Dickens's Narrative Style:
A Statistical Approach to Chronological Variation*

Tomoji Tabata

 

Rˇsumˇ. Cette communication est consacrˇe aux analyses statistiques du langue romanesque de Dickens. Le probl¸me spˇcifique est de trouver des marqueurs linguistiques de la chronologie dans les narrations de l'auteur. Dans les analyses qui suivent, je me sers de l'analyse du composant principal et du Ētest t de StudentČ pour examiner la distribution des frˇquences de vocables les plus communs dans certains extraits de son oeuvre. Cette approche fait ressortir le contraste entre les premi¸res oeuvres et les plus rˇcentes. Les facteurs diffˇrentiateurs entre les premi¸res narrations et les suivantes sugg¸rent une mutation du complexe style formel en style oral.

Key-words: Language of Dickens, chronology, style, narrative, principal component analysis, Student's t-test, computational stylistics, literary statistics, very common words

Mots-clˇs: Langue de Dickens, chronologie, style, narration, analyse des principaux composants , test t de Student, stylistique computationelle, statistiques littˇraires, vocables communs.

1. Introduction
This essay is a sequel to the approach that I adopted in my previous paper (Tabata 1993), in which I applied statistical methods to the stylistic study of four volumes of literary texts by Dickens. Two of the four were published in his twenties: Sketches by Boz (1836) and Oliver Twist (1837-8), while Christmas Books were written in his thirties (1843-8). The last of the set was his last and unfinished novel, The Mystery of Edwin Drood (1870). The result of analysis suggested gradual changes of Dickens's style in some dimensions, such as elaboration, pronominal reference, nominalisation, and so on, though not in exact chronological terms. While the set I studied was not large enough to allow any chronological generalisation, the length of his career (from his early twenties to his death at the age of fifty-eight) encourages a longitudinal study of style. The purpose of this paper therefore is to present a test for chronological variation of Dickens's style by enlarging the size and the scope of text corpus, with due modification of the methods.

The corpus draws on four novels written in the 1830s as an early subset, and six novels written after 1849, when he started David Copperfield, as a later subset (see KEY TO TABLE AND FIGURES).(1) An interval of ten years was allowed between the two sets in an effort to make interpretation of the results as straightforward as possible. As a result, Christmas Books were forced out of the set. Each text is represented by approximately twenty-thousand words from the beginning of the novel, and the language of "pure-narrative" is extracted as a basis for comparison.(2) Bleak House, since it is recounted by two irregularly alternating narrators, provides two narratives: one is the first-person retrospective narrative by Esther Summerson, who uses the past tense, while the other is recounted by an anonymous third-person narrator, who constantly uses the present tense.  

     Label       Narrator [TEXT] & Date              Word-tokens  Segments



     Works Written in 1830s 
     SB #1-3     Sketches by Boz              (1836)      12569      3
     PP #1-2     The Pickwick Papers        (1836-7)      11081      2
     OT #1-4     Oliver Twist               (1837-8)      16677      4
     NN #1-3     Nicholas Nickleby          (1838-9)      12863      3



     Works Written after 1849
     David #1-5  David [David Copperfield] (1849-50)      20145      5
     Esther #1-4 Esther [Bleak House]       (1852-3)      18399      4
     BH #1-2     Bleak House                (1852-3)       7389      2
     TTC #1-3    A Tale of Two Cities         (1859)      12798      3
     Pip #1-4    Pip [Great Expectations]   (1860-1)      18359      4
     OMF #1-3    Our Mutual Friend          (1864-5)      13117      3
     ED #1-3     The Mystery of Edwin Drood   (1870)      11973      3



                 Groups          
                 1     Early Third-person Narrative       53190     12
                 2     Late First-person Narrative        56903     13
                 3     Late Third-person Narrative        45277     11



KEY TO TABLE AND FIGURES
David Copperfield, Great Expectations, and Esther's narrative are chosen with an eye to first-person narrative. The current corpus, seen in terms of person, consists of three first-person narratives and eight third-person narratives. To keep the balance, three shorter first-person narratives, "The stroller's tale", "The convict's return", which are interpolated in The Pickwick Papers, and Miss Wade's narrative, "The History of a Self-Tormentor", which is embedded in Little Dorrit (1855-7), will also be added to analysis in due course.(3) Of the eleven main narratives, three are narrated in the present tense (Sketches by Boz and Edwin Drood in addition to the anonymous narrator of Bleak House).

Since differences in point of view are obvious between first-person narratives and third-person narratives, between narratives told in the past tense and those in the present tense, it is desirable to exclude deictic words, such as personal pronouns and finite verbs, from statistical analysis so as to diminish the overshadowing effect of what is already evident. Otherwise vectors that differentiate first-person narratives from third-person narratives and texts couched in the past tense from those in the present tense will be so strong that other determinants of style may be submerged. This exclusion of personal pronouns and finite verbs deprives my data of some of the interesting subjects for computational stylistics.(4) In return, however, it makes them sensitive to evidence of subtler stylistic differences.

2. Analysis and Interpretation
2.1 Principal component analysis of the 74 very common word-types
The following statistical work is based on the seventy-four commonest non-deictic word-types (as listed in TABLE 1) whose occurrence in segments of text is frequent enough to allow multivariate analysis.(5) Initially I picked up one hundred commonest words, which included thirteen personal pronouns and as many finite verbs.(6) Even after these twenty-six words are excluded, the remaining seventy-four words still account for more than 40 % of all the word-tokens in the pure-narrative. 
 Rank Word-types   SB    PP    OT    NN  David Esther  BH   TTC   Pip   OMF    ED Total(raw)(%)



   1  the        7.606 9.097 7.327 6.320 4.433 4.723 6.834 7.462 5.817 6.602 6.515  9935  6.394
   2  and        3.914 4.088 3.598 4.019 3.927 4.310 3.424 4.329 4.210 3.690 3.792  6164  3.967
   3  of         4.225 3.592 3.190 3.281 2.636 2.462 3.424 3.469 2.511 3.255 3.566  4879  3.140
   4  a          2.912 2.346 2.908 3.001 2.442 2.571 2.774 2.508 2.495 3.171 2.773  4194  2.699
   5  in         2.164 1.661 1.847 1.788 1.812 1.853 2.463 2.016 1.672 2.173 2.096  2983  1.920
   6  to(inf)    1.201 1.101 1.325 1.314 1.524 1.549 1.340 1.188 1.416 1.189 1.111  2055  1.323
   7  with       1.154 1.263 1.091 1.104 1.052 1.277 0.920 1.274 1.149 1.395 1.336  1841  1.185
   8  to(prep)   1.034 1.119 1.091 1.026 1.176 1.163 0.988 1.203 1.024 1.235 1.169  1736  1.117
   9  it         0.676 0.605 0.768 0.700 1.365 1.076 1.272 1.399 1.285 1.113 1.052  1624  1.045
  10  as         0.692 0.957 0.851 1.011 1.082 0.848 0.826 0.938 1.008 1.037 1.128  1476  0.950
  11  at         0.756 0.496 0.738 0.910 1.023 0.962 0.758 0.836 1.100 0.953 0.618  1337  0.861
  12  that(conj) 0.549 0.388 0.660 0.599 0.933 0.989 0.595 0.484 1.002 0.602 0.501  1098  0.707
  13  on(prep)   0.835 0.713 0.522 0.575 0.660 0.554 0.555 0.766 0.757 0.724 0.727  1040  0.669
  14  by         0.525 0.578 0.672 0.536 0.457 0.435 0.528 0.524 0.452 0.640 0.585   826  0.532
  15  which(r)   0.812 0.641 0.762 0.669 0.417 0.424 0.420 0.399 0.381 0.435 0.309   794  0.511
  15  for(prep)  0.732 0.415 0.570 0.459 0.491 0.484 0.568 0.328 0.479 0.343 0.543   762  0.490
  17  but        0.422 0.289 0.414 0.342 0.660 0.582 0.555 0.445 0.523 0.450 0.317   729  0.469
  18  not        0.501 0.262 0.336 0.327 0.551 0.554 0.447 0.391 0.523 0.381 0.267   664  0.427
  19  be         0.461 0.162 0.444 0.435 0.382 0.419 0.663 0.305 0.392 0.335 0.326   603  0.388
  20  from       0.485 0.478 0.402 0.443 0.308 0.326 0.406 0.484 0.376 0.267 0.326   595  0.383
  21  when       0.159 0.343 0.342 0.350 0.536 0.462 0.392 0.344 0.485 0.252 0.309   585  0.377
  22  this       0.294 0.307 0.546 0.498 0.382 0.217 0.298 0.313 0.338 0.442 0.451   579  0.373
  23  all        0.326 0.171 0.300 0.420 0.367 0.451 0.352 0.336 0.479 0.328 0.334   561  0.361
  24  an         0.493 0.433 0.348 0.233 0.308 0.342 0.392 0.336 0.289 0.450 0.443   560  0.360
  25  or         0.398 0.153 0.372 0.365 0.357 0.255 0.379 0.305 0.283 0.381 0.342   505  0.325
  26  out        0.080 0.199 0.222 0.194 0.506 0.364 0.487 0.391 0.414 0.282 0.342   503  0.324
  27  there      0.294 0.217 0.288 0.319 0.387 0.375 0.365 0.328 0.327 0.198 0.234   480  0.309
  28  into       0.382 0.208 0.408 0.365 0.268 0.217 0.392 0.352 0.283 0.160 0.393   474  0.305
  29  who(r)     0.358 0.253 0.384 0.404 0.218 0.223 0.514 0.211 0.245 0.435 0.134   457  0.294
  29  one        0.446 0.235 0.282 0.350 0.323 0.239 0.217 0.367 0.234 0.236 0.309   457  0.294
  31  that(dem)  0.239 0.280 0.222 0.334 0.472 0.239 0.203 0.273 0.272 0.274 0.259   447  0.288
  32  very       0.223 0.244 0.462 0.365 0.338 0.413 0.271 0.211 0.185 0.175 0.150   445  0.286
  33  if         0.207 0.208 0.186 0.272 0.377 0.288 0.352 0.211 0.468 0.198 0.284   443  0.285
  34  little     0.199 0.190 0.222 0.404 0.338 0.413 0.298 0.266 0.169 0.229 0.384   442  0.284
  35  up         0.151 0.280 0.288 0.327 0.333 0.228 0.257 0.289 0.376 0.282 0.200   435  0.280
  36  so(adv.d.) 0.151 0.126 0.174 0.187 0.283 0.554 0.230 0.227 0.278 0.175 0.242   394  0.254
  37  upon       0.191 0.190 0.228 0.334 0.268 0.212 0.284 0.336 0.240 0.206 0.234   382  0.246
  38  no(a)      0.326 0.153 0.258 0.210 0.223 0.163 0.244 0.242 0.267 0.244 0.192   356  0.229
  39  down       0.088 0.117 0.198 0.109 0.228 0.207 0.203 0.313 0.272 0.252 0.209   318  0.205
  40  some       0.263 0.135 0.210 0.179 0.243 0.158 0.203 0.211 0.218 0.168 0.234   316  0.203
  41  more       0.239 0.180 0.210 0.187 0.194 0.207 0.135 0.180 0.196 0.107 0.192   292  0.188
  42  old        0.255 0.162 0.294 0.117 0.114 0.304 0.284 0.094 0.065 0.114 0.284   287  0.185
  43  man        0.294 0.343 0.126 0.155 0.055 0.147 0.176 0.133 0.283 0.259 0.125   285  0.183
  44  then       0.175 0.262 0.126 0.124 0.164 0.136 0.041 0.164 0.278 0.198 0.284   281  0.181
  45  before     0.223 0.108 0.168 0.163 0.179 0.207 0.149 0.211 0.174 0.229 0.117   277  0.178
  46  other      0.271 0.190 0.174 0.264 0.129 0.114 0.108 0.211 0.153 0.206 0.117   269  0.173
  47  over       0.167 0.153 0.192 0.093 0.129 0.125 0.108 0.250 0.174 0.267 0.200   262  0.169
  48  again      0.127 0.171 0.108 0.086 0.238 0.168 0.149 0.203 0.153 0.221 0.192   260  0.167
  49  its        0.247 0.171 0.060 0.155 0.079 0.092 0.325 0.328 0.093 0.145 0.359   258  0.166
  49  that(r)    0.215 0.072 0.090 0.124 0.194 0.130 0.217 0.344 0.153 0.160 0.167   258  0.166
  51  time       0.151 0.126 0.180 0.225 0.208 0.158 0.135 0.133 0.212 0.084 0.125   255  0.164
  51  being      0.088 0.090 0.222 0.225 0.189 0.130 0.149 0.109 0.158 0.168 0.251   255  0.164
  53  two        0.239 0.153 0.120 0.272 0.114 0.125 0.108 0.250 0.120 0.145 0.184   251  0.162
  54  than       0.095 0.099 0.168 0.109 0.174 0.158 0.257 0.219 0.191 0.114 0.184   248  0.160
  55  about      0.167 0.090 0.114 0.179 0.169 0.201 0.149 0.148 0.207 0.091 0.175   245  0.158
  55  head       0.064 0.126 0.204 0.117 0.169 0.163 0.068 0.219 0.125 0.206 0.226   245  0.158
  57  gentleman  0.064 0.334 0.438 0.272 0.070 0.163 0.068 0.078 0.000 0.122 0.033   232  0.149
  58  what       0.095 0.054 0.114 0.187 0.169 0.217 0.122 0.094 0.202 0.114 0.150   226  0.145
  59  after      0.048 0.144 0.186 0.140 0.204 0.141 0.081 0.109 0.169 0.099 0.150   220  0.142
  59  much       0.064 0.126 0.138 0.194 0.134 0.136 0.135 0.078 0.196 0.198 0.134   220  0.142
  61  any        0.199 0.081 0.120 0.086 0.223 0.130 0.217 0.156 0.131 0.084 0.117   219  0.141
  62  face       0.072 0.081 0.150 0.132 0.134 0.136 0.068 0.211 0.076 0.183 0.292   217  0.140
  63  great      0.103 0.180 0.216 0.233 0.089 0.174 0.149 0.125 0.093 0.099 0.058   213  0.137
  64  hand       0.040 0.153 0.120 0.086 0.124 0.082 0.054 0.352 0.142 0.160 0.150   207  0.133
  65  like(prep) 0.080 0.072 0.048 0.047 0.159 0.136 0.122 0.188 0.261 0.168 0.100   204  0.131
  66  eyes       0.048 0.144 0.156 0.140 0.104 0.109 0.027 0.227 0.136 0.137 0.175   202  0.130
  67  mother     0.056 0.000 0.024 0.016 0.874 0.005 0.027 0.000 0.011 0.008 0.050   201  0.129
  68  such       0.151 0.117 0.060 0.155 0.169 0.212 0.135 0.125 0.087 0.076 0.075   196  0.126
  69  on(adv)    0.103 0.117 0.096 0.187 0.099 0.114 0.068 0.086 0.163 0.114 0.167   188  0.121
  70  back       0.088 0.063 0.114 0.086 0.134 0.168 0.054 0.117 0.142 0.122 0.159   186  0.120
  71  way        0.056 0.099 0.150 0.070 0.144 0.158 0.054 0.156 0.131 0.084 0.117   183  0.118
  71  young      0.048 0.090 0.150 0.132 0.025 0.136 0.176 0.141 0.114 0.099 0.251   183  0.118
  73  never      0.143 0.000 0.066 0.054 0.179 0.228 0.217 0.023 0.136 0.061 0.125   181  0.117
  74  now        0.151 0.045 0.078 0.062 0.114 0.141 0.095 0.117 0.163 0.160 0.092   178  0.115
  
 (a) = adjective, (adv) = adverbials, (adv.d.) = adverb of degree, (conj) = conjunction,
 (dem) = demonstrative, (inf) = infinitive, (r) = relative, (prep) = preposition



TABLE 1 Eleven Narrators in Dickens's Novels: Standardised (text-percentage)
frequencies for the 74 very common word-types in the "pure-narrative."
Fig. 1 Eleven Narrators in Dickens's Novels: Word-plot
(for the 74 very common word-types in the corpus)

Fig. 2 Eleven Narrators in Dickens's Novels: Text-plot
(based on the 74 very common word-types in the corpus)

Fig. 1 shows the results of principal component analysis (PCA) of the seventy-four words in the texts divided into thirty-six 4000-word segments. Segmentation of text, one of modifications adopted in this approach, has two objectives. First, to give each variable (word) as appropriate a number of samples as possible in order to reduce the possibility of chance effect. Second, to help observe internal variation (or consistency) in each text. The first step of PCA is to measure correlation of each of the seventy-four words with each of the other seventy-three across the thirty-six segments of text.(7) In the process, a matrix of 2701 correlation coefficients is generated. The next step is to subject the resulting correlation matrix to eigen-analysis. By eigen analysis, the principal components of the matrix are extracted, and it is possible to project the most powerful components in a scatter diagram like Fig. 1 .

The graph gives a picture of the reciprocal relationships among the seventy-four words. Relative distance between the entries reflects similarity or contrast among these words in their concomitant variation over the thirty-six text segments. Words located towards the east and those located towards the west of the graph tend to be mutually opposed: when the frequency of one set of words rises high in a given text-sample, the frequency of the other tends to fall low. The same applies to the vertical axis. To give a crude example, where the relative frequency of "of" runs high "the" also tends to occur frequently, while the relative incidence of words like "when" and "so" (adverb of degree) tend to fall away. The final step is to multiply the eigen-matrix that produced Fig. 1 back through the original frequency table for the 36 segments of text. The purpose is to distribute the text segments on the word-pattern. This procedure produces Fig. 2.

Since this graph is a product of the previous one, they correspond to each other. When one compares the two graphs, one can see that easterly entries of words in Fig. 1 occur more frequently in text-entries lying towards the east of Fig. 2 than in those lying towards the west (and vice versa), while northern entries of words in Fig. 1 predominate in the text-entries situated in the north of Fig. 2 and are outnumbered in texts that find their place in the south (and vice versa). Additionally, words that contribute little to the horizontal and vertical differentiations of texts lie around the middle of the graph.

The most obvious feature of Fig. 2 is the separation between the first-person and the third-person narratives along the horizontal axis. This axis represent the strongest component of the matrix, which accounts for 16.74 % of the complex interrelationships among the 74 words. That the first and third-person distinction predominates along the horizontal axis, with no interpenetration between the two groups, testifies that stylistic difference between the two narrative modes is still remarkably consistent even after the removal of the deictics. Another feature of interest, which is more relevant to the present study, is that the overall contrast can be drawn diagonally between the entries for earlier novels and those for the later. With an imaginary border running from the north-east to the south-west, the segments of earlier texts find themselves in the lower right triangle, while those of the later sit in the upper left. Such a diagonal separation takes place when the two vectors function in conjunction with each other. A notable departure from this pattern is the first segment for the anonymous narrator of Bleak House (BH#1), which goes deep into the middle of the earlier narrators' cluster. The second segment lies just across the border.

Let us scrutinise Fig. 1 to see how reciprocities among the 74 words are related to the overall configuration of Fig. 2. The definite and the indefinite articles, the, a, and an all incline towards the east. The first segments of The Pickwick Papers (PP#1) and Sketches by Boz (SB#1) are among the most given to the use of these words. The easterly locations of the articles seem to reflect the comparative weighting on nominal phrases in the third-person and its comparative avoidance in the first-person narratives. The major prepositions, such as of, by, and in, also leaning towards the east end, add to this interpretation. The relatives which and who, markers of elaboration or identification, go to the far south-east, the territory where the earlier third-person narratives dominate. Among others, SB#1 has frequent recourse to these devices for elaboration.

On the other hand, one can notice that the westerly entries of words betoken intensification and comparison (the adverb of degree so, never, very, much, the prepositional like, and than). Such a group of words represent habits of emphasis shared by the first-person narrators, among whom the first segment of Esther's narrative is an extreme example. Habits of reference provide another revealing contrast. The first-person narrators appear to be more given to reference by pronouns as is illustrated by westerly positions of the pronoun it and the demonstrative that, while the third-person narrators seem to favour more exact and explicit reference, which is implied by a set of determiners and the two relatives lying towards the eastern extremity of the chart.

The word-pattern along the horizontal axis point to contrast between the more formal, generalising style and the more subjective, emphatic style. It may not be inappropriate to associate the former with the overall tendency to describe things and events externally (from a so-called "omniscient" point of view) in the third-person narrative. The latter may illustrate one feature of the language of narrative in which a narrator's sense of value tends to be more freely reflected.

From a chronological viewpoint, the relatives which and who, lying at the south-western end of the chart, are words that discriminate strongly in favour of the texts written in 1830s. What is interesting here is the distance that separates another relative, that, from this set: while the early style is characterised by pronounced recourse to WH-relatives, the later Dickens is inclined in comparative terms to that-relatives. Such different distributions of that-relatives and WH-relatives are in keeping with Beaman's (1984) treatment of the two types of relatives as separate classes. Beaman furthermore finds that-relatives highly associated with spoken narratives and WH-relatives with written narratives. Major prepositions like of and in, which often forms postmodifying phrase, are also found in the early Dickensian quarter, together with the set of articles.

Words situated at the north-eastern, or late Dickensian, end of the chart include the adverbial particles out and down, the pronoun it, the prepositional like (often used as in comparison), words often associated with oral style. Burrows (1992), in treating the chronological shift of English first-person narrative over the past three centuries, notes the increasing incidence of the adverbial particles and the pronoun it towards the present. The mildly increasing incidence of the connectives and and then is also recognised in the later works. Such an overall oral and less formal tendency in the later narrative stands in contrast with the comparatively more elaborate and formal cast of the early narratives, which is reflected in the predominance of WH-relatives, the determiners, and some of major pronouns. The words denoting parts of body, such as hand, eyes, head, and face, also characterise the texts written after 1849.

On a closer study of Fig. 2, another interesting feature is brought into relief. The first segments tend to lie about the outskirts in the chart, whereas other segments are positioned in relative terms nearer to the centre. Nine of the eleven first segments lie towards the various extremities. Three of the four first segments from the 1830s' texts (SB#1, OT#1 and NN#1) are in the south-east, typifying the early Dickens's preference for nominal and prepositional phrases (as is reflected in the comparative weighting on articles and prepositions) and tendency to elaboration. The anachronistically placed BH#1 (as noted earlier) accompanies the trio. Its counterpart in Esther's narrative (Esther#1) are at the westmost end, demonstrating predominance of words related to intensification and negatives.(8) The remarkable contrast between the pair reflects how Dickens, as Page (1990) points out as the result of a manuscript study, took pains to differentiate sharply the two narrators' voice (especially at the outset of each narrative) at different stylistic levels, to say nothing of tense and person. ED#1 and TTC#1, standing furthest from both the early and the first-person narratives, show the most sparing use of "old-fashioned" elaboration.

The seemingly aberrant position of BH#1 in Fig. 2 needs some explanation here. A close inspection of the word frequency table and a concordance for each segment of texts reveals that, whereas the anonymous narrator shares most of the late Dickensian characteristics, he finds more needs to use who than any other late narrator at the outset of the novel where he mentions the people involved in the notorious law suits.(9) This narrator, on the other hand, makes comparatively limited reference to the bodily words eyes, face, head, and hand, to which the narrators of other late novels tend to refer more frequently. Such, rather thematic, factors seem to function both positively and negatively in placing BH#1 apart from the other later narrators. The idea will be given credence in the analysis that I show in the next section.

2.2 Distribution test and the strongly discriminating words
Though the Fig. 2 shows a general contrast between the texts written in the 1830s and those written after 1849, the contrast is somewhat blurred by the fact that the principal vector is dominated by the difference between the first-person and third-person narratives. To diminish the effect of the first-person subset and gain a more rigorous distinction, it seems necessary to establish a basis of comparison excluding the first-person narratives.

Student's t-test is a tool often used for comparison between two groups. It gives a measure of judgement on whether discrepancy between two sets of data are significant enough to be pursued, by calculating the divergences from means. With the help of t-test, I compared the twelve segments of the early third-person narratives with the eleven segments of the later. The summary of the result is shown in TABLE 2.  

 No. Rank Word-types 1830s-mean 1849~-mean  t  Probability  DF



  1. 15   which(r)    0.732      0.387     7.7    0.0000    20
  2.  9   it          0.696      1.201    -6.94   0.0000    20
  3. 26   out         0.1761     0.3662   -5.12   0.0001    20
  4. 65   like(p)     0.0601     0.1465   -4.31   0.0003    20
  5. 48   again       0.1173     0.1966   -3.21   0.0044    20
  6. 32   very        0.341      0.1935    3.12   0.0057    19
  7. 39   down        0.1346     0.2466   -3.03   0.0066    20
  8. 57   gentleman   0.279      0.0726    2.97   0.011     13
  9. 41   more        0.2054     0.1561    2.66   0.018     15
 10. 63   great       0.194      0.1023    2.59   0.021     15
 11. 5    in          1.893      2.166    -2.4    0.026     20
 12. 54   than        0.1230     0.1908   -2.36   0.030     18
 13. 51   time        0.1717     0.1162    2.35   0.030     19
 14. 62   face        0.1130     0.199    -2.36   0.032     15
 15. 36   so(adv.d.)  0.1614     0.2185   -1.99  #0.060     20
 16. 46   other       0.2254     0.1648    1.98  #0.062     20
 17. 49   that(r)     0.128      0.224    -1.98  #0.062     20
 18. 49   its         0.1482     0.286    -2.03  #0.065     12
 19. 47   over        0.1547     0.2133   -1.87  #0.077     20
 20. 64   hand        0.0944     0.187    -1.86  #0.087     12
 21. 16   for(p)      0.559      0.430     1.74  #0.098     18

Words with # are those with higher probability score than "significant" level
(p > 0.05). "Rank" indicates the original word frequency hierarchy. "DF" stands
for degree of freedom. (adv.d.) = adverb of degree, (p) = preposition, (r) =
relative. For probability level, the score shown in MINITAB as "0.0000" is
too small to be round up to 0.0001.



TABLE 2 Dickens's Novels in 1830s versus Novels after 1849:
The 21 most common strongly discriminating words. (Sorted
in the ascending order of "probability" level.)
The t-score is given in the fifth column. The positive value indicates that the word discriminates in favour of the early subset. The negative value means the converse. The sixth column indicates a probability that chance has caused discrepancy in word-frequency between the two sets. Though a probability of 0.05, one in twenty, is often used as a threshold at which the result is considered statistically "significant," I included in the list the words with probability level more than 0.05 and less than 0.1. A score of 0.01, which is regarded as "highly significant," means that difference between the two groups is 99 % reliable. The corollary, however, also must not be forgotten: provided the calculation is accurate, such a difference does happen by chance once in a hundred trials. In such a case the result is dismissed as not significant. However, when such a (or smaller) probability score is gained several times in a hundred trials, it becomes possible to take the result as significant. In the present case, in which seventy-four words are subjected to the test, seven words show the score less than 0.01, and another seven less than 0.05. The discrepancy between the texts in 1830s and those after 1849 in the use of which, it, out, and like , which I noted earlier, are demonstrated again as "very highly significant" (p 0.001).

Fig. 3 Early Style versus Late Style: Text-plot
(Based on the 21 most common "strongly discriminating" words of the corpus.)

Fig. 4 Early Style versus Late Style: Word-plot
(Based on the 21 most common "strongly discriminating" words of the corpus.)

Figs. 3 and 4 are derived from principle component analysis of the twenty-one commonest "strongly discriminating" words listed in TABLE 2. The salient feature of Fig. 3 is sharper differentiation between the texts written in 1830s and those written after 1849, now displayed along the most powerful vector. There is no interpenetration between the two subsets, with BH#1 and BH#2 sitting where they should be reasonably expected. Esther's (#2-4) positions in the "early Dickens" side of the "late Dickens" cluster might seem inappropriate at first thought, but this phenomenon can be accounted for by her marked preference for intensification. In fact, she stands highest among the later narrators in the incidence of "very," "more," and "great." In this respect, she is sharply differentated from other first-person narrator, David and Pip.

Another noteworthy feature is a contrast between the later first-person and the later third-person narratives along the second strongest vector. It is remarkable that the set of twenty-one "chronologically-oriented" words still has a force of separating the two narrative modes, though less sharply than in Fig. 2. Taking into account the fact that I excluded the first-person narratives when I extracted the set of differentiae, it would not be unreasonable to regard the clusterings as formed by a chronological determinant.

Fig. 5 Early Narratives, Late Narratives and Three Histories: Text-plot
(based on the 21 most common "strongly discriminating" words of the corpus)

The force of the distinction based on the twenty-one words (or to say sceptically the possibility that this result came out of mere freak of chance) can be tested by introducing a new set of texts. Here three shorter first-person narrative will be employed as a control set. Fig. 5 summarises the result in a convincing way, by placing the new set in appropriate clusters. Two histories, "the stroller's tale" and "the convict's return" (both from The Pickwick Papers) lie in the 1830s' cluster, while Miss Wade's narrative (from Little Dorrit) lies in the "after-1849" cluster. Interestingly, while Miss Wade finds herself in the "first-person domain", neither the stroller nor the convict are distinguished from the third-person narratives. The essential configuration of the graph is not dissimilar to Fig. 3. Since the three new texts did not participate in the compilation of the original word-list or in the selection of the discriminating words, I take this evidence as confirming that Dickens's early and later styles can be distinguished from each other in quantitative terms.

3. Final Remarks
The evidence this paper has shown is not only consistently in keeping with my earlier findings from a smaller set of texts, but also capable of distinguishing DickensÕs early and later styles with a sharpness that could not be achieved before. Whether to regard such a change in oneÕs style as a development or not is a matter of aesthetic judgements, and the approach I pursued in this study does not lead to any such value judgements. But the evidence from very common vocabulary of DickensÕs narrative at least betokens a movement from a style that characterises Sketches by Boz and The Pickwick Papers to a style typically demonstrated in David Copperfield, from a more formal and elaborate style to a style of an increasing oral tendency.

In this approach, I excluded personal pronouns and finite verbs due to the difference in narrative mode and tense. By limiting comparison to a particular narrative mode (either first-person or third-person), frequency distribution of personal pronouns can be examined in much the same way. For the treatment of finite verbs, lemmatisation is another option, by means of which it is possible to process a number of verbs statistically.

My next step, then, is to complement this approach by focusing attention on what I have excluded from this study.

Notes
(*) This study came out of a group project being pursued at the Centre for Literary and Linguistic Computing, the University of Newcastle, NSW, Australia. I would like to thank John Burrows, Alexis Antonia, Hugh Craig, and Wayne McKenna for technical advice given at all times, and to Nicole Cox for her service in the preparation of texts. I am also grateful to John Lambert for software support.

(1) The Oxford Illustrated Dickens, 21 vols (London: OUP) is the source of the copytexts for:

Sketches by Boz (1989, 1st p.: 1836)
David Copperfield (1987, 1st p.: 1850)
Great Expectations (1992, 1st p.: 1860)
Our Mutual Friend (1989, 1st p.: 1865)
The Penguin English Library (Harmondsworth: Penguin Books) is used as copytexts for:
The Pickwick Papers (1986, 1st p.: 1837) ed. Robert L. Patten
Nicholas Nickleby (1978, 1st p.: 1839) ed. Michael Slatrer
Oliver Twist (1985, 1st p.: 1838) ed. Peter Fairclough
Bleak House (1987, 1st p.: 1853) ed. Norman Page
A Tale of Two Cities (1970, 1st p.: 1859) ed. George Woodcock
The Mystery of Edwin Drood (1985, 1st p.: 1870) ed. Arthur J. Cox
(2) In principle, a chapter ending nearest to twenty-thousand words was taken as the end of sample in this study. For Bleak House, however, the end of sample was prolonged to the end of chapter seven to generate two narrative samples. The size of narrative text differs considerably from text to text according to the portion dialogue and free indirect discourse occupy in each sample.
The distinction between narrative and dialogue adopted in this paper is based upon the presence of quotation marks, a crude but tolerably objective distinction. Within the domain of narrative there is a further need to distinguish "pure-narrative" and the special category of fictional discourse, which is often referred to as "free indirect discourse (FID)," and in which the "voice" of the narrator and that of the character that s/he reports are merged into a hybrid style. How and where to set the boundary between pure-narrative and FID is a touchy question: but I base my distinction between pure-narrative and free indirect discourse on my interpretation of the texts.
(3) John Burrows was kind enough to offer the machine-readable set of these three narratives, which were used as Dickens's text-base in his analysis of first person retrospective narratives. See Burrows (1992 and 1994).
(4) Works by Brainerd (1979), Burrows (1987a), and Craig (1991) show stimulating results of statistical analyses of a set of personal pronouns.
(5) "Word-type" will be referred to as "word" henceforward unless it is necessary to distinguish. It may be necessary to add that considerable amounts of time and efforts are needed for post-editing of texts apart from proof-reading: for the purpose of counting, contractions are expanded (e.g. "can't" counts as "can" and "not"), proper names like "Mr Copperfield", on the other hand, are united with asterisk ("Mr*Copperfield") so as to count as one word. Words that are usually hyphenated or treated as one word but do not appear as one word in my texts are also joined ("for ever" is joined as "for*ever"). Furthermore, common homographic forms are tagged to specify each usage: for example, "that" is separated into "that(c)" (conjunctive), "that(r)" (relative), "that(d)" (demonstrative), and so on.
(6) Verbs that ranked within the top 100 word-types are finite forms of "be" and "have," "came," "looked," "made," "replied," "could," and "would."
(7) All the calculation in this study was undertaken using MINITAB Accelerated, release 8.2, The Apple Macintosh Version. (Minitab Inc, State College, PA, 1991).
(8) On negation, Ian Watt (1960) states that "there are no negatives in nature, but only in human consciousness." If we apply Watt's philosophical remarks in the present case, recourse to negation can be considered a measure of subjectivity. Watt, 274.
(9) Conc v.1.71 beta (Summer Institute of Linguistics, Dallas TX, 1992) was used for building a computerised concordance to the texts I analysed.

Bibliography
1. Books
Biber, Douglas. (1988) Variation across Speech and Writing (Cambridge: Cambridge UP).
Burrows, J. F. (1987a) Computation into Criticism: A study of Jane Austen's novels and an experiment in method (Oxford: Clarendon Press).
Cluett, Robert. (1976) Prose Style and Critical Reading (New York and London: Teacher's College Press, Columbia University.
Gordon, Ian A. (1966) The Movement of English Prose (London: Longman, "English Language Series, 2").
Page, Norman. (1990) Bleak House: A Novel of Connections (Boston: Twayne Publishers).
Stevens, James. (1986) Applied Multivariate Statistics for the Social Sciences (Hillsdale, N.J.: Lawrence Erlbaum Associates, Publishers).
Toolan, Michael J. (1988) Narrative: A Critical Linguistic Introduction (London and New York: Routledge, "The Interface Series, 1").

2. Papers
Beaman, Karen (1984). "Coordination and Subordination Revisited: Syntactic Complexity in Spoken and Written Narrative Discourse" in Tannen, pp. 45-80.
Biber, Douglas and Edward Finegan (1989), "Drift and the Evolution of English Style: A History of Three Genres", Language, Volume 65, 3, pp. 487-517.
Biber, Douglas (1993). "The Multi-Dimensional Approach to Linguistic Analyses of Genre Variation: An Overview of Methodology and Findings", Computers and the Humanities, 26, pp. 331-345.
Brainerd, Baron (1979). "Pronouns and Genre in Shakespeare's Drama", Computers and the Humanities, 13, pp. 3-16.
Burrows, J. F. and D. H. Craig (1994). "Lyrical Drama and the "Turbid Mountebanks": Styles of Dialogue in Romantic and Renaissance Tragedy", Computers and the Humanities, 28, 63-86.
Burrows, J. F. (1987b). "Word-Patterns and Story-Shapes: The Statistical Analysis of Narrative Style", Literary and Linguistic Computing, 2, pp. 61-70.
---. (1992) "Computers and the Study of Literature" in Butler (1992), pp. 167-204.
---. (1994) "Tiptoeing into the Infinite: Testing for National Differences in the Language of English Narrative" in Hockey and Ide.
Butler, Christopher (1992), ed. Computers and Written Texts (Oxford: Blackwell).
Craig, D. H. (1991). "Plural Pronouns in Roman Plays by Shakespeare and Johnson", Literary and Linguistic Computing, 6, pp. 180-186.
---. (1992), "Authorial Styles and the Frequencies of Very Common Words: Jonson, Shakespeare and the Additions to The Spanish Tragedy", Style, Volume 26, 2, pp. 199-220.
Hockey, Susan and Nancy Ide (1996), eds. Research in Humanities Computing 4 (Oxford: Clarendon Press)
Tabata, Tomoji. (1993) "The Language of Dickens and Its Computer-Based Evidence: A Step towards a Chronological Study", Kumamoto Studies in English Language and Literature, 36, pp. 116-134.
Tannen, Deborah, ed. (1984) Coherence in Spoken and Written Discourse (Norwood, N.J.: Ablex).
Watt, Ian (1960). "The first Paragraph of The Ambassadors: an explication", Essays in Criticism, 10, pp. 250-74.