporal relations. In Second Joint Conference on Lexical and Computational Semantics (*SEM), Volume 2: Proceedings of the Seventh International Workshop on Semantic Evaluation (SemEval 2013), pages 1-9, Atlanta, Georgia, USA, June. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF26": {
                "ref_id": "b26",
                "title": "Semeval-2007 task 15: Tempeval temporal relation identification",
                "authors": [
                    {
                        "first": "Marc",
                        "middle": [],
                        "last": "Verhagen",
                        "suffix": ""
                    },
                    {
                        "first": "Robert",
                        "middle": [],
                        "last": "Gaizauskas",
                        "suffix": ""
                    },
                    {
                        "first": "Frank",
                        "middle": [],
                        "last": "Schilder",
                        "suffix": ""
                    },
                    {
                        "first": "Mark",
                        "middle": [],
                        "last": "Hepple",
                        "suffix": ""
                    },
                    {
                        "first": "Graham",
                        "middle": [],
                        "last": "Katz",
                        "suffix": ""
                    },
                    {
                        "first": "James",
                        "middle": [],
                        "last": "Pustejovsky",
                        "suffix": ""
                    }
                ],
                "year": 2007,
                "venue": "Proceedings of the Fourth International Workshop on Semantic Evaluations (SemEval-2007)",
                "volume": "",
                "issue": "",
                "pages": "75--80",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Marc Verhagen, Robert Gaizauskas, Frank Schilder, Mark Hepple, Graham Katz, and James Pustejovsky. 2007. Semeval-2007 task 15: Tempeval temporal relation identification. In Proceedings of the Fourth International Workshop on Semantic Evaluations (SemEval-2007), pages 75-80, Prague, Czech Republic, June. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF27": {
                "ref_id": "b27",
                "title": "Semeval-2010 task 13: Tempeval-2",
                "authors": [
                    {
                        "first": "Marc",
                        "middle": [],
                        "last": "Verhagen",
                        "suffix": ""
                    },
                    {
                        "first": "Roser",
                        "middle": [],
                        "last": "Sauri",
                        "suffix": ""
                    },
                    {
                        "first": "Tommaso",
                        "middle": [],
                        "last": "Caselli",
                        "suffix": ""
                    },
                    {
                        "first": "James",
                        "middle": [],
                        "last": "Pustejovsky",
                        "suffix": ""
                    }
                ],
                "year": 2010,
                "venue": "Proceedings of the 5th International Workshop on Semantic Evaluation",
                "volume": "",
                "issue": "",
                "pages": "57--62",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Marc Verhagen, Roser Sauri, Tommaso Caselli, and James Pustejovsky. 2010. Semeval-2010 task 13: Tempeval-2. In Proceedings of the 5th International Workshop on Semantic Evaluation, pages 57-62, Uppsala, Sweden, July. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF28": {
                "ref_id": "b28",
                "title": "Qualitative CSP, finite CSP, and SAT: Comparing Methods for Qualitative Constraint-based Reasoning",
                "authors": [
                    {
                        "first": "Matthias",
                        "middle": [],
                        "last": "Westphal",
                        "suffix": ""
                    },
                    {
                        "first": "Stefan",
                        "middle": [],
                        "last": "W\u00f6lfl",
                        "suffix": ""
                    }
                ],
                "year": 2009,
                "venue": "Proceedings of the 21st International Joint Conference on Artifical Intelligence, IJCAI'09",
                "volume": "",
                "issue": "",
                "pages": "628--633",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Matthias Westphal and Stefan W\u00f6lfl. 2009. Qualitative CSP, finite CSP, and SAT: Comparing Methods for Qualitative Constraint-based Reasoning. In Proceedings of the 21st International Joint Conference on Artifical Intelligence, IJCAI'09, pages 628-633, San Francisco, CA, USA. Morgan Kaufmann Publishers Inc.",
                "links": null
            },
            "BIBREF29": {
                "ref_id": "b29",
                "title": "Restarts and nogood recording in qualitative constraint-based reasoning",
                "authors": [
                    {
                        "first": "Matthias",
                        "middle": [],
                        "last": "Westphal",
                        "suffix": ""
                    },
                    {
                        "first": "Stefan",
                        "middle": [],
                        "last": "W\u00f6lfl",
                        "suffix": ""
                    },
                    {
                        "first": "Jason Jingshi",
                        "middle": [],
                        "last": "Li",
                        "suffix": ""
                    }
                ],
                "year": 2010,
                "venue": "ECAI 2010 -19th European Conference on Artificial Intelligence, Lisbon, Portugal",
                "volume": "",
                "issue": "",
                "pages": "1093--1094",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Matthias Westphal, Stefan W\u00f6lfl, and Jason Jingshi Li. 2010. Restarts and nogood recording in qualitative constraint-based reasoning. In ECAI 2010 -19th European Conference on Artificial Intelligence, Lisbon, Portu- gal, August 16-20, 2010, Proceedings, pages 1093-1094.",
                "links": null
            },
            "BIBREF30": {
                "ref_id": "b30",
                "title": "Models of causation and the semantics of causal verbs",
                "authors": [
                    {
                        "first": "Phillip",
                        "middle": [],
                        "last": "Wolff",
                        "suffix": ""
                    },
                    {
                        "first": "Grace",
                        "middle": [],
                        "last": "Song",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "Cognitive Psychology",
                "volume": "47",
                "issue": "3",
                "pages": "276--332",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Phillip Wolff and Grace Song. 2003. Models of causation and the semantics of causal verbs. Cognitive Psychology, 47(3):276-332.",
                "links": null
            },
            "BIBREF31": {
                "ref_id": "b31",
                "title": "Representing causation",
                "authors": [
                    {
                        "first": "Phillip",
                        "middle": [],
                        "last": "Wolff",
                        "suffix": ""
                    }
                ],
                "year": 2007,
                "venue": "Journal of experimental psychology: General",
                "volume": "136",
                "issue": "1",
                "pages": "82--111",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Phillip Wolff. 2007. Representing causation. Journal of experimental psychology: General, 136(1):82-111.",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF0": {
                "text": "System architecture of CATENA",
                "uris": null,
                "num": null,
                "type_str": "figure"
            },
            "TABREF0": {
                "content": "<table/>",
                "type_str": "table",
                "num": null,
                "text": "and > for BEFORE and AFTER \u2022 o and o \u22121 for DURING and DURING INV \u2022 d and d \u22121 for IS INCLUDED and INCLUDES \u2022 s and s \u22121 for BEGINS and BEGUN BY \u2022 f and f \u22121 for ENDS and ENDED BY",
                "html": null
            },
            "TABREF1": {
                "content": "<table><tr><td colspan=\"2\">Morphosyntactic information</td><td/><td/><td/><td/><td/></tr><tr><td colspan=\"7\">PoS one-hot phraseChunk x x x x x x x x one-hot Shallow phrase chunk of e1 and e2.</td></tr><tr><td>samePoS</td><td/><td>x</td><td>x</td><td>x</td><td>binary</td><td>Whether e1 and e2 have the same PoS.</td></tr><tr><td>Textual context</td><td/><td/><td/><td/><td/><td/></tr><tr><td>entityOrder</td><td/><td>x</td><td/><td/><td>binary</td><td>Appearance order of e1 and e2 in the text. 5</td></tr><tr><td colspan=\"7\">sentenceDistance 0 entityDistance x x x binary x x x binary 0 if e1 and e2 are adjacent, 1 otherwise.</td></tr><tr><td>EVENT attributes</td><td/><td/><td/><td/><td/><td/></tr><tr><td>class</td><td>x</td><td>x</td><td>x</td><td>x</td><td>one-hot</td><td/></tr><tr><td>tense aspect</td><td>x x</td><td>x x</td><td>x x</td><td>x x</td><td>one-hot one-hot</td><td>EVENT attributes as specified in TimeML.</td></tr><tr><td>polarity</td><td>x</td><td>x</td><td>x</td><td>x</td><td>one-hot</td><td/></tr><tr><td>sameClass</td><td/><td/><td>x</td><td>x</td><td>binary</td><td/></tr><tr><td>sameTenseAspect</td><td/><td/><td>x</td><td>x</td><td>binary</td><td>Whether e1 and e2 have the same EVENT attributes.</td></tr><tr><td>samePolarity</td><td/><td/><td>x</td><td>x</td><td>binary</td><td/></tr><tr><td>TIMEX3 attributes</td><td/><td/><td/><td/><td/><td/></tr><tr><td>type</td><td>x</td><td>x</td><td/><td/><td colspan=\"2\">one-hot TIMEX3 attributes as specified in TimeML.</td></tr><tr><td>Dependency information</td><td/><td/><td/><td/><td/><td/></tr><tr><td>dependencyPath</td><td/><td/><td>x</td><td>x</td><td colspan=\"2\">one-hot Dependency path between e1 and e2.</td></tr><tr><td>isMainVerb</td><td>x</td><td>x</td><td>x</td><td>x</td><td>binary</td><td>Whether e1/e2 is the main verb of the sentence.</td></tr><tr><td>Temporal signals</td><td/><td/><td/><td/><td/><td/></tr><tr><td>tempSignalTokens</td><td/><td>x</td><td>x</td><td>x</td><td colspan=\"2\">one-hot Tokens (cluster) of temporal signal around e1 and e2.</td></tr><tr><td>tempSignalPosition</td><td/><td>x</td><td>x</td><td>x</td><td colspan=\"2\">one-hot Temporal signal position w.r.t e1/e2 (BETWEEN, BEFORE, BEGIN, etc.)</td></tr><tr><td>tempSignalDependency</td><td/><td>x</td><td>x</td><td>x</td><td colspan=\"2\">one-hot Temporal signal dependency path between signal tokens and e1/e2.</td></tr><tr><td>Causal signals</td><td/><td/><td/><td/><td/><td/></tr><tr><td>causSignalTokens</td><td/><td/><td/><td>x</td><td colspan=\"2\">one-hot Tokens (cluster) of causal signal around e1 and e2.</td></tr><tr><td>causSignalPosition</td><td/><td/><td/><td>x</td><td colspan=\"2\">one-hot Causal signal position w.r.t e1/e2 (BETWEEN, BEFORE, BEGIN, etc.)</td></tr><tr><td>causSignalDependency</td><td/><td/><td>x</td><td>x</td><td colspan=\"2\">one-hot Causal signal dependency path between signal tokens and e1/e2.</td></tr><tr><td colspan=\"2\">Lexical semantic information</td><td/><td/><td/><td/><td/></tr><tr><td>wnSim</td><td/><td/><td>x</td><td>x</td><td>one-hot</td><td/></tr></table>",
                "type_str": "table",
                "num": null,
                "text": "Part-of-speech tags of e1 and e2. if e1 and e2 are in the same sentence, 1 otherwise. WordNet similarity computed between the lemmas of e1 and e2. TLINK labels from the rule-based sieve timex-DCT label x one-hot The TLINK type of the e2 (timex) and DCT pair (if any). event-DCT label x one-hot The TLINK types of the e1/e2 and DCT pairs (if any).",
                "html": null
            },
            "TABREF2": {
                "content": "<table/>",
                "type_str": "table",
                "num": null,
                "text": "",
                "html": null
            },
            "TABREF4": {
                "content": "<table><tr><td/><td/><td/><td colspan=\"2\">CATENA</td><td/><td/><td/><td>CAEVO</td><td/></tr><tr><td/><td colspan=\"2\">TempEval-3</td><td/><td colspan=\"3\">TimeBank-Dense</td><td colspan=\"3\">TimeBank-Dense</td></tr><tr><td>Sieve</td><td>P</td><td>R</td><td>F1</td><td>P</td><td>R</td><td>F1</td><td>P</td><td>R</td><td>F1</td></tr><tr><td colspan=\"2\">Temporal Relation Identification</td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td colspan=\"3\">.530 .954 .682</td><td>-</td><td>-</td><td>-</td><td>-</td><td>-</td><td>-</td></tr><tr><td colspan=\"3\">Temporal Relation Type Classification</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td>RB</td><td colspan=\"6\">.908 .127 .223 .727 .049 .092</td><td>-</td><td>-</td><td>-</td></tr><tr><td>RB + TR</td><td colspan=\"6\">.921 .163 .278 .713 .076 .138</td><td>-</td><td>-</td><td>-</td></tr><tr><td>ML</td><td colspan=\"9\">.610 .575 .592 .484 .471 .478 .458 .202 .280</td></tr><tr><td>RB + ML</td><td colspan=\"9\">.616 .595 .605 .495 .493 .494 .486 .240 .321</td></tr><tr><td>RB + TR + ML</td><td colspan=\"9\">.626 .613 .619 .512 .510 .511 .505 .328 .398</td></tr><tr><td>RB + TR + ML + AllVague</td><td>-</td><td>-</td><td>-</td><td>-</td><td>-</td><td>-</td><td colspan=\"3\">.508 .506 .507</td></tr><tr><td>Causal Relation Extraction</td><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td>RB</td><td colspan=\"3\">.917 .423 .579</td><td>-</td><td>-</td><td>-</td><td>-</td><td>-</td><td>-</td></tr><tr><td>ML</td><td colspan=\"3\">.429 .115 .182</td><td>-</td><td>-</td><td>-</td><td>-</td><td>-</td><td>-</td></tr><tr><td>RB + ML</td><td colspan=\"3\">.737 .538 .622</td><td>-</td><td>-</td><td>-</td><td>-</td><td>-</td><td>-</td></tr></table>",
                "type_str": "table",
                "num": null,
                "text": "CATENA evaluated on Tempeval-3 data, compared with the two best participating systems according to UzZaman et al. (2013) and the system byLaokulrat et al. (2015) (left). CATENA is also compared with CAEVO on the TimeBank-Dense test set (right).",
                "html": null
            },
            "TABREF5": {
                "content": "<table/>",
                "type_str": "table",
                "num": null,
                "text": "Analysis of classifier performance per sieve. RB: rule-based sieve, ML: machine-learned sieve and TR: temporal reasoner.",
                "html": null
            },
            "TABREF6": {
                "content": "<table><tr><td>E-E pair Sentence</td><td>TE3-gold</td><td>TE-label</td><td colspan=\"2\">CA-label Post-editing</td></tr><tr><td colspan=\"3\">(e IS INCLUDED</td><td>CLINK-R</td><td>AFTER</td></tr><tr><td>on</td><td/><td/><td/></tr></table>",
                "type_str": "table",
                "num": null,
                "text": "32 , e 44 ) The [incident] e32 provoked an international [outcry] e44 ... -SIMULTANEOUS CLINK BEFORE (e 32 , e 45 ) The [incident] e32 provoked an international outcry and -AFTER CLINK BEFORE led to a major [deterioration] e45 in relations... (e 18 , e 19 ) ...the [inspections] e18 were directly linked to the new law -NGOs and the targeted groups' [compliance] e19 with it. (e 4 , e 6 ) A haze akin to volcanic fumes [cloaked] e4 the capital, INCLUDES AFTER CLINK BEFORE causing convulsive [coughing] e6 and...",
                "html": null
            },
            "TABREF7": {
                "content": "<table/>",
                "type_str": "table",
                "num": null,
                "text": "Examples of E-E pairs in the TempEval-3-platinum dataset with gold annotated labels (TE3-gold), labelled by the temporal module (TE-label) and causal module (CA-label) of CATENA. These examples illustrate how TLINK post-editing using CLINK could improve the labelling quality. the recall of the highly precise RB component. The low precision of the ML module is mostly due to dependency parsing mistakes and issues in disambiguating signals such as from, as in \"...passenger cars in China was on track to hit [400 million] T by 2030, up from [90 million] S now.\" Unfortunately, from the total of 5 gold CLINKs in the 20 documents of the TimeBank-Dense test set, none is identified by CATENA.",
                "html": null
            },
            "TABREF9": {
                "content": "<table><tr><td>BEGUN BY</td></tr></table>",
                "type_str": "table",
                "num": null,
                "text": "E-D label rules based on tense and aspect of E. tsense E-T label TIMEPOINT (e.g. in, at, on) IS INCLUDED TIMEPRECEDING (e.g. before) BEFORE TIMEFOLLOWING (e.g. after) AFTER DURATION (e.g. during, throughout) DURING STARTTIME (e.g. from, since)",
                "html": null
            },
            "TABREF10": {
                "content": "<table><tr><td>dep</td><td>e 1 verb info</td><td>E-E label</td><td>Example</td></tr><tr><td>LGS-PMOD</td><td/><td/><td/></tr><tr><td>LOC-PMOD</td><td>*</td><td>IS INCLUDED</td><td/></tr></table>",
                "type_str": "table",
                "num": null,
                "text": "E-T label rules based on the sense of temporal preposition (tsense) introducing T. AFTER \"...reaction [touched] e1 off by the [collapse] e2 of...\" ...enormous [surge] e1 in coal [consumption] e2 ...\" OPRD-IM/OPRD aspectual verb for initiation BEGINS \"...situation [began] e1 to [relax] e2 in...\" aspectual verb for culmination/termination ENDS \"...we 'd [stop] e1 [bidding] e2 .\" aspectual verb for continuation INCLUDES \"...industry 's growth [continues] e1 to [slow] e2 .\" general verb, aspect=PERFECTIVE PROGRESSIVE SIMULTANEOUS \"...have been [working] e1 to [develop] e2 quantum...\" general verb BEFORE \"...consortium [attempted] e1 to [block] e2 ...\"",
                "html": null
            },
            "TABREF11": {
                "content": "<table><tr><td>Relation</td><td>Path</td><td>Example</td></tr><tr><td>between v and e 1</td><td>dep1</td><td/></tr><tr><td>e 1 is subject of v</td><td/><td/></tr></table>",
                "type_str": "table",
                "num": null,
                "text": "SBJThe Pope's[visit]  e1 persuades v Cubans... v is predicative complement of e 1 PRD-IM The [roundup] e1 was to prevent v them... v is modifier of e 1 (nominal) NMOD An [agreement] e1 that permits v the Russian... v is apposition of e 1 APPO ..., with the [crisis] e1 triggered v by... v is general adverbial of e 1 ADV The number [increased] e1 , prompting v ... vis adverbial of purpose/reason of e 1 PRP-IM The major [allocated] e1 funds to help v ... between v and e 2 dep2 e 2 is object of v OBJ ...have provoked v widespread [violence] e2 . e 2 is logical subject of v (passive verb) LGS-PMOD ...triggered v by the [end] e2 of the... e 2 is predicative complement of v (raising/control verb) OPRD ...funds to help v [build] e2 a museum. OPRD-IM ...persuades v Cubans to [break] e2 loose. e 2 is general adverbial of v ADV-PMOD ...protect v them from unspecified [threats] e2 . e 2 is adverbial of direction of v DIR-PMOD ...lead to v a [surge] e2 of inexpensive imports. e 2 is modifier of v (adjective or adverbial) AMOD-PMOD ...related to v [problems] e2 under a contract.",
                "html": null
            },
            "TABREF12": {
                "content": "<table/>",
                "type_str": "table",
                "num": null,
                "text": "",
                "html": null
            }
        }
    }
}