The phonetic realization of laryngeal contrasts in languages with two series of stop consonants is probably the most commonly described domain for investigations into the relationship between phonology and phonetics. This fact is likely due to a single phonetic parameter, voice onset time (VOT; Lisker & Abramson, 1964; Cho et al., 2019), which is described in countless studies of a large number of languages. Indeed, Ladd and Schmid (2018, p. 229) refer to VOT as a “runaway success” for describing voice contrasts, a somewhat unusual (and perhaps slightly sarcastic) characterization for an objectively measurable physical phenomenon. The ubiquity of VOT stems from two factors. First, it is a relatively simple parameter to observe and measure on acoustic displays. Second, VOT appears to show rather categorical behavior with regard to two-series laryngeal contrasts. Most frequently, it is claimed that three categories of VOT are found across two-series languages: voicing lead (pre-voicing, or negative VOT), short-lag VOT, and long-lag VOT. Individual systems typically make use of two of these three categories in constructing their voicing contrast. So-called aspiration languages (see e.g., Beckman et al., 2013), which include English and most of Germanic, contrast short-lag lenis stops with long-lag fortis stops, although pre-voicing is known to appear in Swedish (e.g., Beckman et al., 2011) and some varieties of English (e.g., Jacewicz et al., 2009). So called true-voice or voicing languages (e.g., Slavic and Romance languages) contrast pre-voiced and short-lag stops.

The categoricity of VOT has been interpreted as evidence in support of a theoretical perspective that has come to be known as Laryngeal Realism (LR; Iverson & Salmons, 1995; Avery & Idsardi, 2001; Honeybone, 2005; Beckman et al., 2013; Raimy, 2021). Within LR, the strongest claims related to VOT are made by Beckman et al. (2013), for whom pre-voicing is an instantiation of a feature [voice] (cf. [slack] in Avery & Idsardi’s [2001] model), long-lag VOT is assumed to reflect a feature [spread glottis] ([spread]), while short-lag VOT indicates a lack of laryngeal specification. Thus, short-lag stops are attributed to a putative default category that is both phonetically and phonologically inactive. While the earliest works within LR focus on how differences between voicing and aspiration languages are reflected in phonological phenomena such as assimilation, many working today within LR assume that VOT offers a transparent perspective on the phonetics-phonology interface, by which in two-series systems, initial stops reveal a what-you-see-is-what-you-get relationship between phonetics and underlying phonological specification. According to this view, the representation of the contrast based on aspiration (as in, e.g., English) is phonologically distinct from that of contrasts based on voicing (as in, e.g., French).

Underlying the discussion above is a fundamental issue concerning whether and how phonetic evidence may indicate the presence of an active phonological specification. For LR, phonologically active status should give rise to observable phonetic effects, while a lack of such effects should indicate unmarked status. For the most part, we agree with this perspective, yet given the phonetic complexity of voice contrasts, we feel that it is important to search for these effects across a range of laryngeal cues. This, we believe, is a necessary endeavor, since some oft-cited work within LR (e.g., Beckman et al., 2011, 2013) bases its phonological claims primarily on VOT data.

While VOT suggests phonological interpretations that are relatively straightforward, the picture becomes more complicated when we consider additional phonetic correlates of laryngeal contrasts. Available evidence from one such correlate, consonant-induced f0 contours (Cf0; Kirby & Ladd, 2016) reveals largely parallel behavior in voicing languages (e.g., Kirby & Ladd [2016] for French and Italian; Pinget & Quené [2023] for Dutch; Schwartz et al. [2019] for Polish; Gao & Arai [2019] for Tokyo Japanese) and aspiration languages (e.g., Hanson [2009] for English). These experiments have found that, at the onset of the vowel following the obstruent, voiceless consonants induce f0 raising, while voiced consonants have no effect, relative to an f0 baseline produced by nasals. Kirby and Ladd (2018) suggest that this effect is due to active targeting of voicelessness by speakers, even in voicing languages in which, according to LR, unaspirated voiceless stops are phonologically unspecified. In short, comparisons of VOT with other laryngeal cues may give rise to conflicting interpretations of the phonological representations underlying two-series systems: VOT suggests distinct representations for voicing and aspiration languages. Cf0 in the studies cited above suggests parallel representations.

In examining how different phonetic cues suggest conflicting phonological interpretations, it is worth noting the distinct loci of the cues in question. VOT, of course, focuses our attention on stop closure and release. Meanwhile, a number of other correlates are housed on the vowel preceding or following voiced and voiceless obstruents (for discussion, see e.g., Kingston & Diehl, 1994). These include Cf0 mentioned above, as well as F1 onset frequency (Stevens & Klatt, 1974). An additional vowel-based correlate to laryngeal contrasts is voice quality. Broadly speaking, a given laryngeal category may be expected to induce breathier or stiffer phonation relative to other members of a contrast. These differences may be observed in acoustic measures of spectral tilt, such as H1*-H2*, or H1*-A1* (for more details and description of additional measures, see Section 2.4), based on amplitude differences between harmonics in the vocal wave. Breathier phonation is generally associated with a drop-off in amplitude of higher harmonics, yielding higher positive values for various spectral tilt measures (see e.g., Garellek, 2019). Voice quality may also be characterized acoustically with measures of periodicity such as cepstral peak prominence (CPP) and harmonics-to-noise (HNR) ratios (again, see Section 2.4 for details and description of additional measures), which quantify the amount of noise in the vocal wave. Higher values for periodicity measures are generally associated with modal voice relative to both breathy and creaky phonation (see e.g., Garellek, 2019).

Given the role phonation appears to play in evolutionary connections between voice contrasts and those based on tone or register (see e.g., Brunelle & Kirby, 2016; further discussion will follow in Section 3), we expect oppositions traditionally described as voiced vs. voiceless to have effects on the phonatory properties observed on neighboring vowels. For two-series languages, such effects have been documented in English, where textbook descriptions of allophonic effects on voice quality (e.g., Cruttenden, 2001), including aspiration and glottalization, have given rise to experimental (e.g., Kong et al., , 2012; Penney et al., 2018; Sanker, 2019) and corpus studies (Hejná et al., 2021) documenting phonation as a cue to consonant voicing. For example, Kong et al. (2012) include data from H1*-H2* as a function of initial voicing in English and Japanese, and find breathier phonation (higher values) after voiceless consonants. For other two-series systems, particularly those found in voicing languages, we know of very little published data on how underlying voicing may affect voice quality. Documenting such effects can give us further insight into the phonetic complexity of laryngeal contrasts. This insight, in turn, should be valuable for our understanding the nature of the phonological representations underlying the oppositions and the types of diachronic changes they may undergo.

This paper will describe an acoustic study of the two-way laryngeal contrast of phrase-medial word-initial stops in Polish, a language known to have a “standard” voiced-voiceless distinction in pre-vocalic positions. VOT in Polish has been thoroughly described (see, among others, Keating, 1980; Waniek-Klimczak, 2011; Sypiańska, 2013; Malisz & Żygis, 2015; Schwartz, 2022; Wojtkowiak & Schwartz, 2022), so our focus here is on correlates housed on the vowel following the initial stop, including a large set of acoustic measures of phonation, Cf0, as well as F1. While our primary empirical goal is to document the effects of underlying voicing in initial stop consonants on the phonetic realization of the following vowel, our theoretical aim is to consider the question of how laryngeal contrasts are encoded phonologically and predictions that phonological representations may make for the phonetic implementation of the contrast. In particular, we consider two recent proposals (Raimy, 2021; Schwartz, 2022), which would have us expect voicing-induced phonation differences on vowels but differ in the scope of their predictions, as well as their likely interpretation of the data we present.

While a more thorough presentation is provided in Section 3, the predictions of the Raimy (2021) and Schwartz (2022) proposals may be summed up briefly as follows. Raimy’s account is based on the Avery and Idsardi (2001) model, in which phonological features are organized into dimensions, and two additional levels of representation are posited. Dimensions are linked with a phonology-phonetics level of representation through a process called completion. Another process, referred to as enhancement, may alter specifications between the phonology-phonetics level and an additional level of representation: categorical phonetics. Raimy’s approach would have us expect phonetic effects of laryngeal marking induced by both voiced and voiceless obstruents. In the case of voiced obstruents, effects would be due to a feature [slack] that completes the Glottal Tension dimension. Meanwhile, effects of voiceless obstruents would be due to a postulate that they are enhanced at the level of categorical phonetics with a feature [spread] (for more details, see Section 3). With regard to the phonetic realization of these effects, we may glean some predictions from Raimy’s approach, although they are not stated explicitly. One prediction we might expect is a lowered f0 on the vowel following voiced obstruents (Kingston & Diehl, 1994), since the lesser laryngeal tension dictated by [slack] should reduce the frequency of vocal fold vibration. Meanwhile, Raimy’s approach would presumably foresee raised f0 after voiceless obstruents due to [spread] enhancement, in parallel with Cf0 results from languages where [spread] is specified phonologically (e.g., Hanson, 2009). With regard to phonation, it is somewhat more difficult to make inferences from Raimy’s account. In principle, both [slack] on voiced onsets and [spread] on voiceless onsets might be expected to induce some degree of breathiness (higher values for spectral tilt measures, lower values for periodicity measures). However, since [spread] is a function of phonetic enhancement that should be most perceptually robust after stop release, while [slack] is claimed to be a phonological specification responsible for pre-voicing during stop closure, perhaps we can interpret the Raimy proposal as being predictive of more robust breathiness effects after voiceless onsets than voiced onsets.

Schwartz’s proposal, couched within the Onset Prominence representational framework (OP; see Schwartz, 2022, for an OP account of two-series laryngeal phonology), makes slightly different predictions. Onset Prominence is a non-segmental framework in which manner of articulation is specified structurally, derived from the hierarchy of phonetic events inherent to a stop-vowel sequence. This approach encodes the CV transition phase as a structural node within the primary building block of the model1 (see Section 3 for more details). In accordance with the structural properties of obstruents within OP, Schwartz predicts that effects of onset laryngeal specifications in voicing languages should be expected in the early portion of the vowel following the consonant and that these effects should diminish later in the vowel. Raimy’s proposal makes no explicit predictions about the time course of phonetic effects on the following vowel–presumably they would be attributed to implementation. Schwartz’s other prediction is that phonetic effects on the following vowel should be induced by voiceless onsets but not voiced ones. This is because the latter lack laryngeal specification in the OP account (see Section 3), while the former are specified with a feature [fortis]. At the same time, the Schwartz proposal makes no explicit predictions about what the phonetic effects induced by [fortis] at vowel onset might be. A summary of the predictions gleaned from two phonological accounts is given in Table 1.

The differences between these proposals bear on persistent issues in laryngeal phonology, including markedness, the phonetic underpinnings of phonological feature specifications, the status of the segment, and the use of multiple levels of representation in modeling the phonetics-phonology interface. We will revisit these issues in Section 3.

Returning to our experimental study, we state our research questions below.

We will present data from a wide variety of acoustic measures of phonation, as well as Cf0 and F1, to offer what we believe to be the most comprehensive look to date at the phonetics of the word-initial laryngeal contrast in Polish. Before proceeding, we review previous literature on the phonetics of the Polish word-initial voice contrast.

One of the main goals of our study is to provide data bearing on the phonological representation of the voicing contrast in Polish. The phonological literature on two-series laryngeal systems presents a rather complicated picture, in which we may encounter competing claims as to which feature specifications to employ (Iverson & Salmons, 1995; Avery & Idsardi, 2001), whether feature specifications are binary or privative (Rubach, 1996; Wetzels & Mascaró, 2001; Bennett & Rose, 2017), whether one member of the laryngeal contrast is unmarked for laryngeal specification (e.g., Honeybone, 2005), and whether we need multiple levels of representation to explain a given system’s behavior (e.g., Raimy, 2021). These questions focus attention on which (if any) phonetic aspects of laryngeal realization should be encoded in phonological systems, with wider theoretical implications for the relationship between phonetics and phonology.

Debate about these issues has centered on whether voicing languages and aspiration languages show equivalent phonological representations of the laryngeal contrast. In practice, this question has boiled down to whether to treat the VOT typology as a function of phonological specification. One influential early proposal is that of Keating (1984), who posits binary [voice] in phonological representation for both system types and an additional level of phonetic representation to capture the three basic VOT categories. Later work (e.g., Harris, 1994; Iverson & Salmons, 1995; Honeybone, 2005) adopted the feature [spread glottis] (or element |H|) to represent aspiration and long VOT, [voice] for negative VOT and characterize stops with short-lag VOT as unmarked. These proposals, which, as mentioned in Section 1, are often referred to under the heading of Laryngeal Realism (LR; see Avery & Idsardi, 2001; Beckman et al., 2013; and Raimy, 2021), assume that voicing and aspiration systems are phonologically distinct. In addition, the representational differences posited by LR are said to be reflected in distinct patterns of assimilation as a function of the laryngeal system (e.g., regressive voicing is apparently not attested in languages without [voice]). Finally, LR has come to adopt privative feature specifications, in which plain voiceless stops with short-lag VOT are assumed to be phonologically unspecified, regardless of the symbol used to transcribe them. For LR, the typological frequency of plain unaspirated voiceless stops (see e.g., Maddieson, 1984) is attributable to their supposed unmarked status phonologically, which is said to reflect phonetic naturalness and a lack of laryngeal adjustment (e.g., Westbury & Keating, 1986). Under this view, the term “unmarked” has at least three distinct meanings, including “phonetically natural,” “typologically frequent,” and “phonologically unspecified” (see Hume, 2011, for discussion of the many uses of the term “markedness” in phonological theory).

At first glance, the insights of LR would seem to be a welcome development for our understanding of laryngeal phonology and its relationship with phonetics. In short, a binary approach must stipulate VOT and assimilation patterns, while LR appears to explain them. However, one of LR’s fundamental claims, that short-lag stops in voicing languages are unspecified for laryngeal features, has proven controversial (e.g., Rubach, 1996; Wetzels & Mascaró, 2001; Bennett & Rose, 2017). In Polish, there is phonological evidence that voicelessness may spread both regressively and progressively to neighboring segments. Rubach (1996) attributes these facts to a binary [voice] specification in which voiceless obstruents are phonologically active (see also Cyran [2014, 2017], who posits a privative specification for voicelessness in two dialect regions of Polish). In this connection, two recent proposals, described in Raimy (2021) and Schwartz (2022), respectively, seek to reconcile the insights of LR in accounting for VOT patterns with the evidence for active voicelessness in Polish. In what follows, we will compare those two proposals and their compatibility with the data presented in this paper.

Raimy (2021) analyzes Polish laryngeal phonology from the perspective of the model developed by Avery and Idsardi (2001). That system may be considered a representative of LR in that it posits that voicing and aspiration systems are phonologically distinct. It is also realist in that it employs features whose labels imply phonetically explicit descriptions of laryngeal activity. In the Avery and Idsardi model, phonological features are organized into hierarchical dimensions, which house antagonistic feature pairings (e.g., [spread] vs. [constricted]). Voicing languages are phonologically specified with the dimension of Glottal Tension (GT), which is then completed with a feature [slack]. Completion in Polish yields negative VOT in stops. By contrast, aspiration systems are phonologically specified on the dimension of Glottal Width, completed by means of a feature [spread]. For Avery and Idsardi, dimensions are the domain of truly phonological features, which may be underspecified, while completion yields an additional level of representation, referred to by Raimy (p. 86) as the phonology-phonetics level. The Avery and Idsardi model also features an additional mechanism, enhancement, to define relations between phonology-phonetics and an additional level of categorical phonetics. For Raimy, Polish voiceless stops are enhanced with the feature [spread]. Notably, this approach appears to explain both the complex patterns of assimilation in the language (see Rubach, 1996; Cyran, 2014), as well as the fact that Polish voiceless stops may exhibit what Raimy (p. 86), citing data from Keating (1984), refers to as “mid-lag” VOT.

A summary of Raimy’s (2021) proposal for Polish compared with English is given in Table 4, which shows derivational steps from phonology to surface phonetics. Note that Raimy describes English /b d ɡ/ as “voiceless unaspirated” (rather than “lenis”). Additionally, the “Larynx-gt” and “Larynx-gw” markers indicate phonological underspecification, indicated by a use of lower-case letters in “gt” and “gw,” which may be altered at later stages of the derivation.

Schwartz (2022) analyses Polish laryngeal phonology (and its interaction with English in the context of bilingual and L2 speech) from the perspective of the Onset Prominence representational environment (OP; e.g., Schwartz, 2016). A key aspect of OP is that manner of articulation is a structural property, such that stop consonants are represented as the top three layers of a four-layer hierarchical tree structure. The layers of the OP hierarchy are derived from the closure, release, CV transition, and vowel steady-state phases (Closure, Noise, Vocalic Onset, and Vocalic Target nodes within OP representations) inherent in the production of stop-vowel sequence. Crucially, a complete stop-vowel CV unit, rather than skeletal positions for individual segments, represents the fundamental building block of the representations, which unite segmental phonology and prosodic constituents into a single ontological whole, rather than connecting them by means of association lines. OP representations for two-series laryngeal systems are shown in Figure 11.

According to the OP approach (Schwartz, 2022, p. 111), the difference between aspiration languages and voicing languages is based on the hierarchical level of assignment of a single laryngeal feature, labeled [fortis]. When [fortis] is assigned at the Closure level (C), the feature specification occupies the top three nodes, including the Noise node (N), and VOT effects are expected. When it is assigned to VO, the Noise node lacks laryngeal specification, and short-lag VOT is expected. Notably, in both aspiration and voicing systems, the VO node of the structure serves as a docking point for the [fortis] feature,6 thus predicting parallel effects of voicelessness on the phonetics of the following vowel. At the same time, no predictions are made as to the appearance of pre-voicing: Pre-voiced and unvoiced /b d ɡ/ have equivalent representations. In other words, unlike in LR, pre-voicing in OP need not reflect the presence of an underlying phonological feature.

Returning to the data presented in this paper, it may be said that the results are broadly supportive of both of these theoretical approaches. Both proposals predict active phonetic effects resulting from consonant voicelessness, thus challenging the claim, made explicit in some versions of LR (e.g., Beckman et al., 2013), that voiceless stops are phonologically inert. We found effects of voicelessness across a range of voice quality measures. These effects were indicative of breathier (yet still modal) phonation. For Raimy (2021), the breathiness effect would presumably be due to the enhancement mechanism by which a feature [spread] is added at the categorical phonetic level of representation. At the same time, since it is fundamentally a skeleton-based model, Raimy’s proposal makes no explicit predictions as to the time course of the enhancement effects that we observed in our data. Presumably, for Raimy, the timing of the acoustic correlates of [spread] at vowel onset would be an inherent aspect of the implementation of enhancement. For Schwartz, the localization of the effects of voiceless consonants falls out directly from the [fortis] specification on the VO node of structure, which aligns with the CV transition portion of the signal in stop-vowel sequences. At the same time, Schwartz’s proposal refrains from stating just what those [fortis]-like effects might be. In essence, [fortis] is a purely abstract feature specification whose phonetic realization is not explicitly defined. In sum, Raimy is more explicit about what the phonetic effects of voicelessness should be (breathier phonation resulting from [spread]), while Schwartz is more explicit about where those effects may be found (at vowel onset), but both predict that effects should be present.

Nevertheless, the two proposals diverge in a number of ways. The first difference is in their treatment of the voiced series of stops. In Raimy’s proposal (2021), voiced stops in Polish are specified for a phonological feature [slack] on the Glottal Tension dimension, which is responsible for the presence of negative VOT. In the OP proposal, the voiced series is unspecified for any laryngeal feature, so pre-voicing and regressive voicing assimilation are not the result of a phonological specification—they are due to purely phonetic effects (see Cyran, 2014, for similar postulates for southern and western varieties of Polish).7 The data described in this paper do not reveal strong evidence for a [slack] or [voice] feature in Polish beyond VOT, whose perceptual weight in voicing languages appears to be somewhat less robust than in aspiration languages (see Keating et al., 1981; Schwartz, 2022). That is, voiced onsets seem to have only negligible, if any, effects on the acoustics of the following vowel. To explain the lack of effect of voiced consonants on the following vowel, Raimy could perhaps suggest that the [slack] specification on voiced stops does not need enhancement, since it already bears phonological status. This suggestion certainly has a plausible phonetic explanation—gestures responsible for pre-voicing may be timed earlier so as to align with stop closure rather than the following vowel. At the same time, such an interpretation would entail an implicit claim that phonetic enhancement targets only phonologically unspecified segments, but is unnecessary for marked objects. It is unclear whether this claim would stand up to further empirical testing.8

There is also reason to question the conventional wisdom that [voice] (or [slack]) is phonologically active in Polish, despite the presence of pre-voicing and regressive voicing assimilation in the language. Beyond the lesser perceptual weight of VOT mentioned earlier, the status of [voice] may be challenged by findings from L2 and bilingual speech research, in which perceptual equivalence classification (Flege, 1987) gives rise to cross-language phonetic interaction. In a number of studies of interactions between voicing and aspiration systems (for an overview, see Schwartz, 2022), we observe more robust interaction in the voiced series than in the voiceless series. Such findings suggest that Polish-style pre-voiced stops and English-style unvoiced lenis stops are perceptually more similar to each other, with a common phonological representation (unspecified for [voice]/[slack]), than short-lag and long-lag fortis stops. Bilinguals and L2 learners are more likely to keep /p t k/ distinct across languages than /b d g/, suggesting distinct representations for the former, but phonological equivalence for the latter. In our data, the presence of pre-voicing was the only phonetic cue suggesting active [voice]—nothing was found on the vowel-based cues. This fact in itself, of course, does not rule out the feature’s phonological status. However, when considered in light of data from both VOT perception and L2/bilingual speech, our findings are easier to interpret if we assume that voiced stops in Polish are phonologically unspecified.

Independent of the status of [voice]/[slack], a more fundamental difference between the Raimy and Schwartz proposals is that the former posits several levels of representation, while representations in the latter are said to map directly to speech (see Schwartz, 2016, p. 40). Perhaps somewhat ironically, this direct mapping does not imply phonetically precise feature specifications. As pointed out earlier, OP is explicit about [fortis] effects appearing on vowel onset, but says nothing about what those effects might be—the fact that they involve f0 raising and breathier phonation, rather than f0 lowering and stiffer phonation, is essentially arbitrary, even if a phonetic explanation for them is readily available. Raimy’s use of [slack] and [spread] feature specifications, by contrast, makes explicit claims about laryngeal activity. The question that remains is whether such phonetically precise specifications, and increasingly detailed levels of representation are, in fact, desirable.

In this connection, consider Ladd (2011), who argues against a level of categorical phonetic representation assumed in many mainstream approaches. In his discussion, Ladd discusses Cho and Ladefoged’s (1999) VOT data from 18 languages. In Figure 9 (p. 223) of that paper, Cho and Ladefoged group their VOT results into four categories using boxes above the relevant bars. Ladd (2011) questions whether those boxes indeed capture distinct VOT categories that would be identifiable if the corpus included a larger number of languages and/or tokens. Likewise, Raimy’s (2021) proposal apparently assumes a categorical distinction between short-lag VOT (without enhancement) in some true-voice languages and mid-lag VOT in Polish (with enhancement). Based on what we know about Polish VOT (see references cited in Section 1), by which, in addition to mid-lag stops, larger data sets also reveal plenty of truly short-lag stops, Raimy would presumably have to claim that the language has both non-[spread]-enhanced (short-lag) and [spread]-enhanced (mid-lag) stops, begging an explanation of the factors that determine when [spread]-enhancement occurs and when it does not. A more plausible characterization of Polish voiceless stops, it seems to us, is that they are indeed short-lag, but with a VOT distribution that is skewed towards longer values than is found for unvoiced lenis stops. The OP approach is compatible with such a characterization: A [fortis] specification on the VO node with a bare Noise node (see Figure 6), encodes their short-lag status and also explains the results of the vowel-based cues presented in this paper, while the skew towards longer VOT values is a product of statistical noise inherent to phonetic data.

From a wider perspective on two-series systems, including their potential diachronic development into tone and register contrasts, the theoretical difficulties associated with phonetically precise phonological features comes into sharper focus. Brunelle and Kirby (2016, p. 193) observe that register systems, which apparently developed historically from contrasts in voicing, are based on a set of acoustic properties that typically co-occur. Low register, deriving from voiced stops, typically involves lower f0, breathy phonation, lowered F1, and longer VOT. Meanwhile, High register is associated with higher f0, stiffer phonation, raised F1 and shorter VOT.

Considering possible links between register and standard Polish-style voicing systems from which they may have evolved, we can see a number of contradictions. Table 5 provides a selection of Polish laryngeal correlates. An additional column shows how the realization of Polish voiceless stops would correspond to the related cue in register systems. The f0 and F1 effects suggest a link between Polish voiceless stops and High register, while the phonation effects show a link between Polish voiceless stops and Low register. Meanwhile, it is not entirely clear how to characterize the VOT link. Brunelle and Kirby (2016) associated plain stops and short-lag VOT with High register, but do not include mid-lag VOT.

The fact that the phonetic properties associated with the Polish contrast show mismatches with those of register contrasts argues against positing a large amount of phonetic detail in the phonology. The reason for this is that if phonology includes increasing amounts of phonetic detail in successive levels of representation, it is not entirely clear how phonetic divergences between laryngeal contrasts might develop diachronically. In other words, if a phonological system indeed includes mechanisms such as completion and enhancement, encoded with feature specifications based on explicitly defined laryngeal activity, we might expect less dramatic cross-language mismatches of the type shown in Table 5, since the grammar would give explicit phonetic instructions to speakers, diminishing the potential for diachronic instability.9 By contrast, purely abstract laryngeal specifications such as [fortis] leave the analyst with a clearer narrative for describing mismatches and sound change. Changes leading to mismatches, such as those shown in Table 5, may develop due to perceptual reinterpretation (Ohala, 1981; Blevins, 2004) of a rather ambiguous, but quite numerous, set of acoustic correlates.

This article has presented data on how the Polish laryngeal contrast may affect the phonetic realization of vowels following voiced and voiceless stops. Over a range of acoustic measures of voice quality, as well as f0 and vowel height (F1), active effects of consonant voicelessness were observed. In particular, voiceless obstruents appear to induce modal phonation on the following vowel that is acoustically closer to breathy voice than what is observed after voiced consonants. Similarly, effects of voicelessness on f0 and F1 (both higher after voiceless consonants), compatible with previous findings in the literature, were also observed. Meanwhile, voiced onsets appeared to have negligible, if any, effects on the phonation, f0, and F1 of the following vowel.

The implications of these findings for the phonological representation of two-series voice contrasts were also considered. Our results present a challenge to theoretical perspectives that assume phonologically inactive voiceless obstruents in true-voice languages. Additionally, our results suggest that pre-voicing may be the only reliable correlate of a [voice] or [slack] feature in Polish. Since the locus of the effects we observed is at the onset of the vowel following stop release, we suggest that theories of phonological representation should have ways of encoding the time course of such effects. At the same time, we suggest that laryngeal representations need not be defined in terms of phonetically explicit phonological features.