Introducing the Pattern-Finder

We would like to present the Pattern-Finder, a new computational tool for analyzing texts with specific application to the close reading of poetry. It was designed and created by Smolinsky & Sokoloff, and can be found at www. patternfinder.net. Our program was
invented to examine whether (and if so, how) the phonetic/
phonological structure of a poem may contribute to its meaning and emotional power—to take a close look at how the ‘music’ of poetry actually works. Two Pattern-Finder studies have already been completed (on Keats’ Bright Star sonnet, and on the Wallace Stevens lyrics
Bantams in Pine-Woods and Fabliau of Florida). These, we claim, uncover aspects of poetic technique
inaccessible to the naked ear; displays from these
studies will form part of our presentation. A third study, on
Hopkins’ Terrible Sonnet No Worst, There Is None, is being proposed as a paper for this conference. Below,
we describe briefly the theoretical background to our
invention; then we describe how the Pattern-Finder is used
on a text, and the type of sound patterning it is designed to bring to light.
Our approach comes from combining two sources, one literary and one linguistic. The first is the traditional
critical close-reading focus on sound repetitions, on
phenomena like alliteration, assonance, rhyme and meter
(for the closest predecessor to our approach, see Burke,
1973). The second derives from the study of a topic in linguistics known as phonetic symbolism. This is an
ongoing area of research, which started in the U.S. over 70 years ago with Edward Sapir’s (1929) experimental Investigation in phonetic symbolism. Following Sapir, many researchers have continued investigating whether isolated speech sounds are not in themselves experienced
as carrying some intrinsic meaning. There is growing
evidence that they can indeed trigger attribution of qualities,
such as smallness or largeness, brightness or darkness. (See, among many others, Chastaing (1958, 1962)
Fischer-Jørgensen (1978) Fonagy (1979) Magnuse (1999) Newman (1933) Smolinsky (2001) and Tanz (1971)).
In one part of Sapir’s experiment, subjects were given two made-up words for “table,” mal and mil (/mal/ and /miyl/ in the SAMPA (Wells, 1997) transcription system we are using). The subjects then had to decide which table was ‘bigger’ and which was ‘smaller.’ Since the vowels in these words were varied while the surrounding frame of consonants were kept the same, in effect, it was the perceived ‘size’ of the vowels which was being tested. Sapir, testing a range of vowels, found there was significant consistency of response, and Newman (1933) continuing
Sapir’s experiment four years later, found enough
consistency in subjects’ responses to go as far as setting up scales for vowels and consonants, showing brightest to darkest and largest to smallest.
Certain regularities in both Sapir’s and Newman’s
findings suggest that the way such attributions of qualities
to speech sounds may work is by features: “Phonetic propert[ies] used to classify sounds,” (Ladefoged, 1982, p 38). For example, one such feature of consonants is voicing (whether they are made with or without vibration of the vocal cords, as in the voiceless /t/ and /s/, which have as their voiced counterparts /d/ and /z/). To illustrate, there is a pronunciation rule in English that when a noun ends in a voiceless consonant, its plural suffix will also be voiceless: cat /kAt/ pluralized is cats /kAts/. But if the noun ends in a voiced consonant, for example dog /dQg/, then its plural will also be voiced: dogs /dQgz/. This rule also has an important extra clause: if, as in midge /mIj/ or roach /rowC/, the noun ends in a voiced or voiceless consonant which is coronal (i.e., one made by raising the blade of the tongue: (Chomsky and Halle, 1968) and
sibilant (characterized by marked loudness and a sustained
high-frequency noise (Ladefoged, op cit.), then the plural suffix is realized as a whole syllable: midges /mIjIz/ or roaches /rowCiz/.
For our purposes, what is relevant about this kind of pronunciation rule, operating below the conscious
level, is that if native English-speakers are given a series of made-up nouns, such as drobe /drowb/, bloph /blQf/ or nutch /nVC/, and asked to pluralize them aloud, they
can, without any apparent effort, produce the correct
plurals for all three unknown words. It seems that they automatically scan each noun’s last sound for the presence of, respectively, voiced, voiceless and sibilant-plus-coronal features, and assign the suffixes accordingly. Thus, it can be strongly argued that features like voiced or coronal
must be available in our subconscious for obeying
pronunciation rules. If so, then they should also be available for symbolizing use, as in Sapir’s “large” and “small” mal and mil examples. We, as readers, might be affected by such symbolizing material in a poem (for example, a sense of spaciousness or constriction) without quite knowing what in its music has triggered our response,
any more than we ‘know’ why the plural of mirage is mirages with its extra syllable. The exploration of this hypothesis—that feature-patterning is the driving
force in the ‘music’ of poetry—was the idea behind the
creation of the Pattern-Finder.
The process of Pattern-Finder analysis is as follows: Firstly, we put the text into a (broad) phonetic transcription; this is an essential step, since a consistent relationship
between sounds and their written representations is
needed—one hardly provided by normal English
orthography. We have chosen the SAMPA phonetic transcription alphabet (see Wells, op.cit.) because it uses the same characters as a normal keyboard: once the
reader has the key to the system, it is immediately available
for use. Prosodic features (three degrees of stress, two pause lengths) are also transcribed with keyboard
characters. The text is then input to the Pattern-Finder with a search command for a feature or features; next, the program analyzes it for the feature(s) required, and finally the analyzed version of the text is output with
highlighted displays of the distribution of the feature(s), together with the frequency count(s).
Most of the Pattern-Finder’s features are the conventional ones found in any phonetics textbook, such as high for a vowel, or bilabial for a consonant. But a number of extra features were added, wider-ranging ones (such as upper, which includes both high and mid-high vowels, or labial, which includes any consonant made with the lips) and more narrowly-focused ones, such as breathy and sibilant, which distinguish between higher-pitched
and lower-pitched sounds involving frication. In
designing our range of features, the intention was to imitate as much as possible of the range of unconscious
phonetic/phonological distinctions in the mind of the
English-speaking poet and reader. In this way, we hoped to capture more of the sound-patterning, broad or narrow, which we hypothesize underlies the music of a poem. The Pattern-Finder offers an extensive range of analyses, single or combinatory, segmental and/or prosodic. Users may analyze the text for up to four features in one search.
They may request “one-highlight” displays, i.e., of a
single set of speech sounds identified either by one feature
(e.g., all the high vowels) or by two features (e.g., all
the high and back vowels). Or they may request two-highlight “versus” dsiplays, i.e., of two sets of speech sounds identified by either one feature (all the high versus all the low vowels), or by two features (all the high and back versus all the low and back vowels). The availability of prosodic features means that users may also search, e.g., for all the stressed and high vowels in a text, or stressed and high vowels versus pre-pause and voiceless consonants.
The resulting highlighted displays of speech sounds,
picked out by one or two features, allow the user to track
their distribution throughout the text. In this way, one is able to match feature-patterning (for example, whether this sound appears in clusters at one or two places in the text, or more sparsely but at regularly-spaced intervals, or is almost, or totally, absent) to meaning (such as contrasts
of atmosphere or subject, correspondences between
apparently disparate images, locus of climax, and so on).
Naturally, we are most eager for the chance to share this program with our peers. (Besides English, we can also
offer clear explanations in French, Russian, Spanish,
Catalan, and Hindi.) We feel that we have invented a kind of X-ray machine for poetry—or, indeed, for longer texts—and we are keen to invite all who love literature to join with us in using this new instrument for investigating
its sound-structures.
References
Phonetics and Phonological Features
Brosnahan, L F. and B. Malmberg (1970) Introduction to phonetics, Cambridge, UK: Cambridge University
Press
Chomsky, N. and M. Halle (1968) The sound-pattern of English. New York: Harper & Row
Gimson, A. C. (1989) Introduction to the pronunciation of English. (4th edition) London: Edwin Arnold
Hyman, L. M. 1957 Phonology: theory and analysis. New York: Holt, Rinehart & Winston
Jakobson, R., G. Fant and M. Halle (1951) Preliminaries
to speech analysis: the distinctive features and their correlates. Cambridge MA: MIT Press
——— and M. Halle (1971) Fundamentals of language. The Hague: Mouton
Ladefoged, P. (1962) Elements of acoustic phonetics. Chicago and London: University of Chicago Press
——— (1982) A course in phonetics. (2nd edition) New York: Harcourt, Brace, Jovanovich Inc.
Peterson, G. E and H. L. Barney (1952) Control methods
used in a study of the identification of vowels,
Journal of the Acoustical Society of America
v 24 #2, pp 175-184
——— and I. Lehiste (1960) Duration of syllable
nucleii in English. Journal of the Acoustical Society of America v 32, pp 693-703
Pickett, J. M. (1980) The sounds of speech communication.
Baltimore, MA: University Park Press
Phonetic Symbolism and Literary Criticism
Burke, K. (1973) “On Musicality in Verse” in The
Philosophy of Literary Form. Berkley, CA: University
of California Press
Chastaing, M. (1958) Le symbolisme des voyelles,
signification d’i, Journal de Psychologie v 55, pp 403-423 & 461-481
——— (1962) La brillance des voyelles, Archivum
Linguisticum v 14, pp 1-13
Fisher-Jørgensen, El. (1978) On the universal character
of phonetic symbolism with special reference to vowels, Studia Linguistica v 32, pp 80-90
Fónagy, I. (1979) La métaphore en phonétique. Ottawa: Marcel Didier
——— (1983) La vive voix. Paris: Payot
Grammont, M. (1946) Traité de phonétique. Paris:
Delagrave Jakobson, R. (1978) Language in literature. Cambridge, MA: Belknap Press of Harvard University
Jespersen, O. (1933) The symbolic value of the vowel i in Linguistica: Selected writings of Otto Jespersen. Copenhagen: Levin & Munksgard pp 283-303
Magnus, M. (1999) Gods of the Word. Kirksville MO: Thomas Jefferson University Press
Marchand, H. (1959) Phonetic symbolism in English word-formation, Indogermanische Forschungen
v 64, pp 146-68, 256-277
Newman, S. S. (1933) Further experiments in phonetic symbolism, American Journal of Psychology v 45, pp 53-75
Ohala, J. J. (1994) The frequency code underlies the sound-symbolic use of voice-pitch, in Sound
Symbolism (eds Hinton, L., J. Nichols and J. J. Ohala)
Cambridge, UK: Cambridge University Press
Sapir, E. (1927) Language as a form of human behavior, The English Journal v 16, pp 421-433
——— (1929) A study in phonetic symbolism, Journal of Experimental Psychology v 12, 225-239
Smolinsky, Stephanie (2001) Brilliance, energy and size in vowels: a cross-linguistic study of phonetic symbolism. Unpublished Ph.D. dissertation, CUNY Graduate Center, NYC.
Tanz, C. (1971) Sound symbolism in words relating to proximity and distance, Language and Speech v 17, pp 87-94
Transcription
SAMPA www.phon.ucl.ac.uk/home/sampa
Wells, J. C., (1997) ‘SAMPA computer readable phonetic alphabet’. In Gibbon, D., R. Moore and R. Winski, (eds.) Handbook of Standards and Resources for Spoken Language Systems. Berlin and New York: Mouton de Gruyter. Part IV, section B.