1
Formal Description of Slavic Languages
20-22 November 2024
Brno
Gaps in initial RT clusters in Czech:
accidental or systematic?
Evidence from dichotic listening
Tobias Scheer Université Côte d'Azur, CNRS 7320
Lucie Braunerová Masaryk University Brno
Martin Lemaître Université Côte d'Azur, CNRS 7320
Anna Poĺomská Masaryk University Brno
2
Intro: dichotic listening
put
• pay in the left ear: pay (L)
lay pay
play
English natives will perceive
• play
Day, R. S. 1969. Fusion in dichotic listening. Ph.D
dissertation, Stanford University.
Cutting, James E. 1975. Aspects of Phonological
Fusion. Journal of Experimental Psychology 104:
105-120.
this is called dichotic fusion:
• the percept combines items of two
(incongruent) inputs that come in
through two different channels
• here audio-audio
• but also audio-visio: McGurk effect
fusion rate
• fusion does not always succeed:
• about 35% (Cutting 1975)
• lay in the right ear: lay (R)
3
Intro: dichotic listening
put in a temporal lag
lay
100 ms
later
pay
play
0
English natives still perceive play
lay (R) pay (L)
100 ms
time
• then play pay (L)
• wait 100 ms
• first play lay (R)
why?
• because play exists in the lexicon, but lpay does not.
• because English phonology disallows #lp phonological effect
lexical effect
Day, R. S. 1969. Fusion in dichotic listening. Ph.D
dissertation, Stanford University.
Cutting, James E. 1975. Aspects of Phonological Fusion.
Journal of Experimental Psychology 104: 105-120.
4
Intro: dichotic listening
lexical vs. phonological effect
if we can isolate a phonological effect,
• it should depend on the phonology of the speaker
• by using nonce words
• they don't exist in the lexicon
• and therefore cannot influence perception
• how can the phonological effect be isolated?
TR-only languages
• word-initially, only TR clusters occur: *#RT, *#RR, *#TT
• English, French, German etc.
• here the phonology should impose a TR-percept, since #RT is illegal
anything-goes languages
• word-initially, any sonority combination occurs: #TR, #RT, #RR, #TT
• (Modern) Greek, many Slavic languages: Russian, Polish, Czech, Ukrainian, etc.
• here the phonology is happy with non-TR percepts, which should therefore be produced
T = obstruent
R = sonorant
5
prediction #1
prediction #1
when simultaneously (no lag) confronted with #T in one ear and #R in the other,
TR-only languages:
• speakers will perceive #TR
• because their phonology does not tolerate any other combination of T and R.
anything-goes languages
• speakers will perceive either #TR or #RT at chance level
• because their phonology does not prefer or disprefer either
• and they have no evidence for going in one direction or the other
recall
• this supposes the elimination of a putative lexical effect:
• otherwise we are not sure what the driving force of the perception is.
6
Experiment #1
real vs. nonce words in
French
7
experiment #1: French
plan
• in a TR-only language, French, try to isolate
• the lexical and the
• phonological effect.
• then see in an anything-goes language, Czech, whether speakers indeed perceive
#TR / #RT at chance level, as predicted.
8
experimental settings
• terminology
• source word = the two input words, e.g. buder (R) and ruder (L)
• target word = the intended fusion of the source words, bruder
• alignV
• in the doctored file that combines the two source words, e.g. buder (R) and ruder
(L),
• the onset of the vowel following T in buder is aligned (= simultaneous) with the
onset of the vowel following R in ruder.
experiment #1: French
9
experimental settings
• 50 participants
• 64 stimuli
• 32 real words
• 32 nonce words
• participants are exposed to the 64 stimuli once. Thus 1600 (real) + 1600 (nonce) =
3200 trials
• randomization of the order in which stimuli are presented
• randomization of left - right ear (to level out putative one-sided hearing deficiency)
experiment #1: French
10
experimental settings
• real words
• the source words (dainer, rainer) do not exist (to avoid a bias in favour of existing
items)
• the target word exists: drainer
• nonce words
• the source words (buder, ruder) do not exist
• the target word does not exist either: bruder
experiment #1: French
11
experimental settings
• free choice
• participants hear the doctored stereo file three times in a row through a headset
(750ms interval)
• and are then asked to say what they have heard (keyboard input)
experiment #1: French
12
results
(simultaneous R-L)
1417
171
2 7 3
1149
435
3 9 4
0
200
400
600
800
1000
1200
1400
1600
CV TR RT TT other
nbresponses
nonce vs. real
(simultaneous)
nonce real
• CV: response buder or ruder for target
bruder
• TR: response bruder for target bruder
• RT: response rbuder for target bruder
• TT: response bduder for target bruder
• other: blank, uninterpretable
experiment #1: French
13
results
fusion rate
• the number of responses with a #CC of
the total of valid trials (i.e. not counting
in "other")
there is a lexical effect
• real (target) words fuse much better than
nonce words
experiment #1: French
12,7
38,9
0,0
10,0
20,0
30,0
40,0
nonce real
%ofvalidtrials
fusion rate
(simultaneous)
*** Chi2
14
results + discussion
TR vs. RT+TT (fused)
• fused responses (presence of a #CC)
• close to 100% of percepts are TR
• RT percepts are at blunder level for real
words (2,7%), a little more than that for
nonce words (5%).
thus there is a phonological effect
• both in nonce and real words
• ==> the TR-only grammar mandates a TR percept.
experiment #1: French
171
435
9 12
0
50
100
150
200
250
300
350
400
450
500
nonce real
nbresponses
TR vs. RT+TT (fused)
TR RT+TT
perception of TR vs. RT+TT should be chance
• since speakers have no evidence for going either way
• the difference between TR and RT+TT percepts is of
course significant, both in nonce and real words.
***
***
15
discussion
• what's new (wrt Cutting 1975 and following)
• there is a lexical effect
• which is eliminated in nonce words
• where only phonology can be responsible for the enforcement of the
almost categorical TR percept.
• Cutting (1975) was right: fusion is phonological (channelled by
phonology). But he couldn't know he was since he didn't have the
lexical effect on his radar.
experiment #1: French Cutting, James E. 1975. Aspects of
Phonological Fusion. Journal of
Experimental Psychology 104: 105-120.
16
Experiment #2
real vs. nonce words in
Czech
17
plan
recall our plan
step 1 completed
• in a TR-only language like French, grammar enforces a TR percept
now step 2
• in an anything-goes language like Czech, it is predicted that speakers perceive TR
vs. RT at chance level.
• grammar does not interfere: it is equally happy with TR and RT percepts.
18
experiment #2: Czech
experimental settings
• identical wrt to experiment #1 on French
• alignV
• 50 participants
• 66 stimuli
• 30 real words
• 36 nonce words
• participants are exposed to the 66 stimuli once. Thus 1500 (real) + 1800 (nonce)
= 3300 trials
• randomization of the order in which stimuli are presented
• randomization of left - right ear (to level out putative one-sided hearing deficiency)
19
experimental settings
• identical wrt to experiment #1 on French
• source words don't exist, targets of nonce words don't exist.
• free choice
• participants hear the doctored stereo file three times in a row through a
headset (750ms interval)
• and are then asked to say what they have heard (keyboard input)
• one difference:
• in French, targets of existing words can only be TR
• in Czech they could be TR or RT
• in order to create identical conditions, in the Czech experiment all targets of
existing words are TR.
experiment #2: Czech
20
results
1417
171
2 7 3
1149
435
3 9 4
0
500
1000
1500
CV
TR
RT
TT
other
nbresponses
nonce vs. real
nonce real
• what springs to the eye when comparing
with French is the presence of significant
RT responses (which were about absent
in French) in both nonce and real words.
• the bars of TR and RT are about the
same size, both for nonce and real.
1248
256 275
16 5
985
290
215
7 3
0
500
1000
1500
CV TR RT TT other
nbresponses
nonce vs. real
(simultaneous)
nonce real
French recalledexperiment #2: Czech
21
results
• there is a lexical effect in Czech
• p=.023 (Chi2)
• thus from now on (next slide and in experiment
#3), like in French, we eliminate the lexical
effect by using only nonce words.
30,5
34,2
0,0
20,0
40,0
60,0
80,0
100,0
nonce real
%ofvalidtrials
fusion rate
(simultaneous)
*
experiment #2: Czech
22
results
main result
• in nonce words, the difference between
TR and RT+TT percepts is nonsignificant:
p = .46.
• TR and RT+TT are statistically
indistinguishable
RM Anova
F(1,49) = .555; p = .46; η2 = .005
Estimated Marginal Means
95% Confidence
Interval
Mean SE Lower Upper
TR 5.12 0.537 4.04 6.2
RT+TT 5.82 0.842 4.13 7.51
experiment #2: Czech nonce words
23
discussion
• as predicted,
• there is no phonological effect in Czech
• whether speakers pick TR or RT+TT is chance.
• when lacking evidence for going either way, Czech grammar does not interfere:
it is just as happy to have TR perceived as it is to have RT/TT perceived.
• dichotic perception of initial CC clusters
• depends on the grammar of speakers
• TR-only languages mandate TR percepts
• anything-goes languages do not care for what is perceived: in absence of
evidence, the selection is chance.
experiment #2: Czech
24
why is that interesting?
• we now have evidence that an anything-goes grammar is contributing no bias
when speakers lack evidence for choosing between
• #TR
• #RT / #TT
• our ultimate goal is to see whether this is also the case when speakers need to
choose between
• existing #RTs
• non-existing #RTs
the initial CV and gaps in #RT inventories
this is the purpose of experiment #3
25
Gaps in
#RT inventories
26
TR-only languages
• instantiate ALL logically possible T+Liquid (TL) clusters (except #tl, #dl)
anything-goes languages
• #TL
• also possess all logically possible TL clusters
• #RT
• may also instantiate ALL logically possible #RTs: Moroccan Arabic, Berber
• or may exhibit only a subset thereof: all Slavic languages in point, (Modern)
Greek
TR-only vs. anything-goes languages
a closer look at anything-goes languages
27
#RT in Slavic
#RT in Slavic:
exhaustive record
Corpus:
http://sites.unice.fr/scheer/tobweb/classes.htm#sldata
Scheer, Tobias 2007. On the Status of Word-Initial
Clusters in Slavic (And Elsewhere). Annual Workshop
on Formal Approaches to Slavic Linguistics. The
Toronto Meeting 2006, edited by Richard Compton,
Magdalena Goledzinowska & Ulyana Savchenko,
346-364. Ann Arbor: Michigan Slavic Publications.
Scheer, Tobias 2012. Direct Interface and One-Channel
Translation. A Non-Diacritic Theory of the
Morphosyntax-Phonology Interface. Vol.2 of A Lateral
Theory of phonology. Berlin: de Gruyter.
28
Polish selection:
20 out of 126 possible #RTs
16%
Czech selection:
28 out of 108 possible #RTs
26%
within Slavic, the two
languages with the biggest
number of #RTs still only
instantiate a small subset of
possible #RTs.
#RT in Slavic
29
#RT inventories: are the gaps
• systematic, i.e. enforced by grammar,
• or accidental ?
#RT in Slavic
like in other cases where a distribution is non-arbitrary,
• the zero hypothesis is that the gaps are systematic, i.e. that
• grammar actively rules out non-existing #RTs
• just like it actively rules our #RT in TR-only languages.
30
The initial CV
predicts that gaps are accidental
31
Interface without diacritics
carriers of morpho-syntactic
information in phonology
• are not diacritics (#, ω, φ etc.)
Lowenstamm, Jean 1999. The beginning of the word.
Phonologica 1996, edited by John Rennison & Klaus
Kühnhammer, 153-166. La Hague: Holland Academic
Graphics. WEB.
Scheer, Tobias 2009. External sandhi: what the initial CV is
initial of. Studi e Saggi Linguistici 47: 43-82. WEB.
Scheer, Tobias 2012. Direct Interface and One-Channel
Translation. A Non-Diacritic Theory of the MorphosyntaxPhonology
Interface. Vol.2 of A Lateral Theory of phonology.
Berlin: de Gruyter.
Scheer, Tobias 2014. The initial CV: Herald of a non-diacritic
interface theory. The Form of Structure, the Structure of
Form. Essays in Honor of Jean Lowenstamm, edited by
Sabrina Bendjaballah, Noam Faust, Mohamed Lahrouchi &
Nicola Lampitelli, 315-330. Amsterdam: Benjamins.
• but rather truly phonological items, i.e.
pieces of the proprietary phonological
vocabulary
• they bear on phonological computation
by their mere presence, like all other
phonological items.
• #, ω and the like only produce an effect
when they are called on by some rule or
constraint.
32
Lowenstamm, Jean 1999. The
beginning of the word. Phonologica
1996, edited by John Rennison &
Klaus Kühnhammer, 153-166. La
Hague: Holland Academic Graphics.
WEB.
the initial CV
• the phonological identity of the
beginning of the word is extra
syllabic space
# CV
Interface without diacritics
Scheer, Tobias 2012. Direct Interface
and One-Channel Translation. A NonDiacritic
Theory of the MorphosyntaxPhonology
Interface. Vol.2 of A Lateral
Theory of phonology. Berlin: de
Gruyter.
• depending on theoretical inclincation,
• a mora
• an x-slot
• an empty CV unit
33
cross-linguistically stable effects of the beginning of the word
1. word-initial clusters
• in some languages, initial clusters are restricted to #TR.
• in others they have the same distribution as internal clusters.
• but there is no language where they are restricted to #RT (#TT, #RR).
Scheer (2004, 2009, 2012)
2. strength of word-initial consonants
• in some languages, word-initial consonants are especially strong.
• in others, they do not have any peculiar behaviour regarding strength.
• but there is no language where they are especially weak.
3. deletion of the first vowel of the word
• in some languages, the first vowel of words is unable to alternate with zero.
• in others it does not show any peculiar behaviour when compared to vowels in
other positions.
• But there is no language where non-initial vowels are unable to alternate with
zero, while initial vowels do.
Interface without diacritics
34
the beginning of the word
• has stable effects across languages.
• for each of the three phenomena mentioned, there appear to be two and only two
parametric options.
• whatever the phonological identity of the beginning of the word, it must be
responsible for this parametric pattern.
• phonological identities that allow for anything and its reverse to happen are
inaccurate.
Scheer (2004, 2009, 2012)Interface without diacritics
35
diacritics such as #, ω, φ
• are arbitrarily chosen (# is a typewriting symbol) and
interchangeable
• have no intrinsic properties (phonological or other)
• therefore can produce any effect and its reverse
• # and ω could equally well mandate that
• #CC be restricted to TR attested
• #CC be restricted to RT not on record
• language does not work like that
• effects of the beginning of the word are not random
• they are cross-linguistically stable
• # and ω could equally well mandate that
• the beginning of the word is strong attested
• the beginning of the word is weak not on record
Scheer, Tobias 2009. External sandhi:
what the initial CV is initial of. Studi
e Saggi Linguistici 47: 43-82. WEB.
Interface without diacritics
Scheer, Tobias 2012. Direct Interface
and One-Channel Translation. A NonDiacritic
Theory of the MorphosyntaxPhonology
Interface. Vol.2 of A
Lateral Theory of phonology. Berlin:
de Gruyter.
36
workings of the initial CV
let us look at the
deletion of the first vowel of the word
language type A
• the first vowel of a word may alternate with zero
• Slavic: Russian, Czech, Polish
• Cz lev - lv-a "lion Nsg, Gpl"
language type B
• the first vowel of a word may not alternate with zero
• German, Belarusian
• German: all schwas can alternate with zero, except if they are initial
• g[ə]halten, *g'halten "held"
• Belarusian
• lew - i-lw-a "lion Nsg, Gsg"
Scheer, Tobias 2009. External sandhi:
what the initial CV is initial of. Studi
e Saggi Linguistici 47: 43-82. WEB.
Scheer, Tobias 2012. Direct Interface
and One-Channel Translation. A NonDiacritic
Theory of the MorphosyntaxPhonology
Interface. Vol.2 of A
Lateral Theory of phonology. Berlin:
de Gruyter.
37
C V - C V C V
| | |
C C V
deletion of the first vowel of the word
settings
• the nucleus of the initial CV must be governed
• therefore the first nucleus of the root must be contentful: two
empty nuclei in a row are ill-formed
• the box is well-formed as it stands
|
V
gvt
• the box is ill-formed as it stands
• inserting a V makes it
well-formed
• deletion will make the string ill-formed
C V - C V C V
| | |
C C V
V
• this is why (in some languages) initial vowels (rather than other
vowels) may not alternate with zero.
• the initial site is protected by the initial CV.
workings of the initial CV
38
three birds with one stone
Scheer (2009, 2012, 2014), Ségéral & Scheer (2008)
the initial CV and the lateral relations of Strict CV
• predict all three cross-linguistically stable effects
initial CV present initial CV absent
word-initial clusters #TR only #CCs unrestricted
word-initial consonants strong non-strong
first vowel of the word may not
alternate with zero
may
alternate with zero
workings of the initial CV
39
workings of the initial CV
let us now look at
how the initial CV regulates initial clusters
language type A: presence of the initial CV
C1 V1 - C2 V2 C3 V3
| | |
R T V
• RT-initial word
• two empty nuclei in a row are ill-formed: V3 cannot
govern two empty nuclei at a time.
• *#RT (same for #TT, #RR)
• TR-initial word
• there are also two empty nuclei in a row, but V2 is enclosed in a TR cluster,
which (unlike other clusters) is solidary.
• the solidarity of branching onsets is usually expressed by the fact that the onset
branches. The Strict CV version of that is a relationship between T and R
(Infrasegmental Government): <==
• #TR is well-formed because all empty nuclei are taken care of:
• the one enclosed in the TR by <==
• the initial empty nucleus by government from V3
Gvt
C1 V1 - C2 V2 C3 V3
| | |
T <= R V
40
workings of the initial CV
language type B: absence of the initial CV
Gvt
C2 V2 C3 V3
| | |
T R V
R T V
T T V
R R V
• RT-initial word
• there is only one empty nucleus to be taken care of, V2:
it is goverend by V3.
• TR-initial word
• same situation, also for TT and RR.
• the sonority configuration of the #CC does not matter:
• anything goes because V2 will always be governed.
41
workings of the initial CV
in sum
• the initial CV
• puts additional burden on initial clusters
• only solidary TR can survive
• typology
• TR-only languages: initial CV present
• anything-goes languages: initial CV absent
prediction
• there is no third possibility: the initial CV can only be present or absent
• there are only two types of languages in the world: those that have the initial CV, and
those that don't. The typology of initial clusters is strictly binary.
thus in anything-goes languages, literally anything goes
• grammar does not impose any restrictions
• gaps in initial clusters are accidental.
42
Gaps in #RT inventories are
accidental:
independent arguments
43
argument #1
• anarchic distribution
• neither occurring or non-occurring #RTs are natural
classes in any sense
• people have tried a lot to find out what each set has in
common and what opposes it to the other set – to no
avail.
• especially in Poland this is a national sport.
independent arguments Cyran, Eugeniusz & Edmund Gussmann
1999. Consonant clusters and governing
relations: Polish initial consonant
sequences. The syllable, Views and
Facts, edited by Harry van der Hulst &
Nancy Ritter, 219-248. Berlin, New York:
de Gruyter.
Scheer, Tobias 2007. On the Status of
Word-Initial Clusters in Slavic (And
Elsewhere). Annual Workshop on Formal
Approaches to Slavic Linguistics. The
Toronto Meeting 2006, edited by Richard
Compton, Magdalena Goledzinowska &
Ulyana Savchenko, 346-364. Ann Arbor:
Michigan Slavic Publications. WEB.
44
argument #2
• modern anything-goes languages are merely CS
minus yers (plus eventual repairs):
• where C1 and C2 have random distribution in CS.
• ==> diachronically, non-TRs are lexical accident
independent arguments Cyran, Eugeniusz & Edmund Gussmann
1999. Consonant clusters and governing
relations: Polish initial consonant
sequences. The syllable, Views and
Facts, edited by Harry van der Hulst &
Nancy Ritter, 219-248. Berlin, New York:
de Gruyter.
why do Slavic languages have #mT, but not #nt?
• if anything, the unmarked #nT should occur.
• because there happened to be no lexical item #n-yer-TV… in CS.
#C1-yer-C2V… > #C1C2V…
Scheer, Tobias 2007. On the Status of
Word-Initial Clusters in Slavic (And
Elsewhere). Annual Workshop on Formal
Approaches to Slavic Linguistics. The
Toronto Meeting 2006, edited by Richard
Compton, Magdalena Goledzinowska &
Ulyana Savchenko, 346-364. Ann Arbor:
Michigan Slavic Publications. WEB.
45
Experiment #3
Czech
existing vs. non-existing RTs
46
experiment #3: Czech existing vs. non-existing #RTs
goal
• in order to see whether the prediction that gaps in #RT inventories are accidental,
• dichotic listening may be used
predictions
• if gaps are accidental, speakers should select existing vs. non-existing #RTs at
chance level if they are given no evidence going either way.
• if gaps are systematic, grammar should introduce a bias in favour of existing #RTs.
47
experiment #3: Czech existing vs. non-existing #RTs
how to test that
• probe an existing cluster: rt
• play simultaneous
• týf (L)
• rýf (R)
• ask participants what they have heard
• display three choices
• TR trýf
• RT rtýf
• none of those
• then test a non-existing cluster: rb
• do the same procedure
prediction
• the number of #RT percepts will be
the same for existing rt and nonexisting
rb.
48
experimental settings
• alignV
• 50 participants
• 54 stimuli
• only nonce words
• source words don't exist
• participants are exposed to the 54 stimuli once. Thus 2700 trials.
• randomization of the order in which stimuli are presented
• randomization of left - right ear (to level out putative one-sided hearing deficiency)
• randomization of left / right location of TR / RT competitor on the screen
experiment #3: Czech existing vs. non-existing #RTs
49
clusters tested
• existing #RTs: rt, rd, lb
• non-existing #RTs: rb, rk
stimuli design
• only Liquid-Stop clusters are tested
• source words: C + long V + C lbéch, rdůj, rkýš, rbách, etc.
experiment #3: Czech existing vs. non-existing #RTs
50
main test
experiment #3: Czech
existing (rt, rd, lb) vs.
non-existing (rb, lk) RTs
descriptive statistics
• fusion rate
• % (of fused)
discussion
• fusion rate is high due to forced choice.
• RT percepts are high
• it is unclear why RT percepts are overwhelming: more
than 80%.
• we are far away from the 50-50 RT-TR of experiment #2.
1071
163 116
1056
209
85
0
500
1000
1500
RT
nbtrials
nb of trials
existing RTs (rt, rd, lb)
non-existing RTs (rb, rk)
TR neither
86,8
13,2
83,5
16,6
0
20
40
60
80
100
RT
%(fused)
% (fused)
existing RTs (rt, rd, lb)
non-existing RTs (rb, rk)
TR
• nb of trials
91,5 91,1
0
20
40
60
80
100
existing RTs
(rt, rd, lb)
non-existing
RTs (rb, rk)
%ofalltrials
fusion rate
51
main test
experiment #3: Czech
the difference between existing (rt, rd, lb)
and non-existing (rb, lk) RTs
• is non significant: p = .532 (RM
Anova)
• = the two sets are statistically
indistinguishable
Within Subjects Effects
Sum of
Squares
df
Mean
Square
F p η²
test 2.25 1 2.25 0.396 0.532 0.002
Residual 278.25 49 5.68
Note. Type 3 Sums of Squares
Estimated Marginal Means
95% Confidence
Interval
test Mean SE Lower Upper
existing RTs (rt, rd, lb) 21.4 0.505 20.4 22.4
non-existing RTs (rb, rk) 21.1 0.585 19.9 22.3
52
Conclusion
53
conclusion
evidence from dichotic listening suggests that
grammar
• in TR-only languages: actively mandates the perception of
initial clusters
• in anything-goes languages: does not mandate or forbid
anything, is happy with #TR and #RT alike.
#RT clusters in anything-goes languages
• are not restricted by grammar in any way
• literally anything goes (as far as grammar is concerned)
• gaps in RT inventories are accidental
• the binary typology of initial clusters predicted by the initial
CV is borne out
French (exp.#1) vs.
Czech (exp.#2)
Czech (exp.#3)
• these are the results for one anything-goes language, Czech
• other anything-goes languages will need to be tested…
54
The End
55
Appendix
56
1. Results of T/R lag
57
results
• influence of T/R lag
• 0 = simultaneous T and R
• T100 = T is played 100ms before R
• R100 = R is played 100ms before T
• results are as predicted
• advantaging T in T100 produces more TR percepts than when source words are
administered simultaneously (zero).
• advantaging R in R100 produces less TR percepts than when source words
are administered simultaneously (zero).
experiment #1: French
• differences
• within nonce words are significant, except 0 vs.
T100 (T100 can't help TR to get any higher)
• within real words are all significant.
171
245
102
435
568
272
0
100
200
300
400
500
600
0 T100 R100
nbTRresponses
TR percepts according to T/R lag
nonce real
58
results
influence of T/R lag
• 0 = simultaneous T and R
• T100 = T is played 100ms before R
• R100 = R is played 100ms before T
effect of T100
• as predicted: advantaging T produces more TR percepts than when source words
are administered simultaneously (zero).
effect of R100
• not as predicted: should advantage R, i.e. produce less TR percepts than when
source words are administered simultaneously (zero). But it yields more TR percepts.
differences
• within nonce words are significant, except 0 vs.
R100 (R100 can't help to get more RTs)
• within real words are significant.
experiment #2: Czech
14
46
2219
60
24
0
10
20
30
40
50
60
70
0 T100 R100
%oftotalnonce/real
TR percepts according to T/R lag
nonce real
59
2. Lexical cluster effect
60
possible lexical effect of clusters
lexical effect
• of word: when a word is present in the lexicon, a bias in its favour is introduced
when compared to a word that is absent form the lexicon.
experiment #3: Czech existing vs. non-existing #RTs
controlling for the lexical cluster effect
• the word effect can be eliminated by using nonce words, but we cannot eliminate the
clusters from our stimuli.
• the lexical cluster effect cannot be escaped, but it can be controlled for:
• by the lexical frequency of clusters
• of cluster: when a cluster is present in the lexicon, a bias in its favour is introduced
when compared to a cluster that is absent form the lexicon.
61
experiment #3: Czech existing vs. non-existing #RTs
possible lexical cluster effects
• in exp. #3, on each trial, an #RT cluster competes with a #TR cluster
• RT cluster effect
• bias in favour of high frequency #RTs (as compared to low frequency #RTs)
• low frq #lb will get less #RT responses than high frq #rt
existing #RTs
• may be subjected to both a lexical RT and a lexical TR effect.
non-existing #RTs
• have no lexical RT effect
• but may have a lexical TR effect.
• TR cluster effect
• bias in favour of high frequency #TRs (as compared to low frequency #TRs)
• when comparing non-existing rp and rb, their respective TRs have high (pr) or low
(br) frq. Thus upon rp vs. pr, high frq pr may produce more TR percepts than low
frq br upon the competition of rb vs. br.
62
possible lexical cluster effect
experiment #3: Czech existing vs. non-existing #RTs
lexical frequencies
(Czech National Corpus)
RT RT frq competing
TR
TR frq
(token, in
thousand)
(token, in
million)
existing rt 111.3 tr 23.3
rd 12.1 dr 16.4
lb 3.3 bl 6.7
non-existing rp pr 180.5
rk kr 23.0
rb br 11.7
63
results
experiment #3: Czech existing vs. non-existing #RTs
is there a TR cluster effect?
• yes
• comparison of the three non-existing RTs
rp, rk, rb
• measure: RT responses
• comparison: RM Anova
RT RT frq competing
TR
TR frq
(token, in
thousand)
(token, in
million)
existing rt 111.3 tr 23.3
rd 12.1 dr 16.4
lb 3.3 bl 6.7
non-existing rp pr 180.5
rk kr 23.0
rb br 11.7
0
50
100
150
200
rp rb rk
lexicalfrq(token)
TR cluster frequency effect
n.s.
***
**
discussion
• high TR frq
• increases TR percepts
• = lowers RT percepts
64
results
experiment #3: Czech existing vs. non-existing #RTs
is there an RT cluster effect?
• probably
• comparison of the three existing RTs
rt, rd, lb
• measure: RT responses
• comparison: RM Anova
RT RT frq competing
TR
TR frq
(token, in
thousand)
(token, in
million)
existing rt 111.3 tr 23.3
rd 12.1 dr 16.4
lb 3.3 bl 6.7
non-existing rp pr 180.5
rk kr 23.0
rb br 11.7
0
20
40
60
80
100
120
rt rd lb
lexicalfrq(token)
RT cluster frequency effect
**
n.s.
***
discussion
• the two upper frq RTs produce significantly more
RT percepts than the low frq RT.
• this cannot be due to TR frq: if anything, high
TR frq should increase TR percepts = lower RT
percepts.
• it is unclear why the difference between rt and
rd is non significant.
65
minimizing the lexical cluster effect
experiment #3: Czech existing vs. non-existing #RTs
TR cluster effect
• our goal is to minimize lexical cluster effects, so that the purely phonological effect
emerges.
• non-existing RTs
• the best item is the one with the lowest TR frequency, since it is closest to
eliminating the documented TR cluster effect.
• rb, rk
• low TR frq (11.7, 23.0): their bias in favour of TR will be small.
• we take them in = use them in our main test.
• rp
• high TR frq (180.5): will introduce a strong bias in favour if TR.
• we don't take it in = don't use it in our main test.
66
main test
experiment #3: Czech existing vs. non-existing #RTs
• we compare existing vs. non-existing RTs
• existing RTs
• rt, rd, lb
• = all three clusters
• knowing that there is probably an RT cluster effect that will produce a bias.
• non-existing RTs
• rb, rk
• but not the high TR frq item rp