Janet B. Pierrehumbert

Phonetic diversity, statistical learning, and acquisition of phonology.

In learning to perceive and produce speech, children master complex language-specific patterns. Daunting language-specific variation is found both in the segmental domain and in the domain of prosody and intonation. This article reviews the challenges posed by results in phonetic typology and sociolinguistics for the theory of language acquisition. It argues that categories are initiated bottom-up from statistical modesin use of the phonetic space, and sketches how exemplar theory can be used to model the updating of categories once they are initiated. It also argues that bottom-upinitiation of categories is successful thanks to the perceptionproduction loop operating in the speech community. The behavior of this loop means that the superficial statistical properties of speech available to the infant indirectly reflect the contrastiveness and discriminability of categories in the adult grammar. The article also argues that the developing system is refined using internal feedback from type statistics over the lexicon, once the lexicon is well-developed. The application of type statistics to a system initiated with surface statistics does not cause a fundamental reorganization of the system. Instead, it exploits confluences across levels of representation which characterize human language and make bootstrapping possible.

The timing of prenuclear high accents in English

In English, the alignment of intonation peaks with their syllables exhibits a great deal of contextually governed variation. Understanding this variation is of theoretical interest, and modeling it correctly is important for good quality intonation synthesis. An experimental study of the alignment of prenuclear accent peaks with their associated syllables will be described. Two speakers produced repetitions of names of the form “Ma Lemm,” “Mom LeMann,” “Mamalie Lemonick,” and “Mama Lemonick,” with all combinations of the four first names and three surnames. Segmental durations and the F0 peak location in the first name were measured. Results show that although both speaking rate and prosodic context affect syllable duration, they exert different influences on peak alignment. Specifically, when a syllable is lengthened by a word boundary (e.g., Ma Le Man versus Mama Lemm) or stress clash (e.g., Ma Lemm), the peak falls disproportionately earlier in the vowel. This seems to be related to the syllable‐intern...

The next toolkit

Language is an example of collective behavior, and it is a type of collective behavior for which people are highly adapted. Comparable systems in other creatures in no way approach the level of organization found in human languages. Language provides a canonical example of a complex system: robust, adaptable, and selfassembling. Self-assembly occurs both in the cognitive system (as children bootstrap a complex generative system from limited and varied exposure to its manifestations) and in populations, as people match their language systems to each other, and group themselves into social networks of people who share the same language. Linguistic systems arise within the affordances of physics, biology, and society. Take vowels as an example. The fact that vowels can be well characterized using three formant trajectories (disregarding all resonances of the vocal tract above the third one) derives from several facts. One is that the glottal source spectrum rolls off rapidly, with the result that upper formants are not reliably excited. Another is that the hearing systems of people and all other animals effectively compute a spectral representation of the sound, from which the resonances of the sound source can be estimated. Yet another is that the auditory system is most sensitive right in the region of the first three formants. The confluence of these factors as the phonetic grounding of the vowel map is discussed at more length in Pierrehumbert (2000). Not all combinations of three resonances correspond to possible vowels, of course; the entire threedimensional vowel space is effectively circumscribed by the articulatory range of human beings. Thus, physics and biology define a vowel solid, much as the color vision system defines a color solid. Specific languages preferentially exploit different regions of this solid. For example, the region corresponding to cardinal [u] is not exploited in many dialects of American English. In the American South, the closest correspondant to [u] is much fronter; Labov, Ash, and Boberg (2006) report F2 values in the range of 1400–2000Hz for this region. Fronter than cardinal [u], but not as front as French /y/, the Southern AE [u] corresponds to a particular region in the three-dimensional vowel space, much as the color word ‘‘teal’’ corresponds to a particular region of the color solid which may not have a precise label in some other languages. Now, we can also consider the cluster of people who share any given variant of /u/. In a dialect atlas, such as Labov et al. (2006), the areas in which a variant predominates are marked as regions on a map. Of course this

The perception of fundamental frequency declination

A series of experiments was carried out to investigate how fundamental frequency declination is perceived by speakers of English. Using linear predictor coded speech, nonsense sentences were constructed in which fundamental frequency on the last stressed syllable had been systematically varied. Listeners were asked to judge which stressed syllable was higher in pitch. Their judgments were found to reflect normalization for expected declination; in general, when two stressed syllables sounded equal in pitch, the second was actually lower. The pattern of normalization reflected certain major features of production patterns: A greater correction for declination was made for wide pitch range stimuli than for narrow pitch range stimuli. The slope of expected declination was less for longer stimuli than for shorter ones. Lastly, amplitude was found to have a significant effect on judgments, suggesting that the amplitude downdrift which normally accompanies fundamental frequency declination may have an important role in the perception of phrasing.

Categories of Tonal Alignment in English

This paper reports the results of an inquiry into the question of category versus continuum in intonation. Variants of the English rise-fall-rise pattern were used to study whether tonal alignment is a categorical or gradient distinction. LPC resynthesis was used to construct a set of stimuli in which the alignment of the F0 rise-fall varied in small steps. Subjects heard the stimuli in randomized order and imitated what they heard. The position of the F0 peak relative to the onset of the stressed vowel was measured in each response. Systematic deviations between the peak placement in the stimuli and those in the responses revealed the existence of two categories. We conclude that tonal alignment functions as a binary distinction in English intonation.

Beyond Word Frequency: Bursts, Lulls, and Scaling in the Temporal Distributions of Words

Background Zipfs discovery that word frequency distributions obey a power law established parallels between biological and physical processes, and language, laying the groundwork for a complex systems perspective on human communication. More recent research has also identified scaling regularities in the dynamics underlying the successive occurrences of events, suggesting the possibility of similar findings for language as well. Methodology/Principal Findings By considering frequent words in USENET discussion groups and in disparate databases where the language has different levels of formality, here we show that the distributions of distances between successive occurrences of the same word display bursty deviations from a Poisson process and are well characterized by a stretched exponential (Weibull) scaling. The extent of this deviation depends strongly on semantic type – a measure of the logicality of each word – and less strongly on frequency. We develop a generative model of this behavior that fully determines the dynamics of word usage. Conclusions/Significance Recurrence patterns of words are well described by a stretched exponential distribution of recurrence times, an empirical scaling that cannot be anticipated from Zipfs law. Because the use of words provides a uniquely precise and powerful lens on human thought and activity, our findings also have implications for other overt manifestations of collective human dynamics.

The influence of sexual orientation on vowel production (L)

Vowel production in gay, lesbian, bisexual (GLB), and heterosexual speakers was examined. Differences in the acoustic characteristics of vowels were found as a function of sexual orientation. Lesbian and bisexual women produced less fronted /u/ and /ɑ/ than heterosexual women. Gay men produced a more expanded vowel space than heterosexual men. However, the vowels of GLB speakers were not generally shifted toward vowel patterns typical of the opposite sex. These results are inconsistent with the conjecture that innate biological factors have a broadly feminizing influence on the speech of gay men and a broadly masculinizing influence on the speech of lesbian/bisexual women. They are consistent with the idea that innate biological factors influence GLB speech patterns indirectly by causing selective adoption of certain speech patterns characteristic of the opposite sex.

Tonal Elements and Their Alignment

In English, many different melodies are possible on any given word or phrase. Even a monosyllabic word, such as Anne can be produced with many qualitatively different melodic patterns, as illustrated in Figure 1. This situation provides a contrast with languages such as Mandarin, in which the tonal pattern is an intrinsic part of the lexical representation. In English, the choice of the melody is not entailed by the choice of words, but rather functions independently to convey pragmatic information. Specifically, it conveys information about how the utterance is related to the discourse and to the mutual beliefs which interlocutors build up during the course of the discourse, as discussed in Ward and Hirschberg (1985) and Pierrehumbert and Hirschberg (1990).

Synthesis by rule of english intonation patterns

This papet reports work on synthesizing English F0 contours. One motivation for this work is to improve the naturalness and liveliness of the prosody in speech synthesis systems. However, our main goal is to develop a theory of the dimensions of variation controlling intonation, and of their interaction.

Why phonological constraints are so coarse-grained

Current models of speech perception are divided with regard to the status of ‘‘phonology’’, or general implicit knowledge of the sound patterns of a language. In the TRACE model (McClelland & Elman, 1986) the phonotactic and prosodic constraints of phonology are treated as epiphenomenal from regularities in the lexicon. In contrast, Norris (1994), Vitevich and Luce (1998) and Merge (Norris, McQueen, & Cutler, 2000) respond to a growing body of experimental literature indicating that low-level encoding of the speech signal (the level whose result is passed up to the lexicon for potential word matches) is sensitive to phonotactic and prosodic constraints. Here, I will explore the consequences of the assumption that the architecture of the speech perception system includes a fast phonological preprocessor (hereafter, an FPP) which uses languagespeci!c, but still general, prosodic and phonotactic patterns to chunk the speech stream on its way up to the lexical network. By integrating such information, the FPP imputes possible word boundaries to particular temporal locations in the speech signal. An important question is what types of phonological patterns are candidates for being encoded in the FPP: Can the processing system exploit any statistical regularities whatsoever in the shape of words? Does absolutely any structural description which is logically possible in phonology provide a usable constraint? I will explore the causes and consequences of one observation about this issue, namely that viable constraints are coarse-grained. Although the logical apparatus of phonological theory would make it possible to state extremely !ne-grained constraints (e.g., constraints containing complex and detailed combinations of features and structural positions), the constraints for which we now have linguistic or psycholinguistic evidence are considerably simpler.