Volume 1, Issue 2 - Special issue: Phonetic detail in the lexicon: Papers from the Eleventh Conference on Laboratory Phonology

This paper explores issues relating to signaling word boundaries from the perspective of Aylett's Smooth Signal Redundancy proposal (Aylett, Stochastic suprasegmentals: Relationships between redundancy, prosodic structure and care of articulation in spontaneous speech, University of Edinburgh, 2000, Aylett and Turk, Language and Speech 47: 31–56, 2004) that language has evolved to spread redundancy, i.e. recognition likelihood, evenly throughout utterances. In Aylett's proposal, information that enables listeners to identify sequences of elements in an utterance (signal redundancy) comes from two sources: a) language redundancy, recognition likelihood based on lexical, syntactic, semantic, pragmatic and other factors, and b) acoustic redundancy, recognition likelihood based on acoustic salience. Smooth signal redundancy is achieved by a complementary relationship between language redundancy and acoustic redundancy that is implemented via prosodic structure. While Aylett and Turk (Language and Speech 47: 31–56, 2004) present the case for prosodic prominence as a lever for modulating the acoustic salience of syllables, the current paper proposes that prosodic constituency also fulfils this function for words. The current paper proposes that the signal redundancy, or recognition likelihood, of words can be manipulated by signaling their boundaries, and that the occurrence and strength of these boundary markers correlates inversely with language redundancy. Prosodic constituency implements the complementary relationship between language redundancy and word boundary salience. Smooth Signal Redundancy provides an integrated explanation for a set of properties relating to prosodic constituent structure.

Over the last three decades, priming and masking experiments and corpus frequency studies have dominated attempts to find ranking in the decomposability of words containing morphological affixes. Here we establish feasibility of using another experimental probe based on audiovisually incongruent speech stimuli. In response to such stimuli, a proportion of participants report percepts that differ in place of articulation from either the audio or the visual signal, typically reporting percept /t/ when receiving audio /p/ dubbed onto visual /k/. We study the systematic variation of this proportion, the McGurk fusion rate, using a small corpus with affixes -ing, -s, -er , and -y/-ie bound to simple English stems. We discuss experimental arrangements designed to ensure that participants are not experiencing confusion due to noisy signals, nor experiencing experimenter effects. Generally we find that embedding in a complex word lowers fusion rate, but the different affixes cause different amounts of lowering. We argue that fusion-rate lowering is a measure of the degree of decomposability, and relate this measure to others. These include measures derived from phonetic stress changes brought about by affixing and measures derived from relative frequencies of stem and affixed derived or inflected words in corpora.

Humans recognize previously heard spoken words better when repetitions of the words involve identical productions than productions by a different speaker. Such findings have been taken as evidence that perceived instances of words or sub-lexical units are stored in a detailed form in memory, and that collections of these memory traces comprise or are linked to mental lexical representations. This study tested a different possibility, that detailed acoustic memory occurs during spoken language processing but does not necessarily correspond to words or other traditionally defined units. Two experiments examined lexical access and recognition memory for continuous speech sequences, extracted from a spoken language corpus, as a function of sequence length and onset phase (with respect to word onset), and speaker. Qualitatively different patterns between word identification and memory performance based on these three variables provide little evidence for a role of the word level of representation in memory for the sequences, and suggest that memory-based processing may more independent of this level than has been assumed.

Listeners learn from their past experience of listening to spoken words, and use this learning to maximise the efficiency of future word recognition. This paper summarises evidence that the facilitatory effects of drawing on past experience are mediated by abstraction, enabling learning to be generalised across new words and new listening situations. Phoneme category retuning, which allows adaptation to speaker-specific articulatory characteristics, is generalised on the basis of relatively brief experience to words previously unheard from that speaker. Abstract knowledge of prosodic regularities is applied to recognition even of novel words for which these regularities were violated. Prosodic word-boundary regularities drive segmentation of speech into words independently of the membership of the lexical candidate set resulting from the segmentation operation. Each of these different cases illustrates how abstraction from past listening experience has contributed to the efficiency of lexical recognition.

When they first begin to talk, children show characteristic consonant errors, which are often described in terms that recall Neogrammarian sound change. For example, a Japanese child's production of the word kimono might be transcribed with an initial postalveolar affricate, as in typical velar-softening sound changes. Broad-stroke reviews of errors list striking commonalities across children acquiring different languages, whereas quantitative studies reveal enormous variability across children, some of which seems related to differences in consonant frequencies across different lexicons. This paper asks whether the appearance of commonalities across children acquiring different languages might be reconciled with the observed variability by referring to the ways in which sound change might affect frequencies in the lexicon. Correlational analyses were used to assess relationships between consonant accuracy in a database of recordings of toddlers acquiring Cantonese, English, Greek, or Japanese and two measures of consonant frequency: one specific to the lexicon being acquired, the other an average frequency calculated for the other three languages. Results showed generally positive trends, although the strength of the trends differed across measures and across languages. Many outliers in plots depicting the relationships suggested historical contingencies that have conspired to make for unexpected paths, much as in biological evolution.

This paper demonstrates that a speaker's judgments of well-formedness for novel forms is based directly on their lexical knowledge by showing individual differences in performance in metalinguistic processing of novel words by 68 monolingual English speakers and 30 Spanish-English bilinguals. Monolingual participants were given a well-formedness judgment task and vocabulary assessment in English. Bilingual participants participated in well-formedness judgment tasks and vocabulary assessments in both English and Spanish. An influence of onset-rime phonotactic probability on well-formedness judgments was demonstrated, replicating and extending previous work to a bilingual population. For the bilinguals, there was no evidence of interference between the two languages in the well-formedness task. Individual differences in well-formedness judgments were examined by looking at relations between well-formedness judgments and vocabulary knowledge. Evidence supporting a connection between lexical knowledge and well-formedness task performance was found in the English data for both monolingual and bilingual participants, but this finding was not replicated in the Spanish data. Participants with a larger vocabulary in English were more accepting of low probability nonwords in English. It appears that those with greater vocabulary knowledge are more likely to have experienced improbable phonological constituents, and may also have a lower threshold for “unacceptable” nonwords, if their threshold is based on a likelihood estimate from their individual lexicon. The lack of a lexical effect for Spanish may reflect the lack of a comparable vocabulary test for Spanish. Overall, it appears that performance on well-formedness judgment tasks for nonwords is shown to be related to emergent generalizations based on the individual's linguistic experience with a language, as reflected in their lexical knowledge.

Russian velar palatalization changes velars into alveopalatals before certain suffixes, including the stem extension -i and the diminutive suffixes -ok and -ek/ik. While velar palatalization always applies before the relevant suffixes in the established lexicon, it often fails with nonce loanwords before -i and -ik but not before -ok or -ek. This is shown to be predicted by the Minimal Generalization Learner (MGL), a model of rule induction and weighting developed by Albright and Hayes (Cognition 90: 119–161, 2003), by a novel version of Network Theory (Bybee, Morphology: A study of the relation between meaning and form, John Benjamins, 1985, Phonology and language use, Cambridge University Press, 2001), which uses competing unconditional product-oriented schemas weighted by type frequency and paradigm uniformity constraints, and by stochastic Optimality Theory with language-specific constraints learned using the Gradual Learning Algorithm (GLA, Boersma, Proceedings of the Institute of Phonetic Sciences of the University of Amsterdam 21: 43–58, 1997). The successful models are shown to predict that a morphophonological rule will fail if the triggering suffix comes to attach to inputs that are not eligible to undergo the rule. This prediction is confirmed in an artificial grammar learning experiment. Under either model, the choice between generalizations or output forms is shown to be stochastic, which requires retrieving known word-forms from the lexicon as wholes, rather than generating them through the grammar. Furthermore, MGL and GLA are shown to succeed only if the suffix and the stem shape are chosen simultaneously, as opposed to the suffix being chosen first and then triggering (or failing to trigger) a stem change. In addition, the GLA is shown to require output-output faithfulness to be ranked above markedness at the beginning of learning (Hayes, Phonological acquisition in Optimality Theory: the early stages, Cambridge University Press, 2004) to account for the present data.

While cooccurrence avoidance of similar or identical consonants within lexical items is well-documented, similar investigations of vowel patterns are lacking. Phonetic patterns suggest that an opposite tendency may exist, and lead to observed asymmetries in the phonological patterning of vowels and consonants. The results of corpus studies of lexical vowel cooccurrence in Croatian and Spanish support this conjecture. Cooccurrence avoidance of similar segments is seen in only one of the languages (Spanish), while the other shows the opposite tendency, with similar vowels preferred in close proximity (Croatian). This finding suggests that vowel cooccurrence can be exempt from restrictions shown for consonants, and in at least some cases, subject to the opposite restriction.

This commentary responds to Abigail Cohn's review of successes and challenges in Laboratory Phonology at the LabPhon10 meeting in 2006 (Cohn, Laboratory Phonology: Past successes and current questions, challenges and goals, Mouton de Gruyter, 2010), and her call for “truly integrated” theoretical models. In my view, the sort of integrated theory that Cohn endorsed relies on the availability of a scientific paradigm to organise our approach to data, methodology and theory. I propose that, from its codification as an approach by the first LabPhon meeting in 1987 (Beckman and Kingston, Introduction, Cambridge University Press, 1990), research in Laboratory Phonology has been (and still is) in the process of acquiring such a paradigm. My commentary argues that the occurrence of categorisation at various levels of granularity, together with evidence for gradient category membership at all levels, are central findings in Laboratory Phonology. These sit in the paradigm alongside the core research question about how the physical, cognitive and social properties of speech sounds are related, and a commitment to using hybrid methodologies from phonetics and phonology, computational linguistics, psycholinguistics and sociolinguistics. My commentary highlights these aspects of the emerging LabPhon paradigm in the papers in this issue by Cutler, Beckman and Edwards, Frisch and Bréa-Spahn, Kapatsinski, and Walter.

The perception of prosodic prominence in spontaneous speech is investigated through an online task of prosody transcription using untrained listeners. Prominence is indexed through a probabilistic prominence score assigned to each word based on the proportion of transcribers who perceived the word as prominent. Correlation and regression analyses between perceived prominence, acoustic measures and measures of a word's information status are conducted to test three hypotheses: (i) prominence perception is signal-driven, influenced by acoustic factors reflecting speakers' productions; (ii) perception is expectation-driven, influenced by the listener's prior experience of word frequency and repetition; (iii) any observed influence of word frequency on perceived prominence is mediated through the acoustic signal. Results show correlates of perceived prominence in acoustic measures, in word log-frequency and in the repetition index of a word, consistent with both signal-driven and expectation-driven hypotheses of prominence perception. But the acoustic correlates of perceived prominence differ somewhat from the correlates of word frequency, suggesting an independent effect of frequency on prominence perception. A speech processing account is offered as a model of signal-driven and expectation-driven effects on prominence perception, where prominence ratings are a function of the ease of lexical processing, as measured through the activation levels of lexical and sub-lexical units.