Detecting Irony in Shakespeare's Sonnets with SPARSAR

Abstract

English. In this paper we propose a novel approach to irony detection in Shakespeare's Sonnets, a well-known data set that is statistically valuable. In order to produce a meaningful experiment, we created a gold standard by collecting opinions from famous literary critics on the same data focusing on irony. In the experiment, we use SPARSAR a system for English poetry analysis and reciting by TTS. The system produces a deep linguistically based representation at phonetic, syntactic and semantic level. It has been used to detect irony with a novel approach based on phonetic processing and sentiment analysis. At first the evaluation was very disappointing, only 50% of the sonnets matched the gold standard. Eventually, taking advantage of the semantic representation produced by the system at propositional level, the logical structure of the sonnet has been highlighted by computing the discourse relations of the couplet and/or the final quatrain. In this way we managed to improve accuracy by 17% up to 66.88%1.

-

Italiano. In questo articolo si propone un nuovo approccio per l’individuazione dell’ironia nei Sonetti di Shakespeare, un dataset che è statisticamente valido. Allo scopo di produrre esperimenti significativi, abbiamo creato un gold standard raccogliendo le opinioni di famosi critici letterari sullo stesso corpus, con l’ironia come tema. Nell’esperimento abbiamo usato SPARSAR un sistema per l’analisi e la 1Copyright c 2019 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0) recitazione della poesia inglese con TTS. Il sistema produce una rappresentazione linguistica profonda a livello fonetico, sintattico e semantico. E’ stata usata per individuare l’ironia sulla base dell’analisi fonetica e del sentiment. All’inizio la valutazione è stata molto deludente, solo il 50% di tutti i sonetti erano inclusi nel gold standard. Poi sulla base della rappresentazione semantica prodotta dal sistema a livello proposizionale, è stata messa in luce la struttura logica del sonetto calcolando le relazioni del discorso del distico e/o della quartina finale. In questo modo abbiamo ottenuto un miglioramento dell’accuracy del 17% raggiungendo il 66.88%.

1 Introduction

Shakespeare’s Sonnets are a collection of 154 poems which is renowned for being full of ironic content (Weiser, 1983), (Weiser, 1987) and for its ambiguity thus sometimes reverting the overall interpretation of the sonnet. Lexical ambiguity, i.e. a word with several meanings, emanates from the way in which the author uses words that can be interpreted in more ways not only because inherently polysemous, but because sometimes the additional meaning it evokes is derived on the basis of the sound, i.e. by homophones (see “eye”, “I” in sonnet 152). The sonnets are also full of metaphors which many times require contextualising the content to the historical Elizabethan life and society. Furthermore, the sonnets are full of words related to specific language domains. For instance, there are words related to the language of economy, war, nature and to the discoveries of the modern age, and each of these words may be used as a metaphor of love. Many of the sonnets are organized around a conceptual contrast, an opposition that runs parallel and then diverges, sometimes with the use of the rhetorical figure of the chiasmus. It is just this contrast that generates irony, sometimes satire, sarcasm, and even parody. Irony may be considered in turn as: what one means using language that normally signifies the opposite, typically for humorous or emphatic effect; a state of affairs or an event that seems contrary to what one expects and is amusing as a result. As to sarcasm this may be regarded the use of irony to mock or convey contempt. (Attardo, 1994) Parody is obtained by using the words or thoughts of a person but adapting them to a ridiculously inappropriate subject. There are several types of irony, though we select verbal irony which, in the strict sense, is saying the opposite of what you mean for outcome, and it depends on the extralinguistics context. It is important to remark that in many cases, the linguistic structures on which irony is based, may require the use of nonliteral or figurative language, i.e. the use of metaphors. In our approach we will follow the so-called incongruity presumption or incongruity-resolution presumption. Theories connected to the incongruity presumption are mostly cognitive-based and related to concepts highlighted for instance, in (Attardo, 2000) . The focus of theorization under this presumption is that in humorous texts, or broadly speaking in any humorous situation, there is an opposition between two alternative dimensions. As a result, in our study of the sonnets, produced by the contents of manual classification, we have been looking for contrasting situations; while in the sentiment analysis experiment, we have been concerned with a quantitative count of polarity related items.

Computational research on sentiment analysis has been based on the use of shallow features with a binary choice to train statistical model (Carvalho et al., 2009) that, when optimized for a particular task, will produce acceptable performance. However generalizing the model has proven to be a hard task. In addition, the text addressed by recent research has been limited to tweets, which are in no way comparable to the sonnets contain a lot of nonliteral language. The other common approach used to detect irony, in the majority of the cases, is based on polarity detection (Van Hee et al., 2018) . Sentiment Analysis (Kim and Hovy, 2004) and (Kao and Jurafsky, 2012) is in fact an indiscriminate labeling of texts either on a lexicon basis or on a supervised feature basis where in both cases, it is just a binary - ternary or graded - decision that has to be taken. This is certainly not explanatory of the phenomenon and will not help in understanding what it is that causes humorous reactions to the reading of an ironic piece of text. It certainly is of no help in deciding which phrases, clauses or just multiwords or simply words, contribute to create the ironic meaning (see (Reyes et al., 2012) ; (Reyes and Rosso, 2013) ).

We will not comment here on the work done to produce the gold standard which has already been described in a separate paper (Busetto & Delmonte, 2019 - To appear) but see all the file in the Supplementary materials). We simply say that we considered as ironic or sarcastic all sonnets that have been so defined by at least one of the many literary critics’ comments we looked into2. 2

The Architecture of SPARSAR:

Syntax and Semantics

SPARSAR3 (Delmonte, 2016) builds three representations of the properties and features of each poem: a Phonetic Relational View from the phonological and the phonetic content of each word; a Poetic Relational View where the main poetic devices are addressed, related to rhythm and rhyme, and the overall metrical structure; then a Semantic Relational View where the syntactic, semantic and pragmatic content of the poem is represented, at the lexical semantic level, at the anaphoric level and at the predicate-argument structure. At this level, also the sentiment or overall mood of the poem is computed on the basis of a lean lexically based sentiment analysis. The system uses a modified version of VENSES, a semantically oriented NLP pipeline (Delmonte et al., 2005) . It is accompanied by a module that works at sentence level and produces a whole set of analysis both at quantitative, syntactic and semantic level. As regards syntax, the system makes available chunks and dependency structures. Then the system introduces semantics both in the version of a classifier and by isolating verbal complex in order to verify propositional properties, like presence of negation, to compute factuality from a 2We used criticism from a set of authors including (Frye, 1957) (Calimani, 2009) (Melchiori, 1971) (Eagle, 1916) (Marelli, 2015) (Schoenfeldt, 2010) (Weiser, 1987) (Serpieri, 2002) all listed in the reference section.

3the system is freely downloadable from its website https://sparsar.wordpress.com/ crosscheck with modality, aspectuality – that is derived from the lexica – and tense. On the other hand, the classifier has two different tasks: separating concrete from abstract nouns, identifying highly ambiguous from singleton concepts (from number of possible meanings from WordNet and other similar repositories). Eventually, the system carries out a sentiment analysis of the poem, thus contributing a three-way classification: neutral, negative, positive that can be used as a powerful tool for prosodically related purposes. State of the art semantic systems are based on different theories and representations, but the final aim of the workshop was reaching a consensus on what constituted a reasonably complete semantic representation. Semantics in our case not only refers to predicate-argument structure, negation scope, quantified structures, anaphora resolution and other similar items. It is referred essentially to a propositional level analysis, which is the basis for discourse structure and discourse semantics contained in discourse relations. It also paves the way for a deep sentiment or affective analysis of every utterance, which alone can take into account the various contributions that may come from syntactic structures like NPs and APs, where affectively marked words may be contained. Their contribution needs to be computed in a strictly compositional manner with respect to the meaning associated to the main verb, where negation may be lexically expressed or simply lexically incorporated in the verb meaning itself. The system does low level analyses before semantic modules are activated, that is tokenization, sentence splitting, multiword creation from a large lexical database. Then chunking and syntactic constituency parsing which is done using a rule-based recursive transition network: the parser works in a cascaded recursive way to include higher syntactic structures up to sentence and complex sentence level. These structures are then passed to the first semantic mapping algorithm that looks for subcategorization frames in the lexica made available for English, including VerbNet, FrameNet, WordNet and a proprietor lexicon of some 10K entires, with most frequent verbs, adjectives and nouns, containing also a detailed classification of all grammatical or function words. This mapping is done following LFG principles (Bresnan, 1982) (Bresnan, 2001) , where c-structure is mapped onto fstructure thus obeying uniqueness, completeness and coherence. The output of this mapping is a rich dependency structure, which contains information related also to implicit arguments, i.e. subjects of infinitivals, participials and gerundives. LFG representation also has a semantic role associated to each grammatical function, which is used to identify the syntactic head lemma uniquely in the sentence. Finally it takes care of long distance dependencies for relative and interrogative clauses. When fully coherent and complete predicate argument structures have been built, pronominal binding and anaphora resolution algorithms are fired. Coreferential processed are activated at the semantic level: they include a centering algorithm for topic instantiation and memorization that we do using a three-place stack containing a Main Topic, a Secondary Topic and a Potential Topic. Main Topics are chosen as best candidates for free pronominals - as long as morphological features are matching. In order to become a Main Topic, a Potential Topic must be reiterated. Discourse Level computation is done at propositional level by building a vector of features associated to the main verb of each clause. They include information about tense, aspect, negation, adverbial modifiers, modality. These features are then filtered through a set of rules which have the task to classify a proposition as either objective/subjective, factual/nonfactual, foreground/background. In addition, every lexical predicate is evaluated with respect to a class of discourse relations. Eventually, discourse structure is built, according to criteria of clause dependency where a clause can be classified either as coordinate or subordinate. Factuality is used to set apart opinions from facts and subjectivity is also used to contribute positively to the choice of expressing ironic content. 3

The Architecture of SPARSAR:

Phonetics and Poetic Devices

The second module is a rule-based system that converts graphemes of each poem into phonetic characters, it divides words into stressed/unstressed syllables and computes rhyming schemes at line and stanza level. To this end it uses grapheme to phoneme translations made available by different sources, amounting to some 500K entries, and include CMU dictionary 4, MRC Psycholinguistic Database 5, Celex Database (H. et al., 1995) , plus a proprietor database made of some 20,000 entries. Out of vocabulary words are computed by means of a prosodic parser implemented in a previous project (Bacalu and Delmonte, 1999) containing a big pronunciation dictionary which covers 170,000 entries approximately. Besides the need to cover the majority of grapheme to phoneme conversions by the use of appropriate dictionaries, remaining problems to be solved are related to ambiguous homographs like “import” (verb) and “import” (noun) and are treated on the basis of their lexical category derived from previous tagging. Eventually there is always a certain number of Out Of Vocabulary (OOV) words. The simplest case is constituted by differences in spelling determined by British vs. American pronunciation. This is taken care of by a dictionary of graphemic correspondances. However, whenever the word is not found the system proceeds by morphological decomposition, splitting at first the word from its prefix and if that still does not work, its derivational suffix. As a last resource, an orthographically based version of the same dictionary is used to try and match the longest possible string in coincidence with current OOVW. Then the remaining portion of word is dealt with by guessing its morphological nature, and if that fails a grapheme-to-phoneme parser is used. Some words thus reconstructed are wayfarer, gangrened, krog, copperplate, splendor, filmy, seraphic, unstarred.

Other words we had to reconstruct are: shrive, slipstream, fossicking, unplotted, corpuscle, thither, wraiths, etc. In some cases, the problem that made the system fail was the presence of a syllable which was not available in VESD, our database of syllable durations. This problem has been coped with partly by manually inserting the missing syllable and by computing its duration from the component phonemes; but also from the closest similar syllable available in the database. We only had to add 12 new syllables for a set of approximately 1000 poems that the system computed. The system has no limitation on 4It is available online at <http://www.speech.cs.cmu.edu/cgi-bin/cmudict/> 5Previously, data for POS were merged in from a different dictionary (MRC Psycholinguistic Database, <http://lcb.unc.edu/software/multimrc/multimrc.zip>, which uses British English pronunciation) type of poetic and rhetoric devices, however it is dependent on language: Italian line verse requires a certain number of beats and metric accents which are different from the ones contained in an English iambic pentameter. Rules implemented can demote or promote word-stress on a certain syllable depending on selected language, linelevel syllable length and contextual information. This includes knowledge about a word being part of a dependency structure either as dependent or as head. 4

The Experiment for the Automatic

Annotation of the Sonnets using

SPARSAR The experiment we devised was organized as follows: we downloaded SPARSAR from its dedicated website https://sparsar.wordpress.com/. At first, following (Tsur, 1992) , pag.15 and (Fonagy, 1971) , and on the basis of the complete Phonological description of each word in the poem (see (Delmonte, 2016) ), the system creates a relation between sound and mood or attitude by means of the module for sentiment analysis. In particular, it collapses together unvoiced, obstruent consonants with high and back vowels to represent hatred and struggle, mystic obscurity, sad and aggressive mood; the opposite is represented by voiced, sonorants and continuants consonants associated to low and front vowels. These oppositions are then applied to the one created by polarity values, negative vs. positive. We use these quantities to check an existing correlation, by using ratios. Basic relations are reported already in (Delmonte, 2016) , where however mood of each sonnet was manually computed. We report here relations intervening between the output of the system, comparing ratios derived from sound relations with those from polarity. As said above, polarity values are computed according to a lexicalized approach to sentiment analysis which takes into account also negation at propositional level (see (Taboada et al., 2011) A ratio lower than 1 indicates a majority of Negative items, higher than 1 a majority of Positive items. The same would apply to the remaining ratios. We compute the mean value for the three indices – Contrasting Vowels, Contrasting Consonants, Contrasting Voicing to indicate a generic sound related mood, Positive when the mean is higher than 1 and negative when it is lower. We then compare Results for polarity from sentiment analysis with those obtained from sound evaluations. We mark sonnets with a clash between the two parameters with 1 and with 0 whenever they converge to the same value. From a perusal of the results, a total of 79 sonnets over 98 have a clash, amounting to a remarkably high percentage of 80%. However when we check the system output with the critics’ choice we come up with a different picture: only 77 of all sonnets match with critics opinion, i.e. exactly 50%. This is the list of those 77 sonnets that have been found to match between the critics’ list and the list of the sonnets recognized by the system as having some kind of contrast:

1 2 4 5 6 10 12 14 17 18 19 20 21 27 30 32 33 34 35 37 41 42 47 48 50 56 57 61 65 67 68 69 71 72 74 75 77 78 79 81 82 84 87 92 95 97 98 101 102 104 106 108 109 111 113 114 115 116 123 125 126 127 129 134 136 137 139 142 144 145 146 149 151 152 153 154 4.1

Extracting Couplets from Logical

Structure

Considering the low accuracy reached with the purely quantitative approach, we decided to look into the semantic output of the system. We deemed that one of the possible reasons for the relatively low accuracy of the system could be the use of quantities to generate abstract evaluations: in other words, it is not always the case that a contrast is to be found by counting number of negative vs. positive items present in the sonnet. As to semantic representation created by SPARSAR, we are here referring to the logical structure of the Elizabethan sonnet where the argumentation is developed into three sections and the conclusion usually comes in the final couplet. This conclusion may revert the contents of the logical order as defined by the premises. The poet may defer the conclusion in the couplet to complete the logical argumentation by adding some further motivation. But in some cases the couplet is used to provoke surprise in the reader/hearer, accompanied by laughter or by indignation whenever sarcasm is intended. So eventually the opposition may only be present in the final two lines, and be hinted at by presence of discourse markers like “Yet”, “But”. In that case, it will not be sufficient for the system to ascertain the required quantity for a contrast, unless some specific rule is inserted that triggers such unexpected, unpredictable ending. To this purpose, we proceeded by extracting manually those failed - we list them in the Appendix that the system found without (sufficient) contrast, contrary to the decision of the critics. 6

After a careful perusal of the couplet of each such sonnet we came up with a double list. The result is that for 26 sonnets the couplet is a clear indicator of the subversion of mood, which may go from negative to positive, if the rest of the sonnet was mostly negative; or from positive to negative in the opposite case. As said above, the trigger for the reverted mood was to be found in the presence of a discourse marker at the beginning of the first (sometimes the second) line of the couplet. Appropriate discourse markers for mood reversal are adversatives, like "but", but also concessives, like "yet" and resultatives like "so, then". This only applies to 13 of the sonnets, the remaining couplets are characterized by presence of negation and negative items (while the rest of the poem has a majority of positive items). This rule was added to the system which raised accuracy on all sonnets to 66.88%. Here below the list of 26 reclassified sonnets:

3, 7, 8, 9, 13, 22, 40, 43, 49, 53, 58, 59, 60, 70, 73, 80, 120, 130, 131, 132, 133, 138, 140, 141, 148, 150

The remaining sonnets require the system to look at the previous and last stanza where again an appropriate discourse marker - or a negation plus negative items - must be present to introduce the reversal of mood. However, this additional modification of the system was not fully successful and was abandoned. The list of these 19 sonnets is this:

15, 16, 25, 26, 29, 31, 36, 55, 62, 85, 86, 88, 89, 91, 93, 94, 121, 124, 143 5

Conclusion

In this paper we have presented work carried out to annotate and experiment with the theme of irony in Shakespeare’s sonnets. The gold standard for the experiment has been created by collecting comments produced by literary critics on the presence of some kind of thematic, semantic and syntactic 6What we found is a list of 45 sonnets: 3, 7, 8, 9, 13, 15, 16, 22, 25, 26, 29, 31, 36, 40, 43, 49, 53, 55, 58, 59, 60, 62, 70, 73, 80, 85, 86, 88, 89, 91, 93, 94, 120, 121, 124, 130, 131, 132, 133, 138, 140, 141, 143, 148, 150 opposition in the sonnets as to produce some sort or irony. We have used the system available on the web, SPARSAR, to produce an automatic evaluation based on two parameters, phonetic features collapsed according to the theory that treats certain sounds to induce a negative rather than a positive mood. The second parameter is polarity, derived from the output of the module for sentiment analysis available in the system. From a comparison between the critics’ choices and the system’s the result was at first rather disappointing, it stopped at 50% of all sonnets. We then produced a new and much richer experiment by considering the logical structure of the sonnet and the content of the couplet by means of sentiment analysis, discourse markers and discourse relations. This allowed us to reach a final accuracy of 68.88%.

APPENDIX

List of couplets and quatrains from sonnets which contain a discourse marker for reverted logical structure A

Section 1: Couplets Reverting the

Logical Sequence

Sonnet 3 But if thou live remembered not to be, Die single and thine image dies with thee.

Sonnet 7 So thou, thyself out-going in thy noon, Unlooked on diest unless thou get a son.

Sonnet 8 Whose speechless song, being many, seeming one, Sings this to thee: “Thou single wilt prove none.”

Sonnet 9 No love toward others in that bosom sits That on himself such murd’rous shame commits.

Sonnet 22 Presume not on thy heart when mine is slain; Thou gav’st me thine not to give back again.

Sonnet 40 Lascivious grace, in whom all ill well shows, Kill me with spites; yet we must not be foes.

Sonnet 43 All days are nights to see till I see thee, And nights bright days when dreams do show thee me.

Sonnet 49 To leave poor me, thou hast the strength of laws, Since why to love I can allege no cause.

Sonnet 53 In all external grace you have some part, But you like none, none you, for constant heart.

Sonnet 58 I am to wait, though waiting so be hell, Not blame your pleasure, be it ill or well.

Sonnet 59 O sure I am the wits of former days To subjects worse have giv’n admiring praise.

Sonnet 60 And yet to times in hope my verse shall stand, Praising thy worth, despite his cruel hand.

Sonnet 70 If some suspéct of ill masked not thy show, Then thou alone kingdoms of hearts shouldst owe.

Sonnet 73 This thou perceiv’st, which makes thy love more strong, To love that well which thou must leave ere long.

Sonnet 80 Then, if he thrive and I be cast away, The worst was this: my love was my decay.

Sonnet 120 But that your trespass now becomes a fee; Mine ransoms yours, and yours must ransom me.

Sonnet 130 And yet, by heaven, I think my love as rare As any she belied with false compare.

Sonnet 131 In nothing art thou black save in thy deeds, And thence this slander, as I think, proceeds.

Sonnet 132 Then will I swear beauty herself is black, And all they foul that thy complexion lack.

Sonnet 133 And yet thou wilt, for I being pent in thee, Perforce am thine, and all that is in me.

Sonnet 138 Therefore I lie with her, and she with me, And in our faults by lies we flattered be.

Sonnet 140 That I may not be so, nor thou belied, Bear thine eyes straight, though thy proud heart go wide.

Sonnet 141 Only my plague thus far I count my gain, That she that makes me sin awards me pain.

Sonnet 148 O cunning love! With tears thou keep’st me blind, Lest eyes well seeing thy foul faults should find.

Sonnet 150 If thy unworthiness raised love in me, More worthy I to be beloved of thee. B

Section 2: Couplet + (Part of) Previous

Stanza

Sonnet 15 Then the conceit of this inconstant stay Sets you, most rich in youth, before my sight, Where wasteful time debateth with decay, To change your day of youth to sullied night; And all in war with time for love of you, As he takes from you, I engraft you new.

Sonnet 16 So should the lines of life that life repair Which this time’s pencil or my pupil pen Neither in inward worth nor outward fair Can make you live yourself in eyes of men. To give away yourself keeps yourself still, And you must live, drawn by your own sweet skill.

Sonnet 25 The painful warrior famousèd for worth, After a thousand victories once foiled, Is from the book of honor razèd quite, And all the rest forgot for which he toiled. Then happy I that love and am belovèd Where I may not remove nor be removèd.

Sonnet 26 But that I hope some good conceit of thine In thy soul’s thought, all naked, will bestow it. Till whatsoever star that guides my moving Points on me graciously with fair aspéct And puts apparel on my tattered loving, To show me worthy of thy sweet respect. Then may I dare to boast how I do love thee; Till then, not show my head where thou mayst prove me.

Sonnet 29 Yet in these thoughts myself almost despising, Haply I think on thee, and then my state, Like to the lark at break of day arising From sullen earth, sings hymns at heaven’s gate. For thy sweet love remembered such wealth brings That then I scorn to change my state with kings.

Sonnet 31 But things removed that hidden in thee lie. Thou art the grave where buried love doth live, Hung with the trophies of my lovers gone, Who all their parts of me to thee did give; That due of many now is thine alone. Their images I loved I view in thee, And thou, all they, hast all the all of me.

Sonnet 36 I may not evermore acknowledge thee, Lest my bewailèd guilt should do thee shame; Nor thou with public kindness honor me, Unless thou take that honor from thy name. But do not so; I love thee in such sort, As, thou being mine, mine is thy good report.

Sonnet 55 Even in the eyes of all posterity That wear this world out to the ending doom. So till the judgment that yourself arise, You live in this, and dwell in lovers’ eyes.

Sonnet 62 But when my glass shows me myself indeed, Beated and chopped with tanned antiquity, Mine own self-love quite contrary I read; Self so self-loving were iniquity. ’Tis thee, myself, that for myself I praise, Painting my age with beauty of thy days.

Sonnet 85 But that is in my thought, whose love to you, Though words come hindmost, holds his rank before. Then others for the breath of words respect, Me for my dumb thoughts, speaking in effect.

Sonnet 86 As victors of my silence cannot boast. I was not sick of any fear from thence; But when your countenance filled up his line, Then lacked I matter, that enfeebled mine.

Sonnet 88 The injuries that to myself I do, Doing thee vantage, double vantage me. Such is my love, to thee I so belong, That for thy right myself will bear all wrong.

Sonnet 89 Thy sweet belovèd name no more shall dwell, Lest I, too much profane, should do it wrong And haply of our old acquaintance tell. For thee against myself I’ll vow debate, For I must ne’er love him whom thou dost hate.

Sonnet 91 But these particulars are not my measure; All these I better in one general best. Thy love is better than high birth to me, Richer than wealth, prouder than garments’ cost, Of more delight than hawks or horses be; And having thee, of all men’s pride I boast; Wretched in this alone, that thou mayst take All this away, and me most wretched make.

Sonnet 93 But heav’n in thy creation did decree That in thy face sweet love should ever dwell; Whate’er thy thoughts or thy heart’s workings be, Thy looks should nothing thence but sweetness tell. How like Eve’s apple doth thy beauty grow, If thy sweet virtue answer not thy show.

Sonnet 94 But if that flow’r with base infection meet, The basest weed outbraves his dignity. For sweetest things turn sourest by their deeds; Lilies that fester smell far worse than weeds.

Sonnet 121 Which in their wills count bad what I think good? No, I am that I am, and they that level At my abuses reckon up their own; I may be straight, though they themselves be bevel. By their rank thoughts my deeds must not be shown, Unless this general evil they maintain: All men are bad, and in their badness reign.

Sonnet 124 That it nor grows with heat nor drowns with showers. To this I witness call the fools of time, Which die for goodness, who have lived for crime.

Sonnet 143 So run’st thou after that which flies from thee, Whilst I, thy babe, chase thee afar behind. But if thou catch thy hope, turn back to me, And play the mother’s part, kiss me, be kind. So will I pray that thou mayst have thy Will, If thou turn back and my loud crying still.

References

Salvatore Attardo . 1994 . Linguistic Theories of Humor . Mouton de Gruyter, Berlin New York.

Salvatore Attardo . 2000 . Irony as relevant inappropriateness . Journal of Pragmatics , 84 ( 32 ).

Ciprian Bacalu and Rodolfo Delmonte . 1999 . Prosodic modeling for syllable structures from the vesd - venice english syllable database . In Aspetti Computazionale in Fonetica, Linguistica e Didattica delle Lingue: Modelli e Algoritmi - Atti 9 Convegno GFS-AIA , pages 147 - 160 , Venezia. GFS-AIA.

Joan Bresnan , editor. 1982 . The Mental Representation of Grammatical Relations . The MIT Press, Cambridge MA.

Joan Bresnan . 2001 . Lexical-Functional Syntax . Blackwell Publishing , Oxford.

Dario Calimani . 2009 . William Shakespeare , I sonetti della menzogna . Carrocci , Roma.

P. Carvalho , L. Sarmento , M. Silva , and E. de Oliveira. 2009 . Clues for detecting irony in user-generated contents: oh ...! ! it's so easy;-) . In Proceeding of the 1st international CIKM workshop on Topicsentiment analysis for mass opinion , pages 53 - 56 . ACM.

Rodolfo Delmonte , Sara Tonelli, Marco Aldo Piccolino Boniforti, Antonella Bristot, and Emanuele Pianta . 2005 . Venses - a linguistically-based system for semantic evaluation . In Machine Learning Challenges , pages 344 - 371 , Berlin. Springer.

Rodolfo Delmonte . 2016 . Exploring shakespeare's sonnets with sparsar . volume 4 , pages 61 - 95 .

R.L. Eagle . 1916 . New light on the enigmas of Shakespeare's Sonnets. John Long Limited , London.

Ivan Fonagy . 1971 . The functions of vocal style . In Seymour Chatman, editor, Literary Style: A Symposium , pages 159 - 174 . Oxford UP , London.

Baayen R. H. , R. Piepenbrock , and L. Gulikers . 1995 . The CELEX Lexical Database (CD-ROM) . Linguistic Data Consortium.

Justine Kao and Dan Jurafsky . 2012 . A computational analysis of style, affect, and imagery in contemporary poetry . In Proceedings of NAACL Workshop on Computational Linguistics for Literature , pages 8 - 17 , Stroudsburg, PA, USA. ACL.

S.-M. Kim and E. Hovy . 2004 . Determining the sentiment of opinions . In Proceedings of the 20th international conference on computational linguistics - COLING , pages 1367 - 1373 , Stroudsburg, PA, USA. ACL.

Maria Antonietta Marelli . 2015 . William Shakespeare , I Sonetti - con testo a fronte . Garzanti , Milano.

Giorgio Melchiori . 1971 . Adriatica Editrice, Bari.

Antonio Reyes and Paolo Rosso . 2013 . On the difficulty of automatically detecting irony: beyond a simple case of negation . Knowledge and Information Systems , 40 : 595 - 614 .

Antonio Reyes , Paolo Rosso, and Davide Buscaldi . 2012 . From humor recognition to irony detection: The figurative language of social media . Data & Knowledge Engineering.

Michael Schoenfeldt . 2010 . Cambridge introduction to Shakespeare's poetry . Cambridge University Press, Cambridge.

Alessandro Serpieri . 2002 . Polifonia Shakespeariana . Bulzoni, Roma.

M. Taboada , J. Brooke , M. Tofiloski , K. Voll , and M. Stede . 2011 . Lexicon-based methods for sentiment analysis . Computational Linguistics , 37 ( 2 ): 267 - 307 .

Reuven Tsur . 1992 . What Makes Sound Patterns Expressive: The Poetic Mode of Speech-Perception . Duke UP , Durham N.C.

Cynthia Van Hee , Els Lefever , and Véronique Hoste . 2018 . Semeval-2018 task 3: Irony detection in english tweets .). In Proceedings of The 12th International Workshop on Semantic Evaluation , pages 39 - 50 . ACL.

1983. Shakespearean irony: Neuphilologische Mitteilungen , David K. Weiser . The 'sonnets' . 84 ( 4 ): 456 - 469 .

David K. Weiser . 1987 . Mind in Character - Shakespeare's Speaker in the Sonnets . The University of Missouri Press.