Computational Criticism:
Using Quantitative Methods
for the Interpretation of Poetry
□ Distant reading and close
reading
□ Close reading and distant
reading focus on different
textual aspects
□ Distant reading is often
used to study questions of
a factual nature
Franco Moretti,
Graphs, Maps, Trees
□ Can this gap between
traditional criticism and
computer-based literary
research be bridged?
□ Analysis of the collected
poems of Louis MacNeice
□ Language is "a traffic in
symbols" and that "these
symbols are plastic - an
endless annoyance to the
scientist but God's own
gift to the poet"
(Experiences with Images,
p. 155)
□ Aims of the study: (1) production of data
on literary devices such as rhyme, meter,
alliteration, assonance, onomatopoeia and
other devices; (2) statistical analyses and
visualisations of these data
□ Collections of machine readable versions
□ Basic automated application of TEI
encoding
□ POS tagging and lemmatisation by
Morphadorner
□ Phonetic Transcriptions added via
pronunciation dictionary (MRC
Psycholinguistic Database)
□ Applications and visualisation have been
programmed using Perl, Python, R and
Processing
Perfect rhyme
□ Exact agreement in final phoneme sequences
Down the road someone is practising scales
The notes like little fishes vanish with a wink of tails
(Sunday Morning, ll. 1-2)
□ Difficulties
□ “Near-matches” such as “use” and “loose”;
□ Repeated words at the end of lines;
□ Number of lines to be analysed;
□ Slant rhyme: matches in final consonants or in
final vowels
It's no go your maidenheads it's no go your culture / All
we want is a Dunlop tyre and the devil mend the
puncture
(Bagpipe Music, ll. 13-14)
On shining lines the trams like vast sarcophagi move /
Into the sky plum after sunset merging to duck's egg
barred with mauve
(Birmingham, ll. 25-26)
□ Semi-rhyme: an agreement between a
masculine rhyme and the final stressed syllable
of a feminine rhyme
Finding pathos in dogs and undeveloped handwriting /
And young girls doing their hair and all the castles of
sand
(Wolves, ll. 3-4)
Distant Scansion
□ Data about the meter was obtained from
phonetic transcriptions
□ The patterns were classified:
□ -X-X-X-X-X
Iambic Pentameter
e.g. “To every question gives the same reply”
□ Catalectic iambic trimeter
-X-X-Xe.g. “We are dying, Egypt, dying”
□ X-X-X-X
Catalectic trochaic trimeter
e.g. “Neither sense nor conscience stirred”
□ Difficulty: Regular stress patterns of a word
may change in the context of a poetic line
e.g. The Sunlight on the Garden / Hardens
and grows cold
Alliteration
□ Repetitions of phonemes at the beginning of
stressed syllables.
Frozen into his blood from the fire in his basalt
f--b-- f--b
(Belfast, l. 2)
And hanging smell of sweetest hay / Weavingly laughingly
leavingly weepingly
-hs–sh/wllw
(Western Landscape, ll. 20-21).
□ Images
□ Critics have discussed images such “the
sea”, “trains”, “church bells”, “Wind”,
“Clocks”
□ USAS Semantic Tagger; Most common
categories: B1 (“Anatomy and physiology”),
M6 (“Location and direction”), M1 (“Moving,
coming and going”), O2 (“Objects
generally”), O4 (“Physical attributes”), T1
(“Time”)
□ Topic Modelling / MALLET: Poor results
□ Manually edited word list created using
USAS and WordNet
□ Themes
□ Summaries of secondary literature: 15
themes, in 3 broad categories
□ Themes assigned manually to poems
□ Metrics produced on perfect rhyme, slant
rhyme, semi-rhyme, deibhide rhyme, internal
rhyme, alliteration, assonance, consonance,
imagery, themes, onomatopoeia, meter,
sentiment analysis (positive and negative
connotations)
□ Algorithms for detection of literary
phenomena are inevitably imperfect
□ 102,907 observations about a corpus of
127,555 tokens or 16,782 verse lines
How to count the “extent” of alliteration?
Further who failed last Friday to feel grieved
(Easter Returns)
The flotsam of private property pekinese and
polyanthus
(An Eclogue for Christmas)
Round the corner is sooner or later the sea
(Round the Corner)
Standard deviations
Semi-rhyme and themes
Wind_Z
Water_Z
Time_Z
Thread_Z
Stone_Z
semiRhy me_Z
romanticLonging.nature
romanticLonging.lov e
romanticLonging.escapism
romanticLonging.childhood
romanticLonging
Religion_Z
Quest_Z
Positiv e_Z
perf ectRhy me_Z
onomatopoeia_Z
Negativ e_Z
Mov ement_Z
modernity .nihilism
modernity .massProduction
modernity .doom
modernity .consumerism
modernity .alienation
modernity .agesOf Man
modernity
metaphy sics.plurality
metaphy sics.my stery
metaphy sics
Ireland_Z
internalRhy me_Z
deibide_Z
consonanceTwoLines_Z
consonanceRhy me_Z
consonance_Z
Bell_Z
assonanceRhy me_Z
assonance_Z
alliteration_Z
aicill_Z
aic
a
a
a
B
c
c
c
d
in
Ir
m
m
m
m
m
m
m
m
m
m
M
N
o
p
P
Q
R
r
r
r
r
r
s
S
T
T
W
W
ill_ llit er s son s son ell_Z onso onso onso eibid t ern elan etap etap etap ode ode ode ode ode ode ode ove ega nom erf e os iti ues elig om a om a om a om a om a emiR t one hrea ime_ ate ind
c
ion ntic ntic ntic ntic ntic
v
t_
ati
m
ti
Z
a
r
r
r
r
r
r
r
na na na e_ alR d_
hy _Z d_Z Z r_Z _Z
on anc e anc e
Z hy sic hy sic hy sic nity nity .a nity .a nity .c nity .d nity .m nity .n ent_ v e_Z topo t Rhy e_Z Z
_Z
Z
nc nc nc
L
L
L
L
L
m
h
o
o
o
o
o
y
_Z
_Z Rh
e_ eR eT
ng ng ng ng ng e_Z
eia me
s
s. m s. p
me
Z
ge lie on oo
i
Z
ym
hy w o
ing ing ing ing ing
s O na sum m ass P hilis m
_Z _Z
_Z
y st lura
me Lin
e_
.c h .es .lov .na
fM t ion
rod
ery lity
eri
Z
_Z es
an
ild cap e
t ur
uct
sm
_Z
ho
e
ion
od is m
Conclusions
□ Text mining techniques can be used to identify
literary devices, to some extent
□ Analyses of data of about literary devices can
provoke interpretation of the texts
□ Representative and reproducible findings
□ Human critic is necessary to explain and to
interpret patterns