SUMMARY As indicated in the Acknowledgements (pp. xi-xiv), Phonotactics and Its Acquisition, Representation, and Use: An Experimental-Phonological Study originated as a PhD dissertation completed at Utrecht University, Netherlands, under the supervision of René Kager. The study addresses a range of issues on the acquisition and representation of phonotactic knowledge and how such knowledge is put to use in speech processing; it will be of particular interest to phonologists and psycholinguists working in the areas of speech segmentation and lexical acquisition.
Within the first opening paragraphs of Chapter 1, “General Introduction” (pp. 1-16), both the purpose of the proposed study and its rationale are set out. No language allows for a random combination of phonemes within words, and it is well established that co-occurrence (im)possibilities between phonemes are governed by language-specific constraints, i.e. phonotactic constraints. Speakers of natural languages all possess reliable -- albeit mostly implicit -- knowledge of such constraints, hence our ability to intuitively discriminate between plausible and implausible combinations of phonemes in both L1 and L2 (L2 phonotactic knowledge improving with advanced L2 proficiency). Unlike orthographic systems, however, the speech signal is continuous and lacks well-defined cues to word boundaries, so how do listeners nonetheless succeed in the seemingly effortless task of breaking down the acoustic signal into lexical and sub-lexical units to access the words stored in their mental lexicon? If lexical recognition were a prerequisite to speech segmentation, language-learning infants, who lack lexical knowledge at the onset, would logically be unable to initiate speech segmentation. Hence the author’s hypotheses: speech segmentation must precede lexical acquisition; and speakers must rely on cues other than lexical information in order to identify word boundaries when processing speech. What these cues are and how they are used to facilitate speech segmentation and lexical acquisition is precisely the subject matter of the proposed investigation. To this end, the author provides a brief review of literature on the form of phonotactic representations in use as well as on the acquisition and source of phonotactic knowledge, highlighting in particular the need to shed light on the following questions: a) Can complex phonotactic representations such as OCP-Place -- a typologically well-attested constraint within words restricting the co-occurrence of non-adjacent homorganic consonants by extending across intervening vowels -- facilitate speech segmentation and lexical acquisition? b) Is the use of OCP-Place language-specific, and therefore acquired from the input, or due to some innate universal perceptual bias? c) If acquired, is such phonotactic knowledge sourced from the lexicon or continuous speech? d) Is the acoustic signal processed by means of a trough- or chunk-based segmentation strategy, the former requiring listeners to attend to low transitional probabilities, the latter to high transitional probabilities? e) In lexical acquisition, does knowledge of probabilistic phonotactics interact with knowledge of syllable structure, which is known to affect speech processing?
Subsequent chapters, which all form stand-alone studies in their own right, report and comment in detail on both the methodology and results of a series of artificial speech segmentation and nonword recall experiments designed to address the above issues. Chapter 2, “Does the lexicon bootstrap phonotactics, or vice versa?” (pp. 17-47), showcases an experiment drawing on an artificial language containing trisyllabic /s/-vowel-/s/ sequences, which have a low probability of occurrence in Dutch infant-directed continuous speech but are over-represented in the infant-directed lexicon. Data indicate that Dutch L1 infants prefer locating word boundaries inside rather than outside such phonemic clusters. No such segmentation preference, however, was found for /p/-vowel-/p/ sequences, which are over-represented in both continuous speech and the infant-directed lexicon. These results are consistent with a trough-based segmentation strategy, which relies on cues from low probability sequences for inserting word boundaries at dips in phoneme transitional probability, and suggest that segmentation cues are indeed acquired from continuous speech rather than from the lexicon.
Chapter 3, “OCP-Place in Speech Segmentation” (pp. 49-79), demonstrates that the effects of OCP-Place on speech segmentation cannot be said to originate from a universal functional bias against the ability to encode separately two consonants with a shared place of articulation. Indeed, results from four artificial language learning experiments indicate that, unlike OCP-Coronal, OCP-Labial is effectively used by adult native listeners of Dutch as a cue for speech segmentation. The Dutch lexicon is restricted only by the latter constraint, which suggests that the use of OCP-Place as a segmentation cue in speech processing is language-specifically correlated with Dutch phoneme distribution and that this constraint is therefore acquired from the input.
Building on the above findings, Chapter 4, “OCP-Place for Speech Processing: A Bias on Perception or Acquisition? The Case of Mandarin Chinese” (pp. 81-109), provides supporting evidence of knowledge of OCP-Place as a language-specifically acquired constraint. In particular, it shows that, unlike the native Dutch-speaking informants in the previous chapter, Mandarin Chinese native speakers -- whose lexicon is not restricted by OCP-Place -- do not use this constraint as a speech segmentation cue, as would be predicted if OCP were due to some universal functional or cognitive bias on acquisition. Further data from three groups of Mandarin Chinese L2 learners of Dutch with varying language proficiency and geographical learning settings (i.e. advanced L2 learners of Dutch living in the Netherlands, advanced L2 learners of Dutch living in China, beginning L2 learners of Dutch living in the Netherlands) reveal that OCP-Place is only used as a segmentation cue by advanced L2 learners of Dutch who had acquired Dutch in the Netherlands. It follows that the acquisition of OCP-Place as a speech segmentation cue in L2 requires both advanced L2 proficiency and sufficient L2 exposure to native input.
Chapter 5, “Probabilistic Phonotactics in Lexical Acquisition: The Role of Syllable Complexity” (pp. 111-137), reports on two experiments with adult monolingual native speakers of Dutch, designed to assess whether knowledge of markedness constraints on syllables modulates phonotactic probability effects on lexical acquisition. By showing that the facilitory effect of high probability phonotactics is significantly enhanced in the case of short-term memory recognition tasks involving more complex syllable structures, results suggest that probabilistic phonotactics and structural phonotactic knowledge as part of a phonological grammar belong to two individual but interacting knowledge components at the sub-lexical level.
Chapter 6, “Second Language Probabilistic Phonotactics and Syllable Structure in Short-Term Memory Recognition” (pp. 139-165), provides further evidence on the separate storage of the two aforementioned sub-lexical knowledge components by demonstrating that these can be acquired independently of each other. Indeed, results from a probed recognition task using phonotactically legal monosyllabic nonwords manipulated for biphone frequency indicate that Spanish and Japanese L2 learners of Dutch -- whose L1s allow for less complex syllable structure than the target language (i.e. CVC in Japanese and CCVC in Spanish vs CCCVCCC in Dutch) -- experienced facilitation of L2 biphone frequency even in the case of biphone structures unattested in their respective L1s. Such knowledge of n-phone probabilities was thus acquired independently of their syllable context of occurrence.
Finally, Chapter 7, “General Conclusion” (pp. 167-176), summarizes previous chapters with a focus on integrating their respective findings and discussing their overall significance for phonology and developmental psychology. It further addresses a number of issues for future investigation. In particular, the chapter highlights the need to determine whether knowledge of both under- and over-represented phonotactic patterns may, in fact, play different roles in speech processing: “It might be that under-representations are more usable in speech segmentation, whereas in lexical acquisition, a clear facilitation comes from knowledge of over-represented patterns” (p. 173). The author also calls for research into whether only knowledge of under-representations is acquired from continuous speech in L1 and L2, or if over-represented phonotactic patterns are also acquired in similar fashion.
EVALUATION Overall, Boll-Avetisyan’s study is as informed and innovative as it is insightful and thought-provoking. It provides the reader with a deep understanding of how complex abstract phonotactic representations are used in speech segmentation and lexical acquisition, and how these are acquired in both L1 and L2. The work’s scientific impact and significance rest on a strong body of empirical evidence collected through creative experimentation, with robust parameters designed to address original research questions.
As the author notes, despite the vast body of evidence for its role in speech segmentation and lexical acquisition, “the question of how phonotactic knowledge is actually put to use in speech processing has been heavily neglected” by researchers (p. 4). The reasons are twofold: on the one hand “psychologists often disregard the nature and complexity of linguistic representations”, focussing instead on the mind’s inner working and learning mechanisms; on the other hand “linguists tend to neglect the consequences of the fact that linguistic representations need to be used in speech processing”, referring instead to the internal structure and complexity of linguistic systems (p. 4). Boll-Avetisyan’s research effectively bridges this gap.
Boll-Avetisyan’s investigation extends both the scope and depth of the literature on the role of phonotactics in speech segmentation and lexical acquisition by showing that infants as well as monolingual and L2 adult speakers draw on phonotactic knowledge of a more abstract and complex nature than mere n-phone probabilities when segmenting continuous speech. Contrary to predictions made by functionalist theories -- which, as a result of a claimed universal bias on perception, predict that listeners can only process sequences of homorganic consonants as separate elements by drawing on additional information from the lexicon (pp. 171-172) -- , participants used OCP as a cue for parsing an artificial language stripped of any such lexical cues. This finding strongly suggests that knowledge of OCP as used in speech segmentation is language specific and acquired from continuous speech rather than from the lexicon.
Moreover, by showing that infants use their knowledge of under- but not over-represented phoneme pairs in speech segmentation, Boll-Avetisyan’s study provides new, compelling insight into the informativeness role of phonotactic knowledge for the parsing of continuous speech: phoneme pairs occurring more often within words than across word boundaries -- because speech consists of a concatenation of words -- , there are fewer troughs than peaks in transitional probability and so, in short, “it is more useful to attend to relatively few troughs than to the much larger number of peaks in transitional probability when segmenting speech” (p. 172). By further demonstrating that “the facilitory effect of high probability on lexical acquisition increased with increasing syllable complexity and vice-versa” (p. 172), Boll-Avetisyan’s results support the hypothesis that speakers possess two separate but interacting sub-lexical phonotactic knowledge components, one specialising in probabilistic phonotactics and the other in structural phonotactics.
Given the complexity of the research undertaken and the statistical tools used, it is remarkable that the accounts and analyses expounded are always clear, relevant, effective and highly interesting. The author clearly has an outstanding ability to formulate original, theoretically-informed hypotheses based on previous literature and to subsequently put these to the test through careful experimentation. This, in combination with frequent brief summary reports and a text written to the highest academic standard -- if we except a number of typographical and/or grammatical errors/inconsistencies -- , makes the author's scholarship readily available to the intended readership as well as to the wider linguistic community.
ABOUT THE REVIEWER:
Dr Kevin Mendousse holds a PhD in linguistics and is currently a Senior Lecturer in the School of European Languages and Literatures at the University of Auckland, New Zealand, where he teaches undergraduate and postgraduate courses in both French language and linguistics. He is the author of a number of journal articles, conference papers and invited research seminars focusing primarily on distinctive feature theory and markedness theory, as they apply to the (morpho)phonology of French and/or English, as well as on the history of linguistic ideas. His research interests also include a forthcoming book translation of original linguistic research carried out on the Ua Pou dialect of the Marquesan language.