Abstract

The advent of the digital age has had an enormous impact on the way we research, teach and think about language ( :1). In digital humanities, the application of computer-assisted methods can facilitate investigation of corpora, (i.e. digitised samples of language in use), lead to discoveries barely detectable with the naked eye and help put interpretation by intuition to test. Such methods can also assist in investigation of the language of single texts and make our close reading more effective. This article aims to suggest how we may investigate the narrative voices in a work of fiction through the program Wmatrix ( ). The case study under analysis is Conrad’s Bildungsroman The Shadow-Line, A Confession (1917), a story revolving around a young inexperienced sea-captain who, during his first command of a ship, has to overcome a series of difficulties to accomplish his mission. In this work, the neat division between the I protagonist-narrator’s internal world and the external adult world he has to confront lends itself to investigation of the first person and other personal pronouns in the whole work. Given that stylistic analysis of literary texts is fundamentally a comparative process, I am here interested in comparing the target text, The Shadow-Line to two comparison texts by the same author. The keyness statistics for the pronouns and their detailed analysis through concordances will contribute to placing the I-voice at the centre of the narration and to identifying the foregrounded patterns of pronoun use that convey the I-voice and the other narrative voices throughout the story.

L’avvento dell’era digitale ha avuto un enorme impatto sul modo con cui noi facciamo ricerca, insegniamo e pensiamo il nostro linguaggio ( : 1). L’applicazione di metodi computazionali agli studi umanistici rende fattibile l’analisi di corpora (campioni digitalizzati della lingua in uso), porta a scoperte impossibili con altri strumenti e offre elementi di verifica oggettivi all’interpretazione tradizionale dei testi. Essa aiuta poi efficacemente nell’analisi del linguaggio di singoli testi e rende la lettura ravvicinata più efficace. Questo articolo ha lo scopo di suggerire come si possono analizzare con il software Wmatrix ( ) le voci narranti in un’opera narrativa. Oggetto di studio di questo articolo è il Bildungsroman The Shadow-Line, A Confession (1917) di Conrad, un racconto che ha come protagonista l’io narrante di un giovane e inesperto capitano di marina che durante il suo primo comando deve superare una serie di difficoltà per portare a termine la sua missione. In quest’opera la netta divisione tra il mondo interiore dell’io-protagonista e il mondo esterno con cui si deve confrontare si presta all’analisi dei pronomi di prima persona e degli altri pronomi personali nell’intera opera. Considerato che l’analisi stilistica di testi letterari implica un processo fondamentalmente di comparazione, si confronta il testo target The Shadow-Line con due romanzi simili dello stesso autore per analizzare le categorie grammaticali predominanti dei pronomi. I risultati statistici dei pronomi e la loro analisi tramite concordanze contribuiranno a posizionare l’io narrativo al centro della narrazione e a identificare l’interazione tra l’io narrante e le altre voci narrative.

1. Introduction

By definition, the term ‘corpus-assisted methods’ involves the analysis of large corpora. Biber ( : 15) defines a corpus as a large collection of principled texts stored on computer […] designed to represent a textual domain in a language. However, Biber did not exclude the applicability of corpus linguistics also to literary texts, thus infringing the basic assumptions of the corpus approach concerning size and representativeness ( ). Indeed, within the field of literary stylistics we come across plentiful corpus-based studies that are focussed on an individual target text of a given author, compared against a reference corpus of the same author. For example, Stubbs ( : 5-24) carried out a quantitative analysis of the frequency distribution of single words and the lexico-grammatical patterns in Conrad’s Heart of Darkness with one of the first versions of WordSmith Tools ( ) , and pointed out that collocations help disclose connotations and sometimes the idiosyncratic meanings of words. McIntyre and Archer ( ) employed the UCREL Semantic Annotation System (cf. Section 3) to investigate the distribution of semantic domains in Alan Bennet’s play The Lady in the Van in order to show the potentialities that a quantitative approach brings to the study of a character’s mind style, a phenomenon which has traditionally been studied qualitatively. A further example of corpus-based methods implemented on a single target text is found in my own study A Corpus Linguistic Approach to Literary Language and Characterization: Virginia Woolf’s The Waves ( ). I here investigated quantitatively and qualitatively the claim, subject to much debate in part of the literary arena, whether the language of each of the six characters can be distinguished from each other. This broad issue is tested by means of a series of statistical comparisons of each character’s part-of-speech and semantic domains through the program Wmatrix ( ).

In digital humanities ( ), corpus-aided methods are largely employed in the study of language from different perspectives, amongst which are intra-textual variations in the style of a single author or text, including chronological change, the dialogue or narration of multiple characters or narrators in a single novel, and other perceived shifts in style ( : 17ii); such methods are also increasingly being applied in pedagogical contexts (see, for example, : 253-270) to look at the stylistic properties of literary texts arising from patterns of use of linguistic features.

The stylistic properties investigated in this study are those involving pronouns, a grammatical category that falls into the larger category of function words, which has traditionally been the object of quantitative studies carried out in the field of authorship attribution ( : 271-280; : 59-66), investigations of literary idiolects, and stylistic changes in a given author’s artistic production ( ; ). More recently, the use of pronouns and their role in narratives have been investigated in corpus stylistics ( ; ; ) to study their stylistic features and effects on the reader since pronouns

delineate a narrator’s subject position in relation to objects/others. In effect the pronoun simultaneously determines, designates, identifies, refers to and (re-) affirms a particular narratorial role […]. A narratorial voice often works as a focalizing ‘window’- a particular, positioned perspective on fictional worlds. ( : 2)

In the present analysis, I will investigate the subject position of the I-protagonist and narrator in Conrad’s The Shadow Line, in relation to the other pronouns within the same text. With the aid of Wmatrix tools for POS analysis ( ), the distributions of personal pronouns in the target text can be readily determined, as well as any statistically significant (‘key’) differences in pronoun frequency between the target text and similar texts by the same author. Meanwhile, the concordance lines generated for the pronouns with the highest frequency will enable observation of the I-narrator’s subject position and its dynamics of interaction versus the other participants deployed at key moments in the narrative.

2. The Shadow-Line: the target text compared with Heart of Darkness and The Secret Sharer

The Shadow-Line, A Confession ( , hereafter TSL) is the story of an unnamed young man who, after deciding to leave his position as a mariner on a steamship and return to England for good, is unexpectedly appointed captain of a ship. The protagonist, now older, recalls the sea voyage from Bangkok to Singapore, and the obstacles he had to face. This Bildungsroman or novel of formation ( : 293-311) takes on the structure of a Proppian initiation ritual in which the challenge facing the young captain involves his ability to sail the ship and manage the crew, while encountering severe natural obstacles accompanied by an obscure and menacing adversary (the dead former captain). However, Ransome (the helper) is always on his side from the beginning to the end of his ordeal. In the end the young captain will manage to bring his ship safely back to harbour.

The grammatical usage of pronouns in TSL, as the target text, will be compared quantitatively to that in Heart of Darkness ( , HD) and The Secret Sharer ( , TSS). These two reference texts have been selected since all three works belong to the genre of the novel of formation, and all three are first-person narratives (Marlow in HD, and an unnamed captain in his first command in both TSS and TSL), each recalling in retrospect his first sea voyage as a young captain, the obstacles he had to face and the struggles within himself and with the adult world that challenge his initial naïve, idealised worldview and lead to his maturing process.

3. Methods: text digitisation and POS analysis inWmatrix

The analysis of the personal pronouns in Conrad’s The Shadow Line was carried out using Wmatrix ( ), a web-based environment containing corpus annotation tools for grammatical and semantic analysis developed by UCREL at Lancaster University. Similarly to other text analysis software, such as WordSmith Tools ( ), Antconc ( ), Wmatrix can generate frequency lists, key word lists, n-grams/clusters, collocates and word clouds that display the more frequently occurring words in a larger font for any text uploaded into the system (Antconc does not have this facility). The added value of Wmatrix is that it can perform automatic semantic analysis, which is beyond the scope of the present analysis ( ; ).

To process an automatic analysis in Wmatrix, users need to upload their texts into a web browser, then the Wmatrix Tag Wizard automatically carries out POS (part-of-speech) and USAS (UCREL Semantic Analysis System) tagging. The frequency and relative frequency, grammatical or semantic tags are generated and can be sorted out either alphabetically or in order of frequency; examples of each of these can also be viewed in context in concordance lines. Part-of-speech tagging is processed by CLAWS (Constituent Likelihood Automatic Word-tagging System ( ; ), a tool designed to assign grammatical tags to the words in a text. The first step that the CLAWS tagger carries out is the tokenization of a text and its morphological and grammatical analysis, then it allocates words to their parts-of-speech on a non-contextual basis, assigns each word to its grammatical category and in a subsequent phase, establishes its most probable word-tag taking into account the linguistic context in which the word occurs ( : 63-65). The accuracy rating of CLAWS is between 96-7%; if users want to increase the accuracy level they can manually intervene on the processed data. Once a target text is processed, we can generate key words and key semantic domain profiles by comparing a target corpus against other reference corpora stored in Wmatrix (e.g. samples from the BNC) or uploaded by users in the Wmatrix environment.

For a keyness analysis, two word-frequency lists, one from the target corpus and the other from the reference corpus are compared. Keyness is not measured in terms of relative frequency alone, but using a statistical measure called Log-likelihood (LL hereafter). This statistical test establishes how significant the relative frequencies are between two corpora, or texts, taking into account their respective sizes. The normal level for statistical significance is the 95% confidence level (p < 0.05) Chi-square value, which equates to an LL value of 3.8 or above. The recommended LL values for a word/tag to be statistically significant is above 6.63, as this is the cut-off point representing 99% confidence in its significance ( : 519-549); the higher the LL value the more significant is the difference between the two corpora. Once a keyness analysis is performed we can see the list of the overused word/tags+ or underused words/tags in the target corpus relative to the reference corpus sorted on their log-likelihood values .Here, overuse means ‘used with significantly higher frequency in the target than the comparison corpus’, and underuse means ‘used with significantly lower frequency in the target than the comparison corpus’.

Most studies that perform keyness analysis focus on the lexical level only, whereas the present study is one of the few to extend the concept to the level of grammatical word class (see ; ). From a purely quantitative perspective, keyness analysis helps identify keywords ( ), key POS and key semantic tags in a corpus, while from a qualitative perspective overused or underused words/tags indicate the distinctive features of a given text that are either over-represented or under-represented as compared with a reference corpus ( ). For sake of clarity, I here reproduce in , an extract from a table showing the first four hits obtained after carrying out a keyness analysis of parts of speech (POS) between TSL (target text) and one of the reference texts (TSS), where O1 is the observed frequency in the target text and O2 in the reference text sorted on the item’s LL value. If we look at the PPIO1tag (1st person sing. objective personal pronoun [me]) we note that in O1 this category has a higher number of occurrence (372, equal to a percentage value of 0.99%) and an LL value (19.17) indicating overuse. On the left of the table, we can click on ‘List1’ or ‘List2’ to see the pronoun words comprised in PPIO1, or click on the concordance links if we want see keyword context results for the PPIO1 tag. The word cloud displayed below the table shows the most frequently occurring tags in a larger font, amongst which is the PPIO1 tag.

The keyness analysis of the POS frequencies of the personal pronouns in TSL is produced through comparison between the target text (TSL) and the reference texts (HD and TSS) respectively. The digitised versions of the texts utilised for quantitative analysis are from Project Gutenberg eBooks (https://www.gutenberg.org/). The e-texts come in plain text format (TXT), thus they were already perfectly suitable to be fed into and processed by Wmatrix. However, before loading each text into the system, angle brackets were inserted around the parts of the text I wanted the program to ignore (e.g. the cover page and the chapter numbers). The next step entailed selecting three different sample texts I would use for qualitative analysis. The computer-based analysis of the personal pronouns in TSL follows the stages indicated below:

1. Corpus design:

-choice of target text and reference texts (cf. Section 2).

2. Pronoun retrieval, through the CLAWS POS (part-of-speech) tagger, built into Wmatrix.

- keyness lists, listing the pronoun tags in order of keyness.

- concordances, showing the occurrence of the personal pronouns in their co-text.

- frequency lists, listing the pronouns in order of frequency.

3. Interpreting the data:

-compared to the three steps in Stage 2 which are quantitatively based, this stage of the process, although it makes use of the ‘concordance’ option in Wmatrix, is qualitative.

4. Results: key POS tags, and key pronouns

shows the statistically significant personal pronoun tags sorted in descending order of LL values produced by Wmatrix, when the target corpus The Shadow Line (TSL; corpus size 37,710) is compared with the first reference text Heart of Darkness (HD, corpus size 37.661) and the second reference text The Secret Sharer (TSS, corpus size 14,997).

Key:

APPGE - possessive pronoun, pre-nominal (e.g. my, your, our)

PPIO1 - 1st person sing. objective personal pronoun (me)

PPHS1 - 3rd person sing. subjective personal pronoun (he, she)

PPIS1 - 1st person sing. subjective personal pronoun (I)

More robust results are recorded when the target text is compared with HD than with TSS, producing both a higher number of significantly overused POS tags (cf. Appendix 1), and higher LL values. This is predictable given the evident similarities between TSL and TSS. In this regard, Benson’s ( : 46-56) study of Conrad’s Two Stories of Initiation defines the two novellas as twin stories of initiation into maturity ( : 46) and states that “there is in The Shadow Line a figure analogous to Leggatt of The Secret Sharer. The double in the later work is Ransome […]” ( : 53).

It may be argued that this finding might have been expected, in view of the stronger thematic and narrative similarities that TSS shares with TSL than HD. But what the analysis in Wmatrix provides is important quantitative validation of such a prediction, and (as we shall see with concordances) detailed, systematic insight into how those interconnections play out in the text. In spite of the features that the two texts share against which TSL is compared, the results in show that statistically significant differences occur in the use not only of first-person pronouns (PPIO1: 1st person sing. objective personal pronoun (me); PPIS1: 1st person sing. subjective personal pronoun [I]), but also third-person pronouns (PPHS1: 3rd person sing. subjective personal pronoun [he, she]). Similarly, another key pronominal POS-tag here, APPGE, (possessive pronoun, pre-nominal) represents pronouns with first person and third person, as well as second person, referents (e.g. my, your, our, his/her). If we unpack these POS tags, we can observe that your, our and she hold a much lower frequency of occurrence compared to the over-represented I-he words within the same POS tags. Considered that the story revolves around the masculine activity of a sea voyage, the pronouns such as your, our appear, in their co-text, to be mainly employed by the I narrator-protagonist to refer to a male or other males who, most of the time, is/are part of the crew. Instead, the she pronouns refer in most cases, as is the convention in English, to the ship the young captain left, and the ship he is appointed to. The over-represented narrative I-he pattern is further strengthened in the keyness analysis carried out at word level between TSL and the reference texts. The results are displayed in in order of their statistical significance.

As already noted in the POS tag keyness analysis, higher statistical significance is recorded in the comparison between The Shadow Line and Heart of Darkness. The over-represented pronouns in the target text compared to the reference text (HD) show a significant use of I and he pronouns occurring in The Shadow Line and hint at a narrative centred around an internal self-referential I-voice with a strong focus on the external referential he. From the concordances produced by Wmatrix, the over-represented I-pronouns refer most of the time to the I protagonist, while the he pronouns to the adults in his world, especially to his ship-mates Mr. Burns and Ransome. This suggests that the young captain, during his difficult and challenging voyage, relies on a dyadic I-he type of interaction and tends to exclude I-they and the I-we relationships.

The dynamics of the I-he interaction can be better understood by observing their distribution across the narrative. For this purpose, in Section 5 I look at the concordance lines generated from three sample passages (cf. Appendix 2) taken from the beginning, middle and end of the story, respectively, each representing a key phase of the protagonist’s maturing process.

5. Distribution of the over-represented I-he pattern in three samples from The Shadow-Line

Compared to the quantitative analysis of the pronouns carried out in Section 4, this Section is qualitatively-based as it makes use of the concordance tool in Wmatrix to search for the occurring I-he patterns in three key narrative moments of The Shadow Line. The concordance tool allows us to search for a word/node with a given amount of co-text to its left and to its right ( ; ). In order to look at the concordances for the pronouns in the sample passages, I uploaded each text-sample in Wmatrix, produced their concordance lines, and manually sorted out the lines in which the I/he pronouns occur as subject in order of occurrence in the samples. The only exception is found in Sample 1 where the occurrence of the first-person possessive (‘my’) is shown. Other pronouns occurring in the vicinity of the node (i.e. the search pronouns ‘I’ and ‘he’) in the concordance lines are also considered.

Text Sample 1 is taken from the opening of the story when the unnamed protagonist is telling the narratee (understood in this case, by default, as the reader) about the reasons for quitting his job as a mariner on a steamship. The passage exemplifies an element typically found at the beginning of a novel of formation in which, as noted by Stape and Simmons (in : xlv), the youthful hero […] experiences a sudden discontent, shock, or impulse for change that entails a break with the known routines of family, career or traditional home. In the concordances produced, only I-pronouns appear since the I-narrator, in this segment of text, is the sole occupant of the narrative, which is characterised by an exclusive focalisation on the protagonist’s inner self.

6. Conclusions

In this article I have carried out a computer-aided analysis of the personal pronouns in Conrad’s The Shadow Line compared to two other works by the same author. The WMatrix software was employed to identify and display the pronouns in order of statistical keyness in the target text compared to the reference texts. I then looked at the most frequently occurring pronouns in three different key phases of the story through concordances generated by Wmatrix. This analytical step exemplified how, in this work of fiction, the foregrounded I-protagonist is positioned in the external world during his rite of passage.

Further investigation could be carried out in order to ascertain to what extent the over-represented self-referential I and the other-referential he is a stylistic feature of the Bildungsroman, or Conrad’s fingerprint. Also, the study of the key pronouns in the The Shadow Line through the frequently repeated sequences of words, bundles/clusters ( :989; ) would further contribute to showing the narrator’s interaction with the adult world and his psychological development in this novella of formation.

Though to practitioners of the advanced and sophisticated current corpus-aided methods working with keyness analysis, frequencies and concordances might be regarded as a traditional and superseded method, they are still a highly relevant support in stylistic analyses of literary texts. In this article, I have emphasised that the key POS-tag approach is an important extension of the widely-used keywords technique, but extended to a more abstract level of analysis (grammatical word class, rather than simply lexis). Given the evidence of my case study, key POS-tag analysis, when combined with qualitative analysis (e.g. concordances) is a method that holds promise for literary stylistics more broadly.

References

1. Adolphs, S. 2006. Introducing Electronic Text Analysis: A Practical Guide for Language and Literary Studies. Abingdon: Routledge.

2. Anthony, L. 2014. AntConc [Computer Software] Tokyo: Waseda University.

3. Archer, D., Wilson, A. & Rayson, P. 2002. Introduction to the USAS category system, 1–37.

4. Balossi, G. 2015. A computer-aided approach to I and the World in Conrad’s The Shadow Line. On-line Proceedings of the Annual Conference of the Poetics and Linguistics Association (PALA) pp.1-18.

5. Balossi, G. 2014. A Corpus Linguistic Approach to Literary Language and Characterization: Virginia Woolf’s The Waves. Amsterdam: John Benjamins.

6. Benson, C. 1954. Conrad’s Two Stories of Initiation. PMLA, 69 (1), 46-56. DOI:10.2307/460126

7. Biber, D. 2011. Corpus linguistics and the study of literature: Back to the future? Scientific Study of Literature 1(1) pp.15-23. DOI: 10.1075/ssol.1.1.02bib

8. Biber, D., Conrad, S., Finegan, E., Leech, G. & Johansson, S. 1999. Longman Grammar of Spoken and Written English. Harlow: Longman.

9. Biber, D., Conrad, S. & Reppen, R. (1998). Corpus Linguistics: Investigating Language Structure and Use. Cambridge: CUP.

10. Burrows, J. F. 1987. Computation into Criticism: A Study of Jane Austen’s Novels and an Experiment in Method. Oxford: Clarendon Press.

11. Conrad, J. [1917] 2006. The Shadow-Line. Electronic Version, www.gutenberg.org; [EBook #451].

12. Conrad, J. [1910] 2009. The Secret Sharer. Electronic Version, www.gutenberg.org; [EBook #220].

13. Conrad, J. [1902] 2009. Heart of Darkness. Electronic Version, www.gutenberg.org; [EBook #219].

14. Conrad, J. [1917] 2013. The Shadow-Line, A Confession. The Cambridge Edition to the Works of Joseph Conrad. Stape, J. H., Simmons, A. S. (eds). Cambridge: CUP.

15. Damerau, F.J. 1975. The Use of Function Word Frequencies as Indicators of Style. Computers and the Humanities 9: 271-280.

16. Dunning, T. 1993. Accurate methods for the statistics of surprise and coincidence. Computational Linguistics 19(1): 61-74.

17. Garside, R. 1987. The CLAWS word-tagging system. In The Computational Analysis of English: A Corpus-based Approach. Roger Garside, Geoffrey Leech & Geoffrey Sampson (eds), 30-41. London: Longman.

18. Garside, R. & Smith, N. 1997. A hybrid grammatical tagger: CLAWS4. In Garside, R., Leech, G., McEnery, A. (eds.). Corpus Annotation: Linguistic Information from Computer Text Corpora, 102-121. London: Longman.

19. Gibbons, A. & Macrae, A. (eds.) 2018. Pronouns in Literature: Positions and Perspectives in Language. Basingstoke: Palgrave MacMillan.

20. Hirsch, M. 1979. The Novel of Formation as Genre: Between Great Expectations and Lost Illusions. Genre 12 (3), pp. 293-311.

21. Hoover, D.L. 2017. The microanalysis of style variation. Digital Scholarship in the Humanities 32 (2): ii17–ii30.

22. Kestemont, M. 2014. Function Words in Authorship Attribution. From Black Magic to Theory? Proceedings of the Third Computational Linguistics for Literature Workshop, co-located with EACL 2014 - the 14th Conference of the European Chapter of the Association for Computational Linguistics (27 April 2014, Gothenburg, Sweden): 59-66.

23. Leech, G. 1992. Corpora and theories of linguistic performance. In Jan Svartvik (ed.), Directions in corpus linguistics, pp. 105-122. Berlin & New York: Mouton de Gruyter.

24. Leech, G. 2013. Virginia Woolf meets Wmatrix. Études de stylistique anglaise. URL: http://journals.openedition.org/ esa/1405 DOI: 10.4000/esa.1405.

25. Mahlberg, M. 2013. Corpus Stylistics and Dickens’s Fiction. London: Routledge.

26. Mahlberg, M. & P. Stockwell. 2016. Point and CLiC: teaching literature with corpus stylistic tools. In McIntyre, D. & Archer, D. 2010. A corpus-based approach to mind-style. Journal of Literary Semantics 39(2):167-182.

27. Murphy, S. 2015. I will proclaim myself what I am: Corpus stylistics and the language of Shakespeare’s soliloquies. Language and Literature 24 (4): 338-354.

28. Propp, V. 1968². The Morphology of the Folktale. Trans. by Laurence Scott. Austin: University of Texas Press.

29. Rayson, P. 2008. From key words to key semantic domains. International Journal of Corpus Linguistics. 13 (4): 519-549. DOI: 10.1075/ijcl.13.4.06ray

30. Rayson, P. 2003. Matrix: A statistical method and software tool for linguistic analysis through corpus comparison. Ph.D. thesis, Lancaster University.

31. Rayson, P. Berridge, D. & Francis, B. 2004. Extending the Cochran rule for the comparison of word frequencies between corpora. JADT: 7es Journées internationales d’Analyse statistique des Données Textuelles : 1-12.

32. Schreibman, S., Siemens, R. & Unsworth, J. (eds) 2016². A New Companion to Digital Humanities. Oxford: Wiley-Blackwell.

33. Scott, M. 2019. WordSmith Tools version 8. Stroud: Lexical Analysis Software.

34. Scott, M. 1997. WordSmith Tools version 2, Oxford: Oxford University Press.

35. Scott, M. 1997. PC analysis of key words - and key key words. System 25 (2): 233-245. DOI:10.1016/S0346-251X(97)00011-0

36. Sinclair, J. McH. 1991. Corpus, Concordances, Collocation. Oxford: OUP

37. Stubbs, M. 2005. Conrad in the computer: Examples of quantitative stylistic methods. Language and Literature 14(1), pp. 5-24. DOI: 10.1177/0963947005048873

38. Tabata, T. 1995. Narrative Style and the Frequencies of Very Common Words: A Corpus-Based Approach to Dickens’s First Person and Third Person Narratives. English Corpus Studies 2: 91-109.

Appendix 1: Pronoun tags analysed

Appendix 2