Poster Pre­sen­ta­tions at SFN

There will be two poster pre­sen­ta­tions at SFN in Wash­ing­ton, DC., on the top­ic of audi­to­ry pre­dic­tions in speech per­cep­tion. The first poster, authored by Alexan­dra Ben­dix­en, Math­ias Scharinger, and Jonas Obleser, sum­ma­rizes as follows:

Speech sig­nals are often com­pro­mised by dis­rup­tions orig­i­nat­ing from exter­nal (e.g., mask­ing noise) or inter­nal (e.g., slug­gish artic­u­la­tion) sources. Speech com­pre­hen­sion thus entails detect­ing and replac­ing miss­ing infor­ma­tion based on pre­dic­tive and restora­tive mech­a­nisms. The nature of the under­ly­ing neur­al mech­a­nisms is not yet well under­stood. In the present study, we inves­ti­gat­ed the detec­tion of miss­ing infor­ma­tion by occa­sion­al­ly omit­ting the final con­so­nants of the Ger­man words “Lachs” (salmon) or “Latz” (bib), result­ing in the syl­la­ble “La” (no seman­tic mean­ing). In three dif­fer­ent con­di­tions, stim­u­lus pre­sen­ta­tion was set up so that sub­jects expect­ed only the word “Lachs” (con­di­tion 1), only the word “Latz” (con­di­tion 2), or the words “Lachs” or “Latz” with equal prob­a­bil­i­ty (con­di­tion 3). Thus essen­tial­ly, the final seg­ment was pre­dictable in con­di­tions 1 and 2, but unpre­dictable in con­di­tion 3. Stim­uli were pre­sent­ed out­side the focus of atten­tion while sub­jects were watch­ing a silent video. Brain respons­es were mea­sured with mul­ti-chan­nel elec­troen­cephalo­gram (EEG) record­ings. In all con­di­tions, an omis­sion response was elicit­ed from 125 to 165 ms after the expect­ed onset of the final seg­ment. The omis­sion response shared char­ac­ter­is­tics of the omis­sion mis­match neg­a­tiv­i­ty (MMN) with gen­er­a­tors in audi­to­ry cor­ti­cal areas. Crit­i­cal­ly, the omis­sion response was enhanced in ampli­tude in the two pre­dictable con­di­tions (1, 2) com­pared to the unpre­dictable con­di­tion (3). Vio­lat­ing a strong pre­dic­tion thus elicit­ed a more pro­nounced omis­sion response. Con­sis­tent with a pre­dic­tive cod­ing account, the pro­cess­ing of miss­ing lin­guis­tic infor­ma­tion appears to be mod­u­lat­ed by pre­dic­tive context.

The sec­ond poster looks at sim­i­lar mate­r­i­al, but con­trasts coro­nal [t] with dor­sal [k], yield­ing inter­est­ing asym­me­tries in MMN responses:

Research in audi­to­ry neu­ro­science has lead to a bet­ter under­stand­ing of the neur­al bases of speech per­cep­tion, but the rep­re­sen­ta­tion­al nature of speech sounds with­in words is still a mat­ter of debate. Elec­tro­phys­i­o­log­i­cal research on sin­gle speech sounds pro­vid­ed evi­dence for abstract rep­re­sen­ta­tion­al units that com­prise infor­ma­tion about both acoustic struc­ture and artic­u­la­tor con­fig­u­ra­tion (Phillips et al., 2000), there­by refer­ring to phono­log­i­cal cat­e­gories. Here, we test the pro­cess­ing of word-final con­so­nants dif­fer­ing in their place of artic­u­la­tion (coro­nal [ts] vs. dor­sal [ks]) and acoustic struc­ture, as seen in the time-vary­ing for­mant (res­o­nance) fre­quen­cies. The respec­tive con­so­nants dis­tin­guish between the Ger­man nouns Latz (bib) and Lachs (salmon), record­ed from a female native speak­er. Ini­tial con­so­nant-vow­el sequences were aver­aged across the two nouns in order to avoid coar­tic­u­la­to­ry cues before the release of the con­so­nants. Latz and Lachs served as stan­dard and deviant in a pas­sive odd­ball par­a­digm, while the EEG from 20 par­tic­i­pants was record­ed. The change from stan­dard [ts] to deviant [ks] and vice ver­sa was accom­pa­nied by a dis­cernible Mis­match Neg­a­tiv­i­ty (MMN) response (Näätä­nen et al., 2007). This response showed an intrigu­ing asym­me­try, as seen in a main effect con­di­tion (deviant Latz vs. deviant Lachs, F(1,1920) = 291.84, p < 0.001) of an omnibus mixed-effect mod­el. Cru­cial­ly, the MMN for the deviant Latz was on aver­age more neg­a­tive than the MMN for the deviant Lachs from 135 to 185 ms post deviance onset (p < 0.001). We inter­pret these find­ings as reflect­ing a dif­fer­ence in phono­log­i­cal speci­fici­ty: Fol­low­ing Eulitz and Lahiri, 2004, we assume coro­nal seg­ments ([ts]) to have less spe­cif­ic (‘fea­t­u­ral­ly under­spec­i­fied’) rep­re­sen­ta­tions than dor­sal seg­ments ([ks]). While in stan­dard posi­tion, Lachs acti­vat­ed a mem­o­ry trace with a more spe­cif­ic final con­so­nant for which the deviant pro­vid­ed a stronger mis­match than vice ver­sa, i.e. when Latz acti­vat­ed a mem­o­ry trace with a less spe­cif­ic final con­so­nant. Our results sup­port a mod­el of speech per­cep­tion where sen­so­ry infor­ma­tion is processed in terms of dis­crete units inde­pen­dent of high­er lex­i­cal prop­er­ties, as the asym­me­try can­not be explained by dif­fer­ences in lex­i­cal sur­face fre­quen­cies between Latz and Lachs (both log-fre­quen­cies of 0.69). We can also rule out a fre­quen­cy effect on the seg­men­tal lev­el. Thus, it appears that speech per­cep­tion involves a lev­el of pro­cess­ing where indi­vid­ual seg­men­tal rep­re­sen­ta­tions with­in words are evaluated.