{ "paper_id": "I13-1047", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T07:14:46.124494Z" }, "title": "Using the Semantic-Syntactic Interface for Reliable Arabic Modality Annotation", "authors": [ { "first": "Rania", "middle": [], "last": "Al-Sabbagh", "suffix": "", "affiliation": { "laboratory": "", "institution": "University of Illinois at Urba-na-Champaign", "location": {} }, "email": "alsabba1@illinois.edu" }, { "first": "Jana", "middle": [], "last": "Diesner", "suffix": "", "affiliation": { "laboratory": "School of Library and Infor-mation Science University of Illinois at Urba-na-Champaign", "institution": "", "location": {} }, "email": "jdiesner@illinois.edu" }, { "first": "Roxana", "middle": [], "last": "Girju", "suffix": "", "affiliation": {}, "email": "girju@illinois.edu" } ], "year": "", "venue": null, "identifiers": {}, "abstract": "We introduce a novel modality scheme where triggers are words and phrases that convey modality meanings and subcategorize for clauses and verbal phrases. This semanticsyntactic working definition of modality enables us to design practical and replicable annotation guidelines and procedures that alleviate some shortcomings of current purely semantic modality annotation schemes and yield high inter-annotator agreement rates. We use this scheme to annotate a tweet-based Arabic corpus for modality information. This novel language resource, being the first, initiates NLP research on Arabic modality.", "pdf_parse": { "paper_id": "I13-1047", "_pdf_hash": "", "abstract": [ { "text": "We introduce a novel modality scheme where triggers are words and phrases that convey modality meanings and subcategorize for clauses and verbal phrases. This semanticsyntactic working definition of modality enables us to design practical and replicable annotation guidelines and procedures that alleviate some shortcomings of current purely semantic modality annotation schemes and yield high inter-annotator agreement rates. We use this scheme to annotate a tweet-based Arabic corpus for modality information. This novel language resource, being the first, initiates NLP research on Arabic modality.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Abstract", "sec_num": null } ], "body_text": [ { "text": "Modality is the grammaticalized expression of the \"speaker's subjective attitudes\" (Bybee et al., 1994:176) and \"psychological stances\" (Mitchell and al-Hassan, 1994 :7) towards propositions and events and their factuality status. In NLP applications and domains, modality is considered as one linguistic means to convey and detect attitudes and opinions (Wiebe at al., 2005; Abdul-Mageed and Diab, 2012) , commitments and beliefs (Diab et al., 2009) , power relations , uncertainties and speculations (Szarvas et al., 2008; Matsuyoshi et al., 2010) .", "cite_spans": [ { "start": 83, "end": 107, "text": "(Bybee et al., 1994:176)", "ref_id": null }, { "start": 136, "end": 165, "text": "(Mitchell and al-Hassan, 1994", "ref_id": null }, { "start": 355, "end": 375, "text": "(Wiebe at al., 2005;", "ref_id": "BIBREF19" }, { "start": 376, "end": 404, "text": "Abdul-Mageed and Diab, 2012)", "ref_id": "BIBREF0" }, { "start": 431, "end": 450, "text": "(Diab et al., 2009)", "ref_id": "BIBREF7" }, { "start": 502, "end": 524, "text": "(Szarvas et al., 2008;", "ref_id": "BIBREF18" }, { "start": 525, "end": 549, "text": "Matsuyoshi et al., 2010)", "ref_id": "BIBREF10" } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "We herein present the first work on Arabic modality annotation, which is part of a larger research project to (1) automatically identify modality triggers (i.e. words and phrases conveying modality meanings), holders (i.e. modality experiencers) and scopes (i.e. the range of linguistic constituents modified by the modality triggers), and (2) automatically detect power re-lations among participants in the social network of Twitter by using this modality information.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "Despite solid work on Arabic modality in theoretical linguistics (Mitchell and al-Hassan, 1994; Brustad, 2000; Moshref, 2012) , there are no Arabic corpora annotated for modality, not even the widely used Penn Arabic Treebank. However, there is a plethora of work and annotated corpora for modality in other languages, including English (Saur\u00ed et al, 2006; Baker et al., 2010; Rubinstein et al., 2013) , Portuguese (Hendrickx et al., 2010; Avila and Mello, 2013) , Japanese (Matsuyoshi et al. 2010) and Chinese (Cui and Chi, 2013) .", "cite_spans": [ { "start": 65, "end": 95, "text": "(Mitchell and al-Hassan, 1994;", "ref_id": null }, { "start": 96, "end": 110, "text": "Brustad, 2000;", "ref_id": "BIBREF4" }, { "start": 111, "end": 125, "text": "Moshref, 2012)", "ref_id": "BIBREF12" }, { "start": 337, "end": 356, "text": "(Saur\u00ed et al, 2006;", "ref_id": "BIBREF17" }, { "start": 357, "end": 376, "text": "Baker et al., 2010;", "ref_id": "BIBREF2" }, { "start": 377, "end": 401, "text": "Rubinstein et al., 2013)", "ref_id": null }, { "start": 415, "end": 439, "text": "(Hendrickx et al., 2010;", "ref_id": "BIBREF8" }, { "start": 440, "end": 462, "text": "Avila and Mello, 2013)", "ref_id": null }, { "start": 474, "end": 498, "text": "(Matsuyoshi et al. 2010)", "ref_id": "BIBREF10" }, { "start": 503, "end": 530, "text": "Chinese (Cui and Chi, 2013)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "Arabic modality annotation involves multiple challenges. First, the paradigm of Arabic modality triggers is complex as it includes auxiliaries, lexical verbs, nominals and particles -like many other languages as well. Second, triggers can be lexically and/or semantically ambiguous: a lexically-ambiguous trigger is a word or phrase that may or may not convey a modality meaning based on context. A semantically-ambiguous trigger is a word or phrase that may convey two or more modality meanings. Third, implicit scopes are common in Arabic and annotators have to be made aware of them. Fourth, Arabic word order flexibility allows triggers -especially adverbials -to occur in the scope's initial, medial or final positions, which makes it challenging for annotators to identify scope spans. Finally, modality scopes are not necessarily adjacent to their triggers, which furthermore complicates the detection of their spans.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "The tweets genre on which we work poses an additional challenge due to language variation. We select a random sample of Arabic tweets from the YADAC corpus (Al-Sabbagh and Girju, 2012) posted in Egypt during the first six months of the 2011 revolution. All selected tweets are about the political situation at that time. Tweets are not only in the Egyptian Arabic (EA) dialect, but also in Modern Standard Arabic (MSA), especially tweets from press agencies and celebrities. Therefore, our annotation scheme has to work on both MSA and EA modality.", "cite_spans": [ { "start": 156, "end": 184, "text": "(Al-Sabbagh and Girju, 2012)", "ref_id": "BIBREF1" } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "Arabic and the tweets genre are not the only original aspects of this paper. We present a novel linguistically-motivated annotation scheme with a semantic-syntactic working definition of modality triggers as words and phrases that convey modality meanings and subcategorize for clauses and verbal phrases. Modality meanings are based on Palmer's (1989 Palmer's ( , 2001 ) cross-lingual typology of modality, which is proven valid for both MSA and EA (Mitchell and al-Hassan, 1994; Brustad, 2000; Moshref, 2012) .", "cite_spans": [ { "start": 337, "end": 351, "text": "Palmer's (1989", "ref_id": "BIBREF13" }, { "start": 352, "end": 369, "text": "Palmer's ( , 2001", "ref_id": "BIBREF14" }, { "start": 450, "end": 480, "text": "(Mitchell and al-Hassan, 1994;", "ref_id": null }, { "start": 481, "end": 495, "text": "Brustad, 2000;", "ref_id": "BIBREF4" }, { "start": 496, "end": 510, "text": "Moshref, 2012)", "ref_id": "BIBREF12" } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "The semantic-syntactic interface between modality triggers and their scopes depicted in our definition is well-established in theoretical linguistics for Arabic (Moshref, 2012) and for English (Jackendoff, 1972; Brennan, 1993; Butler, 2003) . Semantics and syntax work simultaneously such that semantics guarantees not to tag all words and phrases that subcategorize for clauses and verbal phrases as modality triggers. Syntax guarantees not to tag words and phrases that share some semantic meanings with modality triggers, but do not subcategorize for clauses and verbal phrases, as modality triggers.", "cite_spans": [ { "start": 161, "end": 176, "text": "(Moshref, 2012)", "ref_id": "BIBREF12" }, { "start": 193, "end": 211, "text": "(Jackendoff, 1972;", "ref_id": "BIBREF9" }, { "start": 212, "end": 226, "text": "Brennan, 1993;", "ref_id": "BIBREF3" }, { "start": 227, "end": 240, "text": "Butler, 2003)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "Subcategorization frames of modality triggers are sporadically mentioned in the NLP literature on modality annotation. For English, Saur\u00ed et al. (2006) stated in their definition of event modality triggers that \"they subcategorize for a that, gerundive or infinitival clause, but also an NP headed by event denoting nouns\" (p.334). For Portuguese, Hendrickx et al. (2010) stated that \"in the majority of the cases, the target [i.e. scope] is a subordinate clause or a verbal phrase ... in some cases, also main clauses can be targets [i.e. scopes]\" (p. 1810). However, no prior work integrates the semantic-syntactic interface into a modality annotation scheme.", "cite_spans": [ { "start": 132, "end": 151, "text": "Saur\u00ed et al. (2006)", "ref_id": "BIBREF17" }, { "start": 348, "end": 371, "text": "Hendrickx et al. (2010)", "ref_id": "BIBREF8" } ], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "Our conceptualization and implementation of this semantic-syntactic interface provide annotators with practical annotation guidelines that yield highly-reliable results, as shown herein. Furthermore, they define modality in terms of concrete syntactic features which we use in our future work for the automatic identification of triggers and their scopes.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "The rest of the paper is structured as follows: section 2 briefly reviews related work. Section 3", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "gives details about Arabic modality and its complexities. Section 4 presents our data, annotation guidelines and procedures. Section 5 reports the annotation results. Finally, we conclude with future work in Section 6.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Introduction", "sec_num": "1" }, { "text": "Recent work on modality annotation focuses on English, Portuguese, Japanese and Chinese. Baker et al. (2010) used an idiosyncratic categorization of English modality that distinguished 8 semantic meanings: requirement, permissive, success, effort, intension, ability, want and belief. They defined each type as a pattern of the form H (modal) P where H is the sentence's agent and P is the proposition (e.g. H permits [P to be true/false]). They obtained an average interannotator agreement rate of 0.82. The error analysis of their modality tagger showed that errors resulted primarily from the triggers' lexical ambiguity.", "cite_spans": [ { "start": 89, "end": 108, "text": "Baker et al. (2010)", "ref_id": "BIBREF2" } ], "ref_spans": [], "eq_spans": [], "section": "Related Work", "sec_num": "2" }, { "text": "Prabhakaran et al. (2012) focused on 5 semantic meanings of English modality, and used the same HP patterns as Baker et al. (2010) for annotation guidelines. They reported an interannotator agreement rate of 0.95. Their modality tagger yielded a 0.44 F 1 score against a goldstandard and 0.79 and 0.91 F 1 scores against different testing sets from their crowdsourced data.", "cite_spans": [ { "start": 111, "end": 130, "text": "Baker et al. (2010)", "ref_id": "BIBREF2" } ], "ref_spans": [], "eq_spans": [], "section": "Related Work", "sec_num": "2" }, { "text": "Rubinstein et al. (2013) used a more standardized typology of English modality that entailed (1) priority modality divided into bouletic, teleological and deontic triggers; and (2) non-priority modality divided into epistemic, circumstantial and ability triggers. Their purely semantic annotation scheme returned an alpha reliability score of 0.89 only when collapsing the subtypes of priority and non-priority triggers. The scheme yielded an alpha reliability score of 0.65 for scope span annotation.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Related Work", "sec_num": "2" }, { "text": "Cui and Chi (2013) applied Rubinstein et al.'s (2013) scheme for modality annotation to the Penn Chinese Treebank. They obtained a reliability score of 0.94 for triggers' annotation using the collapsed binary typology of modality triggers as priority vs. non-priority. Their error analysis reported vagueness in the annotation guidelines as one disagreement factor.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Related Work", "sec_num": "2" }, { "text": "The lack of previous NLP work on Arabic modality, modality annotation in tweets and syntactically-guided modality annotation schemes render direct comparisons to our work impossible. Yet, the two main distinguishing factors of our work are: (1) to guarantee the replicability of our study, we avoid idiosyncratic typologies of modality that were used in some previous work; and (2) to better guide our annotators, we use practical guidelines that rely on both semantics and syntax rather than semantics only as in previous annotation schemes.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Related Work", "sec_num": "2" }, { "text": "Among multiple typologies of modality, Palmer's (1989, 2001 ) was validated for both MSA and EA in theoretical linguistics (Mitchell and al-Hassan, 1994; Brustad, 2000; Moshref, 2012) .", "cite_spans": [ { "start": 48, "end": 54, "text": "(1989,", "ref_id": null }, { "start": 55, "end": 59, "text": "2001", "ref_id": null }, { "start": 123, "end": 153, "text": "(Mitchell and al-Hassan, 1994;", "ref_id": null }, { "start": 154, "end": 168, "text": "Brustad, 2000;", "ref_id": "BIBREF4" }, { "start": 169, "end": 183, "text": "Moshref, 2012)", "ref_id": "BIBREF12" } ], "ref_spans": [], "eq_spans": [], "section": "Background", "sec_num": "3.1" }, { "text": "Palmer distinguishes two main classes of modality: propositional and event. Propositional modality is concerned with the speaker's attitude to the truth-value of a proposition, and includes: \uf0b7 Epistemic modality, which expresses the speaker's judgment about the factual status of the proposition as well as the speaker's opinion and attitude towards that proposition. \uf0b7 Evidential modality, which indicates the evidence the speaker has for his or her judgment or opinion. Evidence can be reported as in hearsay and quotes or sensory. Event modality refers to events that are not actualized but are merely potential, and includes: \uf0b7 Deontic modality, which relates to obligations and permissions that emanate from an external source, and commissives, which originate from an internal source as speakers lay an obligation on themselves for a potential event. \uf0b7 Dynamic modality, which relates to ability, willingness and wishes.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Background", "sec_num": "3.1" }, { "text": "The challenges of Arabic modality annotation are attributed to (1) the complexity of the Arabic modality paradigm, (2) the lexical and semantic ambiguity of Arabic modality triggers, (3) implicit scopes, (4) word order flexibility and (5) potential long dependencies between triggers and their scopes.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Challenges", "sec_num": "3.2" }, { "text": "The paradigm of Arabic modality triggers includes a large set of auxiliaries, lexical verbs, nominals and particles. Except for auxiliaries, adverbs and some particles, all modality triggers inflect for gender, number, person, tense, aspect and mood. Furthermore, generic modality patterns such as \u202b\u0627\ufedf\u0640\u202c \u202b\ufee3\ufee6\u202c * \u202b\u0623\u0646\u202c mn Al-* >n (it is * that), where * is typically an adjective (e.g. \u202b\u0623\u0646\u202c \u202b\u0627\ufedf\ufee4\ufbad\ufee2\u202c \u202b\ufee3\ufee6\u202c mn Almhm >n (it's important that)), are common.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Challenges", "sec_num": "3.2" }, { "text": "Modality triggers can be lexically and/or semantically ambiguous. The noun \u202b\u0632\ufee3\ufe8e\u0646\u202c zmAn is one example of a lexically-ambiguous trigger because in 1 it is an epistemic with a clause scope. Yet, in 2 it is a non-modal standing for era. The MSA particle \u202b\ufefb\ufe91\ufeaa\u202c lAbd is one example of semantically-ambiguous triggers because in 3 it is an epistemic with a clause scope; whereas in 4 it is an obligative with a verbal-phrase scope. Implicit modality scopes are common in Arabic and come in different realizations. In 5, the scope of the permissive \u202b\ufee7\ufeb4\ufee4\ufea2\u202c nsmH (allow) is the deictic \u202b\u0630\ufedf\ufeda\u202c *lk (that) which refers to the clause \u202b\ufbfe\ufbad\ufe8e\u0646\u202c \u202b\u0627\ufedf\ufee4\ufebc\ufeae\ufbfe\ufeee\u0646\u202c \u202b\ufedf\ufee6\u202c ln yhAn AlmSrywn (Egyptians won't be humiliated). That is, the scope of nsmH is actually a clause.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Challenges", "sec_num": "3.2" }, { "text": "5 . \u202b\ufedf\ufee6\u202c \u202b\ufbfe\ufbad\ufe8e\u0646\u202c \u202b\u0627\ufedf\u202c \u202b\ufee4\ufebc\ufeae\ufbfe\u202c \u202b\ufeee\u0646\u202c . . \u202b\ufedf\ufee6\u202c \u202b\ufee7\ufeb4\ufee4\ufea2\u202c \u202b\ufe91\ufeac\ufedf\ufeda\u202c . ln yhAn AlmSrywn. ln nsmH b*lk.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Challenges", "sec_num": "3.2" }, { "text": "Egyptians won't be humiliated. We won't allow it.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Challenges", "sec_num": "3.2" }, { "text": "In 6, the abilitives \u202b\u0627\u202c \u202b\ufecb\ufeae\u0641\u202c AErf (can) and \u202b\ufecb\ufe8e\u0631\u0641\u202c EArf (can) share the same verbal-phrase scope of \u202b\u0627\u202c \u202b\ufeb7\ufeee\u0641\u202c A$wf (see). To avoid redundancy, the speaker elides the scope of the second abilitive -\u202b\u0627\u202c \u202b\ufecb\ufeae\u0641\u202c AErf (can) -and does not replace it with any deictic expression. Thus AErf modifies an implicit verbal-phrase scope. On the surface level, the obligative \u202b\ufefb\u0632\u0645\u202c lAzm (must) in 7 is followed by the noun phrase a real reaction against military trials. Yet, on a deeper level, the tweet is the short version of we must (take) a real reaction against military trials. This means that lAzm has an implicit verbal-phrase scope. Word order flexibility allows for some modality triggers -especially adverbials -to occur before, after or in the middle of their scope(s).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Challenges", "sec_num": "3.2" }, { "text": "Long dependencies between modality triggers and their scope(s) are the last challenge with Arabic modality annotation. The obligative \u202b\u0627\u0637\ufee0\ufe90\u202c ATlb (ask; require) in 8 subcategorizes for a complement clause, which starts 9 words later (affixes excluded). ", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Challenges", "sec_num": "3.2" }, { "text": "We randomly selected a corpus of 1,704 raw tweets (33,349 tokens and 11,013 unique types) from the YADAC corpus (Al-Sabbagh and Girju, 2012) . The considered time span ranges from January 25, 2011 to June 30, 2011. All tweets were posted in Egypt by ordinary individuals, celebrities (e.g. politicians, actors, singers, TV hosts), and the press (e.g. newspapers, TV stations, NGOs, election campaigns).", "cite_spans": [ { "start": 112, "end": 140, "text": "(Al-Sabbagh and Girju, 2012)", "ref_id": "BIBREF1" } ], "ref_spans": [], "eq_spans": [], "section": "Corpus Encoding and Description", "sec_num": "4.1" }, { "text": "The corpus includes tweets in both MSA and EA because press users always post in MSA, while celebrities and ordinary individuals frequently switch between MSA and EA. Based on our manual annotation of user types, we have 1,318 tweets posted by individuals, 369 tweets by celebrities and 17 tweets by the press.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Corpus Encoding and Description", "sec_num": "4.1" }, { "text": "Two EA native speakers performed the annotation. Being linguistics students, they can be assumed to master MSA. They were given a onehour video tutorial covering the annotation guidelines and procedures in Sections 4.3 and 4.4, respectively, followed by a 30-minute workshop dedicated to training and discussion.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Annotators and Annotation Units", "sec_num": "4.2" }, { "text": "Each annotator is required to label each (1) modality trigger, (2) its semantic meaning, (3) its scope type(s), and (4) its scope span(s). We keep holder annotation for future work as it poses additional challenges.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Annotators and Annotation Units", "sec_num": "4.2" }, { "text": "Our core annotation guidelines are summarized in the semantic-syntactic working definition of modality given in Section 1. We define modality triggers as words and phrases that (1) convey a modality meaning from Palmer's (1989, 2001) typology, (2) and subcategorize for clauses and verbal phrases; representing propositions and events, respectively. We also give the annotators a number of supplementary guidelines.", "cite_spans": [ { "start": 221, "end": 227, "text": "(1989,", "ref_id": null }, { "start": 228, "end": 233, "text": "2001)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Annotation Guidelines", "sec_num": "4.3" }, { "text": "Annotators have to label each trigger and its scope(s). Multiple triggers may have the same scope as in 9 where the two epistemic triggers \u202b\ufecb\ufe8e\u0631\u0641\u202c EArf (I know) and \u202b\ufee3\ufe98\ufe84\ufedb\ufeaa\u202c mt>kd (I'm sure) share the clause scope of that Mubarak won't be executed. Annotators have to label all the scopes of the modality trigger for type and then identify their spans. In 10, the obligative \u202b\ufefb\u0632\u0645\u202c lAzm (must) modifies three verbal-phrase scopes linked by the coordinating conjunction \u202b\u0648\u202c w (and). We must follow up with this, not ignore it and investigate it well so it won't happen again.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Annotation Guidelines", "sec_num": "4.3" }, { "text": "Finally, annotators have to retrieve implicit scopes whether they are referred to in-text or using their own real-world knowledge.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "10", "sec_num": null }, { "text": "Annotation proceeded in four stages. For Stage 0, we used our novel, manually-built, large-scale Arabic Modality Lexicon (AML) to automatically pre-highlight candidate modality triggers. AML was built in three steps: \uf0b7 First, we manually generated the person, gender, number, tense, mood and aspect inflections as well as the present and past participle derivations of 276 lemmas compiled from Mitchell and al-Hassan (1994) , Brustad (2000) and Moshref (2012) . \uf0b7 Second, we added a list of triggers including particles, adverbs and multi-word generic expressions that do not inflect for person, gender, number, tense, mood and aspect. \uf0b7 Finally, we labeled each entry for an English ly\" gloss=\"I think\" ambiguity=\"NA\" dialect=\"EA\" semClass=\"epistemic\" features=\"NA\" For Stage 1, annotators labeled each prehighlighted modality trigger for its modality semantic meaning. We defined modality semantic annotation as a synonymy judgment task where the annotators, given a number of synsets, had to decide to which synset the pre-highlighted trigger belongs. We used 8 synsets; each of which featured one modality semantic meaning from Palmer's (1989; 2001) typology. The average size of the synsets is 15 words/phrases to represent different shades of meaning. Yet, due to space limitations, we only included sample synsets in Table 3 . To avoid fatigue, disinterest, and distraction effects, we used counterbalancing and prompted the annotators to provide their own synonym(s) for the pre-highlighted candidate trigger if none of the given synsets seemed synonymous.", "cite_spans": [ { "start": 394, "end": 423, "text": "Mitchell and al-Hassan (1994)", "ref_id": null }, { "start": 426, "end": 440, "text": "Brustad (2000)", "ref_id": "BIBREF4" }, { "start": 445, "end": 459, "text": "Moshref (2012)", "ref_id": "BIBREF12" }, { "start": 1510, "end": 1525, "text": "Palmer's (1989;", "ref_id": "BIBREF13" }, { "start": 1526, "end": 1531, "text": "2001)", "ref_id": "BIBREF14" } ], "ref_spans": [ { "start": 1702, "end": 1709, "text": "Table 3", "ref_id": "TABREF9" } ], "eq_spans": [], "section": "Annotation Procedure", "sec_num": "4.4" }, { "text": "For Stage 2, annotators labeled the syntactic type of the linguistic constituents modified by the pre-highlighted modality trigger (i.e. scope type) where applicable. Annotators had to choose whether the modified constituent was a clause, a verbal phrase, or another type of constituency (e.g. a noun phrase, an adjectival phrase). Once the clause or the verbal phrase option was selected, annotators were prompted to extract that clause or verbal phrase. For Stage 3, we automatically extracted the triggers that followed our semantic-syntactic working definition of modality. That is, triggers labeled as synonymous to one of the synsets in Table 3 AND as modifying a clause or a verbal phrase. In this stage, instances such as \u202b\u0627\ufed3\ufe98\ufedc\ufeae\u0648\u0627\u202c AftkrwA in 11 were automatically excluded: although it modifies a complement clause, it means remember which is a non-modality meaning. Instances such as \u202b\ufed7\ufe8e\u062f\u0631\u202c qAdr (able to) in 13 were automatically admitted as valid modality triggers. It belongs to the dynamic abilitive synset, and modifies the verbal phrase \u202b\ufbfe\ufebc\ufee8\ufeca\u202c \u202b\u0623\u0646\u202c >n ySnE (to make).", "cite_spans": [], "ref_spans": [ { "start": 643, "end": 650, "text": "Table 3", "ref_id": "TABREF9" } ], "eq_spans": [], "section": "Annotation Procedure", "sec_num": "4.4" }, { "text": "The Egyptian people are able to make their own destiny.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": ". \u202b\u0627\ufedf\ufee4\ufebc\ufeae\u064a\u202c \u202b\u0627\ufedf\ufeb8\ufecc\ufe90\u202c \u202b\ufed7\ufe8e\u062f\u0631\u202c \u202b\ufe91\ufbff\ufeaa\u0647\u202c \u202b\ufee3\ufebc\ufbff\ufeae\u0647\u202c \u202b\ufbfe\ufebc\ufee8\ufeca\u202c \u202b\u0623\u0646\u202c Al$Eb AlmSry qAdr >n ySnE mSyrh bydh", "sec_num": "13" }, { "text": "Our annotation procedure pinpoints the efficiency and applicability of each dimension of our definition of modality. These guidelines also speed up the annotation process and increase annotation reliability because they provide annotators with practical and concrete prompts, and elicit well-structured answers that can be automatically converted into the modality annotation profiles described in Section 4.5.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": ". \u202b\u0627\ufedf\ufee4\ufebc\ufeae\u064a\u202c \u202b\u0627\ufedf\ufeb8\ufecc\ufe90\u202c \u202b\ufed7\ufe8e\u062f\u0631\u202c \u202b\ufe91\ufbff\ufeaa\u0647\u202c \u202b\ufee3\ufebc\ufbff\ufeae\u0647\u202c \u202b\ufbfe\ufebc\ufee8\ufeca\u202c \u202b\u0623\u0646\u202c Al$Eb AlmSry qAdr >n ySnE mSyrh bydh", "sec_num": "13" }, { "text": "Twitter terms of services prohibit redistributing raw tweet texts. Thus at the end of the annotation process, a profile was built for each tweet with its user name, tweet ID, and modalityrelated information. Associated software is to be given to help reconstruct tweets using their IDs. Although at the time of writing this paper all tweet IDs are still active, there is a potential of degradation if users delete their tweets or make their accounts private. This does not affect the modality-related profile, however, the complete tweet text will not be available. Modality-related information presents chunks of the tweet texts that represent the trigger word/phrase and the scope clause/verbal phrase. Thus we assume that we are not violating the terms of services.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Modality Annotation Profiles", "sec_num": "4.5" }, { "text": "Triggers are marked with 4-character labels. The first character is T for Trigger. The second two characters indicate the semantic meaning of the trigger {Ep: epistemic, Rp: reported, Sn: sensory, Ob: obligative, Pr: permissive, Cm: commissive, Ab: abilitive, Vl: volitive}. The fourth character is an index to indicate whether the trigger is the 1 st , 2 nd and so on in the tweet and to relate the trigger to its scope(s).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Modality Annotation Profiles", "sec_num": "4.5" }, { "text": "Scopes are marked with 3-character labels. The first is S for Scope. The second represents the syntactic type of the scope -{C: clause, P: verbal phrase}. The last is an index matching that of its trigger. Table 4 shows the modality annotation profiles for examples 1 and 9, respectively. user=\"alaa\" tweet_id=\" 71857458888458240\" [[ (\"zmAn\", \"TEP1\"), ((\"mbArk lsh byHkmnA\", \"SC1\")) ]] user=\"eAiNet\" tweet_id=\"46316910177697792\" [[ (\"EArf\", \"TEP1\"), ((\"An mbArk m$ HytEdm\", \"SC1\"))], [(\"mt>kd\", \"TEP2\"), ((\"An mbArk m$ HytEdm\", \"SC2\")) ]] Table 4 : Example modality annotation profiles 5 Annotation Results", "cite_spans": [], "ref_spans": [ { "start": 206, "end": 213, "text": "Table 4", "ref_id": null }, { "start": 539, "end": 546, "text": "Table 4", "ref_id": null } ], "eq_spans": [], "section": "Modality Annotation Profiles", "sec_num": "4.5" }, { "text": "AML pre-highlighted 2,892 candidate triggers in our 1,704 tweets. We used the kappa \u03ba statistics to measure the Inter-Annotator Agreement (IAA) rates for: \uf0b7 Modality semantic annotation: this labels each candidate trigger as synonymous to one of the synsets in Table 3 featuring Palmer's (1989; 2001) typology. \uf0b7 Modality syntactic annotation: this includes (1) identifying the scope type as to whether it is a clause, a verbal phrase, or none; and (2) identifying the scope span in terms of the beginning and the end of each scope.", "cite_spans": [ { "start": 279, "end": 294, "text": "Palmer's (1989;", "ref_id": "BIBREF13" }, { "start": 295, "end": 300, "text": "2001)", "ref_id": "BIBREF14" } ], "ref_spans": [ { "start": 261, "end": 268, "text": "Table 3", "ref_id": "TABREF9" } ], "eq_spans": [], "section": "Inter-Annotator Agreement Rates and Disagreement Factors", "sec_num": "5.1" }, { "text": "Our macro kappa \u03ba IAA rate for modality semantic annotation is 0.899 (Table 5 ). It is hard to measure if this rate is significantly higher than rates reported in the literature of modality semantic annotation because direct comparison with prior work is not possible as explained in Section 2. Yet, one point to highlight is that we do not use a collapsed typology of modality semantic meanings as in Rubenstein et al. 2013and Cui and Chi (2013), who both collapsed modality semantic meanings into two major classes only: priority vs. non-priority. We attribute our high IAA rate for modality semantic annotation to: (1) the large-scale AML, which provides annotators with an extensive list of candidate triggers; and (2) using synonymy judgments to give annotators practical, selfevident annotation prompts instead of subjective guidelines, defining modality triggers as expressions of alternative states in which the world could be.", "cite_spans": [], "ref_spans": [ { "start": 69, "end": 77, "text": "(Table 5", "ref_id": "TABREF11" } ], "eq_spans": [], "section": "Inter-Annotator Agreement Rates and Disagreement Factors", "sec_num": "5.1" }, { "text": "There are, however, two limitations to using synonymy judgments for modality semantic annotation. First, the quality of the annotation relies on the quality of the used synsets. It is important to select unambiguous triggers to represent the modality semantic meanings in different contexts. This is because triggers interact with other linguistic features such as modification, negation and grammatical mood. Second, to better guide the annotators, especially when working on a morphologically-rich language such as Arabic, it is better to have the synset members inflected for the same person, gender, number, tense, mood and aspect as the candidate trigger.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Inter-Annotator Agreement Rates and Disagreement Factors", "sec_num": "5.1" }, { "text": "It took us three iterations of annotations -each one with two different annotators -to come up with the best final synsets used in this paper. This process is time and labor consuming. Yet, once the synsets have been created, the annotation process is fast and replicable with a potential to be crowdsourced. We will examine this option in future work.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Inter-Annotator Agreement Rates and Disagreement Factors", "sec_num": "5.1" }, { "text": "Highly-ambiguous lemmas are the first disagreement factor for modality semantic annotation. Epistemic lemmas such as \u202b\ufeb7\ufe8e\u0641\u202c $Af (saw), \u202b\ufecb\ufeae\u0641\u202c Erf (knew), \u202b\ufed3\ufbad\ufee2\u202c fhm (understood), \u202b\ufebb\ufeaa\u0642\u202c Sdq (believed) and \u202b\ufed7\ufe8e\u0644\u202c qAl (said) among others have multiple meanings of which one or more might be modality-related. This explains why most of the disagreement scores in Table 6 are between modality and non-modality meanings (i.e. NA). For one annotator \u202b\ufee7\ufed4\ufbad\ufee2\u202c nfhm in 14 is synonymous to \u202b\ufee7\ufeb8\ufeae\u062d\u202c n$rH (explain) and thus does not belong to any of the modality synsets in Table 3 . For the other annotator, this trigger still means explain, but not as in explaining factual information, but as in making people adopt a specific point of view or a belief. Thus it is synonymous to \u202b\ufbfe\ufe86\ufee3\ufee8\ufeee\u202c \u202b\ufee7\ufea0\ufecc\ufee0\ufbad\ufee2\u202c \u202b\u0646\u202c \u202b\u0623\u0646\u202c njElhm y&mnwn >n (make them believe that) and is an epistemic trigger.", "cite_spans": [], "ref_spans": [ { "start": 355, "end": 362, "text": "Table 6", "ref_id": "TABREF12" }, { "start": 555, "end": 562, "text": "Table 3", "ref_id": "TABREF9" } ], "eq_spans": [], "section": "Inter-Annotator Agreement Rates and Disagreement Factors", "sec_num": "5.1" }, { "text": "We must explain to people that the Supreme Council of Armed Forces is one thing and the army is another.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "\u202b\ufee7\ufed4\ufbad\ufee2\u202c \u202b\u0627\ufedf\ufee8\ufe8e\u0633\u202c \u202b\ufea3\ufe8e\ufe9f\ufe94\u202c \u202b\u0648\u0627\ufedf\ufea0\ufbff\ufeb6\u202c \u202b\ufea3\ufe8e\ufe9f\ufe94\u202c \u202b\u0627\ufedf\ufecc\ufeb4\ufedc\ufeae\u064a\u202c \u202b\u0627\ufedf\ufee4\ufea0\ufee0\ufeb2\u202c \u202b\u0625\u0646\u202c lAbd nfhm AlnAs can be either we must protest and say no or it's necessary that we protest and say no. \uf0b7 Implicit scope recovery: implicit scopes with deictic expressions or in-text reference were easy to retrieve unlike implicature-based scopes. For instance, \u202b\ufea7\ufec4\ufe8e\ufe91\ufbab\u202c \u202b\ufebb\ufeaa\ufed7\ufee8\ufe8e\u202c \u202b\u0627\ufea3\ufee8\ufe8e\u202c AHnA SdqnA xTAbh (we believed his speech) was perceived by one annotator as we believed what he said in his speech was true. Thus the annotator selected the clause option for the scope type. The other annotator did not see such an implicature and thus selected NA, meaning that the scope is unrecognizable or is neither a clause nor a verbal phrase. The macro kappa \u03ba IAA rate for scope span recognition in Table 5 is 0.929 which is quite high. We attribute this to the simplicity of the tweet genre, which entails short sentences (140 characters or less) and a writing style that resembles short telegraphic notes more than formal and lengthy sentences. Interjections, adjuncts and subordinate conjunctions are the main reasons for disagreement. In 16, one annotator ends the span of the clause-based scope before \u202b\ufecb\ufee0\ufeb8\ufe8e\u0646\u202c El$An (so that); while the other includes the entire sentence into the scope span. We will add clearer guidelines for when interjections, adjuncts and subordinate clauses should be considered into the scope span in future work. People must go to protests (so that those already protesting won't feel as if left alone).", "cite_spans": [], "ref_spans": [ { "start": 79, "end": 86, "text": "Table 5", "ref_id": "TABREF11" }, { "start": 1117, "end": 1124, "text": "Table 5", "ref_id": "TABREF11" } ], "eq_spans": [], "section": "\u202b\ufee7\ufed4\ufbad\ufee2\u202c \u202b\u0627\ufedf\ufee8\ufe8e\u0633\u202c \u202b\ufea3\ufe8e\ufe9f\ufe94\u202c \u202b\u0648\u0627\ufedf\ufea0\ufbff\ufeb6\u202c \u202b\ufea3\ufe8e\ufe9f\ufe94\u202c \u202b\u0627\ufedf\ufecc\ufeb4\ufedc\ufeae\u064a\u202c \u202b\u0627\ufedf\ufee4\ufea0\ufee0\ufeb2\u202c \u202b\u0625\u0646\u202c lAbd nfhm AlnAs ly (I think) and \u202b\ufecb\ufe8e\ufbfe\ufeb0\u202c EAyz (I want), (2) 15.6% either MSA or EA based on context such as \u202b\ufed7\ufe8e\u0644\u202c qAl (he said) and \u202b\ufecb\ufeae\u0641\u202c Erf (he knew), and (3) 7% MSA-exclusive such as \u202b\u0627\ufeb3\ufe98\ufeee\ufe9f\ufe90\u202c Astwjb (it necessitated) and \u202b\u0648\u062f\u062f\u062a\u202c wddt (I wanted). Table 8 : Modality semantic meanings and scope types Ambiguity accounts for 37% and 5% of the valid modality lemmas for lexical and semantic ambiguity, respectively. Some of the most frequent lemmas from each ambiguity type are illustrated in Tables 9 and 10 Finally, about 82% of the scope heads are adjacent to their triggers. This is expected given that tweets are typically short.", "cite_spans": [], "ref_spans": [ { "start": 335, "end": 342, "text": "Table 8", "ref_id": null }, { "start": 578, "end": 593, "text": "Tables 9 and 10", "ref_id": "TABREF7" } ], "eq_spans": [], "section": "Annotated Corpus Statistics", "sec_num": "5.2" }, { "text": "Clause V", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Annotated Corpus Statistics", "sec_num": "5.2" }, { "text": "We presented a novel modality annotation scheme and applied it to the Arabic language in the tweets genre. This work is part of a larger project to use linguistic modality to detect power relations among participants on Twitter. The presented scheme uses both semantics and syntax to increase annotation reliability. Results show that Arabic modality triggers have regular subcategorization patterns that yield high annotation agreement when used as guidelines.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion and Outlook", "sec_num": "6" }, { "text": "Currently, we are working on an updated version of this corpus with improved guidelines to tackle disagreement factors that emerged here. The new version will also include annotations for modality holders and trigger-related features such as negation, modification and mood.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Conclusion and Outlook", "sec_num": "6" } ], "back_matter": [ { "text": "This work has been partially supported by a grant on social media and mobile computing from the Beckman Institute for Advanced Science and Technology.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Acknowledgement", "sec_num": null } ], "bib_entries": { "BIBREF0": { "ref_id": "b0", "title": "AWATIF: A Multi-Genre Corpus for Modern Standard Arabic Subjectivity and Sentiment Analysis", "authors": [ { "first": "Muhammad", "middle": [], "last": "Abdul-Mageed", "suffix": "" }, { "first": "Mona", "middle": [], "last": "Diab", "suffix": "" } ], "year": 2012, "venue": "The 8th International Conference on Language Resources and Evaluation (LREC'12)", "volume": "", "issue": "", "pages": "21--27", "other_ids": {}, "num": null, "urls": [], "raw_text": "Muhammad Abdul-Mageed and Mona Diab. 2012. AWATIF: A Multi-Genre Corpus for Modern Standard Arabic Subjectivity and Sentiment Analy- sis. The 8th International Conference on Language Resources and Evaluation (LREC'12), Istanbul, Turkey, 21-27 May 2012.", "links": null }, "BIBREF1": { "ref_id": "b1", "title": "YADAC: Yet another Dialectal Arabic Corpus", "authors": [ { "first": "Rania", "middle": [], "last": "Al", "suffix": "" }, { "first": "-", "middle": [], "last": "Sabbagh", "suffix": "" }, { "first": "Roxana", "middle": [], "last": "Girju", "suffix": "" } ], "year": 2012, "venue": "The 8th International Conference on Language Resources and Evaluation (LREC'12)", "volume": "", "issue": "", "pages": "21--27", "other_ids": {}, "num": null, "urls": [], "raw_text": "Rania Al-Sabbagh and Roxana Girju. 2012. YADAC: Yet another Dialectal Arabic Corpus. The 8th In- ternational Conference on Language Resources and Evaluation (LREC'12), Istanbul, Turkey, 21-27 May 2012.", "links": null }, "BIBREF2": { "ref_id": "b2", "title": "A Modality Lexicon and its Use in Automatic Tagging", "authors": [ { "first": "Kathrin", "middle": [], "last": "Baker", "suffix": "" }, { "first": "Michael", "middle": [], "last": "Bloodgood", "suffix": "" }, { "first": "Bonnie", "middle": [], "last": "Dorr", "suffix": "" }, { "first": "Nathaniel", "middle": [ "W" ], "last": "Filardo", "suffix": "" }, { "first": "Lori", "middle": [], "last": "Levin", "suffix": "" }, { "first": "Christine", "middle": [], "last": "Piatko", "suffix": "" } ], "year": 2010, "venue": "The 7th International Conference on Language Resources and Evaluation (LREC'10)", "volume": "", "issue": "", "pages": "19--21", "other_ids": {}, "num": null, "urls": [], "raw_text": "Kathrin Baker, Michael Bloodgood, Bonnie Dorr, Nathaniel W. Filardo, Lori Levin, and Christine Piatko. 2010. A Modality Lexicon and its Use in Automatic Tagging. The 7th International Confer- ence on Language Resources and Evaluation (LREC'10), Valletta, Malta, 19-21 May 2010.", "links": null }, "BIBREF3": { "ref_id": "b3", "title": "Root and Epistemic Modal Auxiliary Verbs", "authors": [ { "first": "Virginia", "middle": [], "last": "Brennan", "suffix": "" } ], "year": 1993, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Virginia Brennan. 1993. Root and Epistemic Modal Auxiliary Verbs, PhD Thesis, University of Massa- chusetts, Amherst.", "links": null }, "BIBREF4": { "ref_id": "b4", "title": "The Syntax of Spoken Arabic: A Comparative Study of Moroccan, Egyptian, Syrian and Kuwaiti Dialects", "authors": [ { "first": "E", "middle": [], "last": "Kristen", "suffix": "" }, { "first": "", "middle": [], "last": "Brustad", "suffix": "" } ], "year": 2000, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Kristen E. Brustad. 2000. The Syntax of Spoken Ara- bic: A Comparative Study of Moroccan, Egyptian, Syrian and Kuwaiti Dialects. Georgetown Univer- sity Press, Washington DC, USA.", "links": null }, "BIBREF6": { "ref_id": "b6", "title": "The Evolution of Grammar: Tense, Aspect and Modality in the Languages of the World", "authors": [ { "first": "Joan", "middle": [ "L" ], "last": "Bybee", "suffix": "" }, { "first": "R", "middle": [ "D" ], "last": "Perkins", "suffix": "" }, { "first": "W", "middle": [], "last": "Pagliuca", "suffix": "" } ], "year": 1994, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Joan L. Bybee, R. D. Perkins and W. Pagliuca. 1994. The Evolution of Grammar: Tense, Aspect and Modality in the Languages of the World. University of Chicago Press, Chicago, USA.", "links": null }, "BIBREF7": { "ref_id": "b7", "title": "Commited Belief Annotation and Tagging", "authors": [ { "first": "Mona", "middle": [], "last": "Diab", "suffix": "" }, { "first": "Lori", "middle": [], "last": "Levin", "suffix": "" }, { "first": "Teruko", "middle": [], "last": "Mitamura", "suffix": "" }, { "first": "Owen", "middle": [], "last": "Rambow", "suffix": "" }, { "first": "Vinodkumar", "middle": [], "last": "Prabhakaran", "suffix": "" }, { "first": "Wei", "middle": [ "Wei" ], "last": "Guo", "suffix": "" } ], "year": 2009, "venue": "Proceedings of the Third Linguistic Annotation Workshop (ACL-IJCNLP '09)", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Mona Diab, Lori Levin, Teruko Mitamura, Owen Rambow, Vinodkumar Prabhakaran and Wei Wei Guo. 2009. Commited Belief Annotation and Tagging. Proceedings of the Third Linguistic Annotation Workshop (ACL-IJCNLP '09), August 2009, Suntec, Singapore.", "links": null }, "BIBREF8": { "ref_id": "b8", "title": "Modality in Text: A Proposal for Corpus Annotation", "authors": [ { "first": "Iris", "middle": [], "last": "Hendrickx", "suffix": "" }, { "first": "Am\u00e1lia", "middle": [], "last": "Mendes", "suffix": "" }, { "first": "Silvia", "middle": [], "last": "Mencarelli", "suffix": "" } ], "year": 2010, "venue": "The 7th International Conference on Language Resources and Evaluation (LREC'10)", "volume": "", "issue": "", "pages": "19--21", "other_ids": {}, "num": null, "urls": [], "raw_text": "Iris Hendrickx, Am\u00e1lia Mendes, and Silvia Mencarelli. 2010. Modality in Text: A Proposal for Corpus Annotation. The 7th International Confer- ence on Language Resources and Evaluation (LREC'10), Valletta, Malta, 19-21 May 2010", "links": null }, "BIBREF9": { "ref_id": "b9", "title": "Semantic Interpretation in Generative Grammar", "authors": [ { "first": "Ray", "middle": [], "last": "Jackendoff", "suffix": "" } ], "year": 1972, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Ray Jackendoff. 1972. Semantic Interpretation in Generative Grammar. Cambridge, MA: The MIT Press.", "links": null }, "BIBREF10": { "ref_id": "b10", "title": "Annotating Event Mentions in Text with Modality, Focus, and Source Information", "authors": [ { "first": "Matsuyoshi", "middle": [], "last": "Suguru", "suffix": "" }, { "first": "Megumi", "middle": [], "last": "Eguchi", "suffix": "" }, { "first": "Chitose", "middle": [], "last": "Sao", "suffix": "" }, { "first": "Koji", "middle": [], "last": "Murakami", "suffix": "" } ], "year": 2010, "venue": "The 7th International Conference on Language Resources and Evaluation (LREC'10)", "volume": "", "issue": "", "pages": "19--21", "other_ids": {}, "num": null, "urls": [], "raw_text": "Suguru, Matsuyoshi, Megumi Eguchi, Chitose Sao, Koji Murakami, Kentaro Inui, and Yuji Matsumo- to. 2010. Annotating Event Mentions in Text with Modality, Focus, and Source Information. The 7th International Conference on Language Resources and Evaluation (LREC'10), Valletta, Malta, 19-21 May 2010", "links": null }, "BIBREF12": { "ref_id": "b12", "title": "Corpus Study of Tense, Aspect, and Modality in Diaglossic Speech in Cairene Arabic", "authors": [ { "first": "Ola", "middle": [], "last": "Moshref", "suffix": "" } ], "year": 2012, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Ola Moshref. 2012. Corpus Study of Tense, Aspect, and Modality in Diaglossic Speech in Cairene Ara- bic. PhD Thesis. University of Illinois at Urbana- Champaign.", "links": null }, "BIBREF13": { "ref_id": "b13", "title": "Mood and Modality", "authors": [ { "first": "R", "middle": [], "last": "Frank", "suffix": "" }, { "first": "", "middle": [], "last": "Palmer", "suffix": "" } ], "year": 1989, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Frank R. Palmer. 1989. Mood and Modality. Cam- bridge University Press, Cambridge, UK.", "links": null }, "BIBREF14": { "ref_id": "b14", "title": "Mood and Modality. 2 nd Edition", "authors": [ { "first": "R", "middle": [], "last": "Frank", "suffix": "" }, { "first": "", "middle": [], "last": "Palmer", "suffix": "" } ], "year": 2001, "venue": "", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Frank R. Palmer. 2001. Mood and Modality. 2 nd Edi- tion. Cambridge University Press, Cambridge, UK.", "links": null }, "BIBREF15": { "ref_id": "b15", "title": "Detecting Power Relations from Written Dialog", "authors": [ { "first": "", "middle": [], "last": "Vinodkumar Prabhakaran", "suffix": "" } ], "year": 2012, "venue": "Proceedings of the 2012 Student Research Workshop", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Vinodkumar Prabhakaran. 2012. Detecting Power Relations from Written Dialog. Proceedings of the 2012 Student Research Workshop, Jeju, Republic of Korea.", "links": null }, "BIBREF16": { "ref_id": "b16", "title": "Statistical Modality Tagging from Rule-Based Annotations and Crowdsourcing", "authors": [ { "first": "Michael", "middle": [], "last": "Vinodkumar Prabhakaran", "suffix": "" }, { "first": "Mona", "middle": [], "last": "Bloodgood", "suffix": "" }, { "first": "Bonnie", "middle": [], "last": "Diab", "suffix": "" }, { "first": "Lori", "middle": [], "last": "Dorr", "suffix": "" }, { "first": "Christine", "middle": [ "D" ], "last": "Levin", "suffix": "" }, { "first": "Owen", "middle": [], "last": "Piatko", "suffix": "" }, { "first": "Benjamin", "middle": [], "last": "Rambow", "suffix": "" }, { "first": "", "middle": [], "last": "Van Durme", "suffix": "" } ], "year": 2012, "venue": "Proceedings of the ACL-2012 Workshop on Extra-Propositional Aspects of Meaning in Computational Linguistics", "volume": "", "issue": "", "pages": "", "other_ids": {}, "num": null, "urls": [], "raw_text": "Vinodkumar Prabhakaran, Michael Bloodgood, Mona Diab, Bonnie Dorr, Lori Levin, Christine D. Piatko, Owen Rambow and Benjamin Van Durme. 2012. Statistical Modality Tagging from Rule- Based Annotations and Crowdsourcing. Proceedings of the ACL-2012 Workshop on Extra- Propositional Aspects of Meaning in Computational Linguistics (ExProM-2012), Jeju, Republic of Korea.", "links": null }, "BIBREF17": { "ref_id": "b17", "title": "Annotating and Recognizing Event Modality in Text", "authors": [ { "first": "Roser", "middle": [], "last": "Saur\u00ed", "suffix": "" }, { "first": "March", "middle": [], "last": "Verhagen", "suffix": "" }, { "first": "James", "middle": [], "last": "Pustejovsky", "suffix": "" } ], "year": 2006, "venue": "The 19th International FLAIRS Conference", "volume": "", "issue": "", "pages": "335--339", "other_ids": {}, "num": null, "urls": [], "raw_text": "Roser Saur\u00ed, March Verhagen and James Pustejovsky. 2006. Annotating and Recognizing Event Modality in Text. The 19th International FLAIRS Confer- ence, pages: 335-339, Florida, USA, May 2006", "links": null }, "BIBREF18": { "ref_id": "b18", "title": "The BioScope Corpus: Annotation for Negation, Uncertainty and their Scope in Biomedical Texts", "authors": [ { "first": "Gy\u00f6rgy", "middle": [], "last": "Szarvas", "suffix": "" }, { "first": "Veronika", "middle": [], "last": "Vincze", "suffix": "" }, { "first": "Rich\u00e1rd", "middle": [], "last": "Farkas", "suffix": "" }, { "first": "J\u00e1nos", "middle": [], "last": "Csirik", "suffix": "" } ], "year": 2008, "venue": "Current Trends in Biomedical Natural Language Processing", "volume": "", "issue": "", "pages": "38--45", "other_ids": {}, "num": null, "urls": [], "raw_text": "Gy\u00f6rgy Szarvas, Veronika Vincze, Rich\u00e1rd Farkas, and J\u00e1nos Csirik. 2008. The BioScope Corpus: Annotation for Negation, Uncertainty and their Scope in Biomedical Texts. BioNLP 2008: Current Trends in Biomedical Natural Language Pro- cessing, pages: 38-45, Columbus, Ohio, USA, June 2008", "links": null }, "BIBREF19": { "ref_id": "b19", "title": "Annotating Expressions of Opinions and Emotions in Language", "authors": [ { "first": "Janyce", "middle": [], "last": "Wiebe", "suffix": "" }, { "first": "Theresa", "middle": [], "last": "Wilson", "suffix": "" }, { "first": "Claire", "middle": [], "last": "Cardie", "suffix": "" } ], "year": 2005, "venue": "Language Resources and Evaluations", "volume": "39", "issue": "", "pages": "165--210", "other_ids": {}, "num": null, "urls": [], "raw_text": "Janyce Wiebe, Theresa Wilson, and Claire Cardie. 2005. Annotating Expressions of Opinions and Emotions in Language. Language Resources and Evaluations, 39:165-210", "links": null } }, "ref_entries": { "TABREF0": { "content": "
\u202b\ufe91\ufbff\ufea4\ufedc\ufee4\ufee8\ufe8e\u202c\u202b\ufedf\ufeb4\ufbab\u202c \u202b\ufee3\ufe92\ufe8e\u0631\u0643\u202c\u202b\u0632\ufee3\ufe8e\u0646\u202c\u202b\ufedb\ufe8e\u0646\u202c\u202b\ufea7\ufe92\ufeae\u0629\u202c \u202b\ufecb\ufee8\ufeaa\u06be\ufe8e\u202c\u202b\u0627\ufedf\ufee0\ufef2\u202c \u202b\u0627\ufedf\ufee8\ufe8e\u0633\u202c \u202b\ufedb\ufefc\u0645\u202c\u202b\ufeb3\ufee4\ufecc\ufee8\ufe8e\u202c \u202b\ufedf\ufeee\u202c.1
lw smEnA klAm AlnAs Ally EndhA xbrp kAn zmAn
mbArk lsh byHkmnA
If we'd listened to the elite, Mubarak would have been
still ruling us.\u202b\ufea7\ufefc\u0635\u202c\u202b\u0627\ufee7\ufe98\ufbad\ufef0\u202c \u202b\u0627\ufedf\ufed4\ufeae\u062f\u064a\u202c \u202b\u0627\ufedf\ufea4\ufedc\ufee2\u202c\u202b\u0632\ufee3\ufe8e\u0646\u202c.2
zmAn AlHkm Alfrdy AnthY xlAS
", "num": null, "type_str": "table", "html": null, "text": "The era of individual rulers has come to an end." }, "TABREF1": { "content": "
\u202b\u0627\ufedf\ufec4\ufe92\ufef2\u202c \u202b\ufeb3\ufeae\ufbfe\ufeae\u0647\u202c \u202b\ufecb\ufee0\ufef0\u202c \u202b\ufbfe\ufeb4\ufe98\ufee0\ufed8\ufef2\u202c \u202b\u0648\u06be\ufeee\u202c\u202b\ufea3\ufeb4\ufbff\ufee6\u202c \u202b\ufebb\ufeaa\u0627\u0645\u202c \u202b\ufe97\ufeac\ufedb\ufeae\u202c\u202b\u0648\u202c \u202b\u0623\ufee7\ufbab\u202c\u202b\ufefb\ufe91\ufeaa\u202c\u202b\ufee3\ufe92\ufe8e\u0631\u0643\u202c\u202b\u0646\u202c\u202b\u0625\u202c.3
\u202b\u0627\ufedf\ufee4\ufea4\ufedc\ufee4\ufe94\u202c \u202b\ufed7\ufe8e\ufecb\ufe94\u202c \u202b\ufed3\ufef2\u202c
<n mbArk lAbd w>nh tzkr SdAm Hsyn whw ystlqy ElY
sryrh AlTby fy qAEp AlmHkmp
It must be that 4 . \u202b\ufefb\ufe91\ufeaa\u202c \u202b\u0627\u0646\u202c \u202b\ufbfe\ufeee\ufebf\ufeca\u202c \u202b\ufee3\ufe92\ufe8e\u0631\u0643\u202c \u202b\u0627\ufedf\ufeb4\ufea0\ufee6\u202c \u202b\ufed3\ufef0\u202c
lAbd An ywDE mbArk fY Alsjn
Mubarak must be put in jail.
", "num": null, "type_str": "table", "html": null, "text": "Mubarak remembered Saddam Hussein as he was lying on his medical bed in the court." }, "TABREF7": { "content": "
gloss, ambiguities {lexical, semantic, both,
NA}, dialects {MSA, EA, both}, modality
semantic meaning and special features {quasi;
inherently-negative, multi-word expression}
as in Table 1.
Currently, AML has 7,584 entries, with the
statistical distributions in Table 2. Despite the
large size of AML, annotators were instructed to
add any words or phrases that match our working
semantic-syntactic definition of modality.
Semantic meaningsAmbiguity
Epistemic3,144 Lexical2,363
Sensory134Semantic155
Reported427Lexical/Semantic116
Obligative1,091 Unambiguous4,950
Permissive815
Commissive132
Abilitive957
Volitive884
DialectsSpecial Features
MSA2,268 Quasi777
EA3,100 Inherently-Neg.788
MSA/EA2,216 MWE276
Table 2: AML statistics
", "num": null, "type_str": "table", "html": null, "text": "An expert from the Arabic Modality Lexicon (AML)" }, "TABREF9": { "content": "", "num": null, "type_str": "table", "html": null, "text": "Sample synsets used for modality semantic annotation" }, "TABREF11": { "content": "
", "num": null, "type_str": "table", "html": null, "text": "Macro kappa \u03ba inter-annotator agreement rates for modality semantic and syntactic annotations" }, "TABREF12": { "content": "
", "num": null, "type_str": "table", "html": null, "text": "Confusion matrix for semantic annotation" }, "TABREF15": { "content": "
477231533548
Rp1976733243
Sn4912120172
Ob54771521167
Pr154751178
Cm20046
Ab39749113
Vl8016701248
NA3660760802
Total91342448992 2377
", "num": null, "type_str": "table", "html": null, "text": "" }, "TABREF16": { "content": "
Lemma \u202b\ufed7\ufe8e\u0644\u202cqAl17140
\u202b\ufee7\ufed4\ufeb2\u202cnfs56186
\u202b\ufeb7\ufe8e\u0641\u202c \u202b\ufed3\ufbad\ufee2\u202c$Af fhm80 88129 45
Table 9: Top frequent lexically-ambiguous lemmas
Lemma Trans. Modality meaningsFreq.
\u202b\ufeb7\ufe8e\u0641\u202c$Afepistemic (think) sensory (watch; witness)47 95
\u202b\ufecb\ufeae\u0641\u202cErfepistemic (know) abilitive (can)41 40
\u202b\ufee3\ufee4\ufedc\ufee6\u202cmmknepistemic (possible that) abilitive (can)28 17
Table 10: Top frequent semantically-ambiguous lemmas
", "num": null, "type_str": "table", "html": null, "text": "." } } } }