Variation in phoneme inventories

Cormac Anderson1, Tiago Tresoldi2,1, Johann-Mattis List1, Simon J. Greenhill1, Robert Forkel1, Russel D. Gray1

1 Max Planck Institute for the Science of Human History, 2 Uppsala University

While originally drawn up for purposes of language description, phoneme inventories have been used in studies on a broad range of topics across cultural evolution. These studies have drawn on large cross-linguistic collections of inventories, such as UPSID (Maddieson 1984), PHOIBLE (Moran and McCloy 2020), and the Global Linguistics Database (Ruhlen 2008). Surprisingly, given the widespread application of these resources, no studies have attempted to assess either their internal consistency in coding data, or the extent to which the data different collections present is comparable.

We devise computational methods for assessing variation in phoneme inventories, both within and between the most used datasets. We link inventories in different collections to unique language identifiers from (Hammarström et al. 2020) and normalise graphemes with the help of a specially designed library (List et al. 2021). We use quantitative metrics for assessing inventory similarity and conduct a qualitative study into systematic differences between the different collections.

Our results show unexpected variation across the different collections, with inventories purportedly for the same language variety sometimes differing widely in terms of both size and the graphemes used. Some of this variation appeared to be random, while more of it was systematic and attributable to the coding policies used when drawing up the different datasets. For example, the Global Linguistics Database uses a single character to indicate the presence of a specific secondary articulation in consonants or non-modal phonation in (Creanza 2015), while other collections rather list all phonemes with these features. This can create quite wide disparities between inventories for languages with these features.

The systematic differences between collections that we uncover are ones well known from the phonological literature. In particular, Chao's (1934) discussion of the problems of phonemic analysis is a useful prism with which to assess the structural divergence of the collections under investigation. Some collections tend towards what he describes as “over-analysis”, writing long vowels, diphthongs, and geminate consonants as unit phonemes, while others tend rather towards “under-analysis”, presenting these instead as clusters.

Our results cast some doubt on the reliability of studies which make use of phoneme inventories as a source of data. They highlight the need for greater research into questions of comparability in this field and better tools for normalising phonological data for quantitative purposes.

References

Chao, Yuen Ren. 1934. “The Non-Uniqueness of Phonemic Solutions of Phonetic Systems.” Bulletin of the Institute of History and Philology 4 (4): 363–97. Creanza, N., M. Ruhlen, T. J. Pemberton, N. A. Rosenberg, M. W. Feldman, and S. Ramachandran. 2015. “A Comparison of Worldwide Phonemic and Genetic Variation in Human Populations.” Proc. Natl. Acad. Sci. U.S.A. 112 (5): 1265–72. Hammarström, Harald, Martin Haspelmath, Robert Forkel, and Sebastian Bank. 2020. Glottolog. Version 4.3. Jena: Max Planck Institute for the Science of Human History. Online: https://glottolog.org. List, Johann-Mattis: Anderson, Cormac: Tresoldi, Tiago and Forkel, Robert. 2021. Cross- Linguistic Transcription Systems. Version 2.0.0. Jena: Max Planck Institute for the Science of Human History. DOI: https://doi.org/10.5281/zenodo.3515744. Maddieson, Ian. 1984. Patterns of Sounds. Cambridge; New York: Cambridge University Press. Moran, Steven; McCloy, Daniel. 2020. PHOIBLE 2.0. Jena: Max Planck Institute for the Science of Human History. Online: https://phoible.org/ Ruhlen, Merritt. 2008. A global linguistic database. Moscow: RGGU.

Units of sound – what should we inventory? Julian Bradfield, [email protected] University of Edinburgh The workshop call describes the tension between, on the one hand, field linguists and descriptivists who routinely use the classical language-specific contrastive phoneme concept to construct inventories, and on the other hand, linguistic theories. Some major theories, such as SPE, sweep contrast under the carpet while maintaining a phoneme-like entity (a feature vector in a UR, for example), and in others, such as autosegmental theories, even the segment exists only as a temporal coincidence of tiers, so the phoneme is at most a contingent entity. In this presentation I take a strongly anti-universalist position to argue that not only is contrast language-specific, but the appropriate entities for sound inventories may be language-specific – and indeed I argue that descriptivists and orthography designers are implicitly aware of this. I consider several well-known and less well-known examples. The southern African San language known as !Xóõ (Traill 1985) or Taa (Naumann 2016) has, in a classical phonemic analysis, more than a hundred consonants. This was used by one of the more notorious phoneme inventory papers (Atkinson 2011) to support linguistic dispersion from southern Africa. Bradfield (2014) argued that in Khoisan at least, and probably also Bantu, rather than having, say, sixty clicks with five places of articulation and twelve manners, there are five independent clicks combining ‘in parallel’ with twelve manner-carrying accompaniments, functioning more or less as phonemes. We might go further, and consider whether the accompaniments are the appropriate unit, or whether finer units reflect both the systematicity of the phonology, and, perhaps, the psy- chological reality. Non-nasal clicks may be voiced or voiceless, and independently ejective or aspirated. Nasal clicks may be voiced, voiceless, or pre-glottalized voiced. Oral stops have a slightly smaller list of manners. The practical orthography (Naumann 2016) uses g, ’, h, n to mark the manners, with phonetic letters for the clicks themselves. Perhaps less contentiously, vowels may have glottalization, breathiness, pharyngealization and nasalization on top of /a,e,i,o,u/ – the modifications behave as ‘phonemes’, and again the orthography represents them as letters ’, h, q, n. Even in some European languages, such as Portuguese, nasalization is arguably, as written, an entity in the mind of the speaker, rather than a device for doubling the number of phonemes. On the other side of the world, many middle and southern American languages have nasality as a supra-segmental, even word-level, feature (Stanton 2017), so we get /-m˜o-,-mbo-/ and /*-mo-, *-mb˜o-/, where allophony resides neither in the vowel nor the consonant, but jointly according to word nasality. For Amuzgo, Kim (2020) argues that nasality makes the entire voicing distinction allophonic, contrary to previous analyses. How can one decide whether a feature (e.g. breathiness, nasality) is merely a feature in analysis, or a unit of sound in the language? I contend that psychophonetic experiments are the best way. For example, most British English speakers are familiar with the use of -h to denote voiceless fricatives: th, ph in English spelling, ch in Scottish English and German loanwords. In my experience, most untrained speakers do not generalize to voiced dh, bh, gh, unless they already have exposure to Gaelic or to transcriptions of other languages using this convention; so ‘spirantization’ is not a unit of sound in English. However, Bradfield (2014) showed that Bantu speakers do generalize orthographic voicing and aspiration markers to novel clicks. I predict that speakers of nasal harmony languages will generalize the allophonies to novel vowels (e.g. [y] which is rare in South America) and consonants (e.g. [q, c]), and that literate Taa speakers will generalize breathiness etc. to novel vowels. References

Atkinson, Quentin (2011). Phonemic diversity supports a serial founder effect model of language expansion from Africa. Science 332(6027), 346–349. Bradfield, Julian (2014). Clicks, concurrency and Khoisan. Phonology 31(1), 1–49. Kim, Yuni (2020). Nasal shielding and the non-phonological status of voicing in amuzgo. Talk presented at Nasality and laryngeal representations in phonology, University of Essex, January 2020. Naumann, Christfried (2016). The phoneme inventory of Taa (West !Xoon dialect). See Vossen and Haacke (2016). Stanton, Juliet (2017). Constraints on the distribution of nasal-stop sequences: an argument for contrast. Ph. D. thesis, Massachusetts Institute of Technology. Traill, Anthony (1985). Phonetic and Phonological Studies of !Xóõ Bushman. Hamburg: Buske. Vossen, Rainer and Wilfrid Haacke (Eds.) (2016). Lone Tree – Scholarship in the Service of the Koon: Essays in memory of Anthony Traill. Rüdiger Köppe Verlag. Investigating Phonemic Inventories without Minimal Pairs: Defining Contrast in the Vowel Inventory of Comox-Sliammon (ʔayʔaǰuθəm) Gloria Mellesmoen and Amanda Cardoso University of British Columbia Cross-linguistic investigations of vowel inventories focus on contrast maintenance between discrete vowel categories (usually phonemes), often expressed as differences in height (F1) and backness (F2) (e.g. Becker-Kristal 2010). Minimal pairs are often used as part of the core argumentation to support proposed phonemic categories and play a crucial role in identifying relevant groupings in the overall vowel inventory analysis. However, this approach presupposes that: (1) the basic units of contrast are phonemes; (2) height and backness are the main dimensions by which vowels contrast; and (3) minimal pairs diagnose contrast across all languages. This paper challenges these assumptions through an investigation of contrast in the vowel inventory of Comox-Sliammon (ʔayʔaǰuθəm). Comox-Sliammon is an endangered Central Salish language spoken in Canada with an estimated 47 L1 speakers (FPCC 2018). True minimal pairs are difficult to find because words are morphologically complex and may contain many inflectional and derivational affixes (Watanabe 2000). Note elusivity of minimal pairs is not merely a reflection of sparse/limited documentation; Rose & Blackmore (2018:587) highlight that polysynthetic and agglutinative languages have “word forms [that] are often phonologically and semantically too complex to lend themselves to descriptions based on minimal pairs” and that minimal pairs are exceedingly difficult to identify in these languages. If minimal pairs are scarcer in Comox-Sliammon than in other languages (e.g. English), it may not be fruitful to search for these forms to identify contrastive units when other approaches may be more suitable for exploring a wider variety of phonemic inventories across languages. Considering height and backness as primary dimensions for identifying vowel categories is also problematic. Mellesmoen & Huijsmans (2019) describe substantial acoustic overlap between Comox-Sliammon phonemic vowels (/a i u ə/) along the F1 and F2 dimensions. It is only when phonological environment is considered that some discreteness can be observed. Further, Blake (2000) argues that the distribution of /ə/ in Comox-Sliammon is predictable and does not need to be specified in the underlying form. She further suggests that /ə/ is differentiated from the underlying (/a i u/) vowels as it is nuclear, but non-moraic. Minimally, this suggests that length may play a role in distinguishing /ə/ from the other vowels. Taken together, these analyses suggest that phonological environment and duration may be important to defining contrast in the vowel inventory of Comox-Sliammon. We explore the relationship between F1, F2, F3, and duration in vowel productions from two speakers of Comox-Sliammon with Principal Component Analysis (PCA) (Jolliffe 2002). Results show that F1, F2, and duration substantially contribute to vowel contrast and plotted groupings do not correspond to the anticipated phonemes, but rather phonological environment. While F1 and F2 are important, duration and environment are crucial to understanding contrast in Comox-Sliammon vowels. This is consistent with predictions based on previous work, but not something captured in standard approaches to contrast. This study demonstrates the need for investigations (especially on understudied languages) which explore contrast in cross-linguistically robust ways. Our understanding of vowel systems’ organizations across languages relies on whether our methods allow cross-linguistic patterns to emerge.

References Becker-Kristal, Roy. 2010. Acoustic typology of vowel inventory and dispersion theory: Insights from a large cross-linguistic corpus. Los Angeles, CA: UCLA Doctoral dissertation. Blake, S. 2000. On the Distribution and Representation of Schwa in Sliammon (Salish). Doctoral dissertation, UBC. FPCC. 2018. Report on the Status of B.C. First Nations Languages http://www.fpcc.ca/ Jolliffe, I. T. (2002). Principal component analysis, 2nd edn. New York, NY: Springer- Verlag. Mellesmoen, G., & Huijsmans, M. 2019. The relationship between pronunciation and orthography: Using acoustic analysis as a practical illustration of ʔayʔaǰuθəm (Comox-Sliammon) vowel quality. In Proceedings of the 19th International Congress of Phonetic Sciences, Melbourne, Australia 2019 (pp. 979-983), eds. S. Calhoun, P. Escudero, M. Tabain & P. Warren. Rose, Y., & Blackmore, S. 2018. Questioning the role of lexical contrastiveness in phonological development: Converging evidence from perception and production studies. Canadian Journal of Linguistics/Revue canadienne de linguistique, 63(4), 580-608. Watanabe, H. 2003. A morphological description of Sliammon, Mainland Comox Salish: With a sketch of syntax. Endangered Languages of the Pacific Rim, Faculty of Informatics, Osaka Gakuin University.

Vowel reduction strategies in the world’s languages: Phonemic and phonetic realisations Sonja Dahlgren University of Helsinki

The inter-relatedness of phonology and phonetics can be seen in the way languages utilise vowel reduction. The process is basically phonological i.e. the system is often a result of the stress system of the language (stress-timed ~ syllable-timed). However, the outcome can either be phonemic or phonetic depending on 1) how far along the language is in the phonologisation of vowel reduction, and 2) what benefits the (L1) language user in terms of perception of the language. According to e.g. Harris (2005: 119-122) & Crosswhite (2001: 21-27), languages have two main strategies in vowel reduction. These include either centripetal i.e. centralisation of peripheral vowels to a schwa-like quality or centrifugal, which takes advantage of the most easily distinguishable phonemic characters, i.e. corner vowels /i, a, u/ (and often schwa). Harris and Crosswhite discuss that the differences result from two formal mechanisms: prominence reducing or contrast enhancing. They do not, however, comment on why different languages choose between the two. I aim to provide a preliminary typology of different vowel reduction strategies. The phonemic/phonetic outcome seems, to a large extent, to depend upon the (morpho- )phonological structures of the languages, whether they are so called ‘consonant-rich’ languages i.e. ‘moderately high’ to ‘high’ on the consonant-vowel ratio (as described in Maddieson 2013) or ‘low’ to ‘average’ languages on the C/VQ. According to Traunmüller (1999) the NW Caucasian languages and Northern Chinese, both with very few vowel qualities and considerable consonant inventories, use the strategy of unstressed vowels adapting to the quality of adjacent consonants (consonant-to- vowel coarticulation). According to Traunmüller, this strategy aids the listener in perceiving the consonant quality more accurately through the phonetic cues provided by the vowel. Many of the ‘low’ to ‘average’ C/VQ ratio languages utilise a reduction strategy of neutralising the vowel quality in unstressed syllables to a centralised vowel (e.g. English, French, German & Spanish (Delattre 2009); Persian (AbolhasaniZadeh, Abdolalizadeh & Sharifi Moghadam 2014), Bulgarian (Harris 2005: 130), and Eastern Ojibwa (Crosswhite 2001: 33). On the other hand, many languages that have a ‘moderately high’ to ‘high’ C/VQ ratio realise vowel reduction in the form of following the adjacent consonants’ quality i.e. have positional vowel allophones conditioned by their consonantal environment, often including the corner vowels and/or schwa. These types of languages can be found in at least the language families of Afroasiatic (Arabic, Bellem 2007: 174-175 & Coptic, cf. Author 2016), the Ndu (Mamanbu; Ndu languages in general as per Aikhenvald 2008: 41; 43), the NW Caucasian languages and Northern Chinese (Traunmüller 1999: 141-143), Uto-Aztecan (e.g. Ute, Givón 2011: 16-20), Uralic (e.g. Nenets, Salminen 1997: 39), and Indo-European (some of the more consonant-rich ones, such as Russian (e.g. Crosswhite 2001: 57-65)). It seems clear that consonantal languages that have a phonemic vowel reduction system based on the corner vowels (e.g. Russian) are further developed regarding phonologisation of vowel reduction than those that use a variable schwa with phonetic residues of consonant-to-vowel coarticulation (e.g. Arabic, Coptic, NW Caucasian and Northern Chinese languages). However, the phonetic residues still follow the place and manner of articulation of the nearby consonants, resulting in front vowel qualities near front (e.g. coronals, palatals) consonants, back vowels near back (velars, uvulars, pharyngeals) and round vowels near consonants that round vowel quality (labials). Therefore, both vowel reduction strategies of the consonantal languages serve the language users in the same way. In addition to being acoustically clearer, the corner vowels also provide useful phonetic information regarding the consonant qualities, representing the phonemic features ‘front’, ‘back’, and ‘rounded’, even if this realisation comes on the phonetic level. I argue that languages that have a phonemic realisation of the reduced vowels allow for more free variation than those that depend on the phonetic realisations of them. Therefore, the realisation of vowel reduction is language-specific and follows the developmental patterns of it.

References

AbolhasaniZadeh, Vahideh, Maryam Abdolalizadeh & Azadeh Sharifi Moghadam. 2014. Vowel reduction in Kermanian accent. Procedia – Social and Behavioral Sciences 136. 21-25. Aikhenvald, Alexandra Y. 2008. The of East , Papua New Guinea. Oxford: Oxford University Press. Author. 2016. Towards a definition of an Egyptian Greek variety. Papers in Historical Phonology 1. 90-108. Bellem, Alex. 2007. Towards a comparative typology of emphatics. University of London, doctoral dissertation.1 Crosswhite, Katherine. 2001. Vowel reduction in Optimality Theory. New York & London: Routledge. Delattre, Pierre. 1969. An acoustic and articulatory study of vowel reduction in four languages. International Review of Applied Linguistics in Language Teaching 7(4). 295-325. Harris, John. 2005. Vowel reduction as information loss. In Ph. Carr, J. Durand and C. J. Ewen (eds.). Headhood, elements, specification and contrastivity, 119–132. Amsterdam: Benjamins. Givón, Talmy. 2011. Ute reference grammar. Amsterdam: Benjamins. Maddieson, Ian. 2013. Consonant-Vowel Ratio. In Matthew S. Dryer & Martin Haspelmath (eds.), The World Atlas of Language Structures Online. Leipzig: Max Planck Institute for Evolutionary Anthropology. Pearce, Mary. 2008. Vowel harmony domains and vowel undershoot. UCL Working papers in Linguistics 20. 115-140. Salminen, Tapani. 1997. Tundra Nenets inflection. Helsinki: Suomalais-ugrilaisen seuran toimituksia 227. Traunmüller, Hartmut. 1999. Coarticulatory effects of consonants on vowels and their reflection in perception. In Proceedings from the XIIth Swedish Phonetics Conference, 141-144.

1 Single-spaced, re-paginated version (PDF) made by the author.

The role of contrastive feature hierarchies in the establishment of phoneme inventories B. Elan Dresher,1 Daniel Currie Hall,2 and Sara Mackenzie3 1University of Toronto, 2Saint Mary’s University, 3Memorial University of Newfoundland

Introduction. Databases such as UPSID (Maddieson 1984), P­base (Mielke 2008), and PHOI­ BLE (Moran & McCloy 2019) represent phonological inventories as sets of IPA symbols, with each symbol standing for a phonetic description akin to a set of fully specified distinctive fea­ tures (as in Chomsky & Halle 1968 and similar systems). Valuable though these resources are, we contend that this approach obscures the fundamental role of the phoneme as a unit in a language­specific system of contrasts. We argue that phoneme inventories are best understood in terms of contrastive feature specifications, assigned in language­specific hierarchies by the Successive Division Algorithm (SDA; Dresher 2009). In the SDA, features are assigned so as to divide the inventory recursively into smaller subsets until each phoneme has a distinct rep­ resentation; no feature is assigned unless it serves to mark some phonemic contrast that has not already been encoded. Specification by the SDA accounts for phonological processes that ig­ nore non­contrastive features, while avoiding problems with other forms of underspecification (see, e.g., Archangeli 1988). Understanding phoneme inventories in terms of contrastive hier­ archies of features has consequences for what kinds of typological generalizations can mean­ ingfully be made about them. The phonetic shapes of inventories and their phonological feature specifications mutually constrain each other, but neither wholly determines the other: Phonetic shapes of inventories constrain (but don’t dictate) feature specifications. The SDA does not stipulate a fixed ordering of features (cf., e.g., Clements’s (2009) Robustness Scale). This means that phonetically similar inventories may be phonologically distinct, even if the same features are used to specify them. For example, consider Mackenzie’s (2013: §2.1) analysis of laryngeal harmony in Ngizim and Hausa. Each language has a three­way contrast among plain voiceless, plain voiced, and implosive stops. Ngizim prohibits voiced pulmonic obstruents from following voiceless ones (*t…d), but the phonetic voicing of implosives is ignored (✓t…ɗ). Hausa disallows homor­ ganic pulmonic and implosive voiced obstruents from co­occurring (*ɗ…d), but allows voice­ less obstruents to occur with implosives (✓ɗ…t). This can be attributed to the different con­ trastive hierarchies in (1) and (2). In Ngizim (1), the harmonizing feature [±voice] is specified only on [−constricted glottis] obstruents; in Hausa (2), [±c.g.] is specified only on [+voice] obstruents. Harmony in each language ignores segments unspecified for the harmonizing fea­ ture. In an inventory with ejectives as well as the segments in (1) and (2), [±voice] and [±c.g.] would fully cross­classify, and neither could be underspecified. But an asymmetrical inventory allows different orders of features to yield different specifications. Feature specifications constrain (but don’t dictate) phonetic shapes of inventories. At the same time, contrastive specification can account for typological patterns in segment inventories, particularly ones that have sometimes been attributed to dispersion (e.g., Flemming 2004). As Hall (2011) points out, features assigned by the SDA can only specify how segments differ: no two phonemes can be given the same value for a feature unless that feature serves to distinguish them from some other phoneme(s), and any two phonemes must have contrasting values for at least one feature. There is, for example, no set of specifications that could be assigned to the unattested vowel inventory */ɨ ɘ ʉ/ that could not also characterize the inventory /i a u/. To the extent that the phonetic realization of segments tends to enhance their phonological feature specifications (Stevens & Keyser 1989), inventories will tend to be phonetically dispersed, even in the absence of any mechanism that explicitly evaluates or enforces phonetic distinctness.

1 (1) Ngizim: /ɗ d t/ (2) Hausa: /ɗ d t/

[+c.g.] [−c.g.] [+voice] [−voice] /ɗ/ /t/ [+voice] [−voice] [+c.g.] [−c.g.] /d/ /t/ /ɗ/ /d/

References Archangeli, Diana (1988). Underspecification in phonology. Phonology 5:2. 183–207. Chomsky, Noam & Morris Halle (1968). The sound pattern of English. New York: Harper and Row. Clements, G. N. (2009). The role of features in phonological inventories. In Eric Raimy & Charles E. Cairns (eds.) Contemporary views on architecture and representations in phonol­ ogy. Cambridge, MA: MIT Press, 19–68. Dresher, B. Elan (2009). The contrastive hierarchy in phonology. Cambridge: Cambridge University Press. Flemming, Edward (2004). Contrast and perceptual distinctiveness. In Bruce Hayes, Robert Kirchner & Donca Steriade (eds.) The phonetic bases of phonological markedness. Cam­ bridge: Cambridge University Press, 232–276. Hall, Daniel Currie (2011). Phonological contrast and its phonetic enhancement: Dispersedness without dispersion. Phonology 28:1. 1–54. Mackenzie, Sara (2013). Laryngeal co­occurrence restrictions in Aymara: Contrastive repre­ sentations and constraint interaction. Phonology 30:2. 297–345. Maddieson, Ian (1984). Patterns of sounds. Cambridge: Cambridge University Press. Mielke, Jeff (2008). The emergence of distinctive features. Oxford: Oxford University Press. Moran, Steven & Daniel McCloy (eds.) (2019). PHOIBLE. 2nd edition. Jena: Max Planck Institute for the Science of Human History. Published online at http://phoible.org/. Stevens, Kenneth N. & Samuel Jay Keyser (1989). Primary features and their enhancement in consonants. Language 65:1. 81–106.

2 The prosodeme inventories of the Danish dialects

The Danish language famously exhibits a contrast between the presence and absence of the laryngeal prosody “stød”. This contrast is historically related to the tonally manifested accent contrast found in most dialects of Norwegian and Swedish (e.g. Goldshtein in press). However, the synchronic comparability of the two systems is disputed. While some claim that they are fundamentally different (e.g. Basbøll 2005: 85), others claim that they are underlyingly very similar and differ mainly in phonetic implementation (e.g. Morén 2005). These discussions, which ultimately center on what is contrastive in the prosodic systems of the North Germanic languages, have generally failed to take into account the vast prosodic variation in the Danish dialects (see Ejskjær 1990). The present talk discusses the inventory of contrastive prosodic phenomena – prosodemes – in the Danish dialects, and their relation to Standard Danish and the North Germanic languages of mainland Scandinavia. It is well known, that some of the southernmost dialects of Danish exhibit a tonally manifested accent system much like what is found in Norwegian and Swedish. Others exhibit stød as in Standard Danish. And others again are said to exhibit neither tonal accent nor stød. What is especially illuminating, but only rarely discussed, is the dialects exhibiting both a tonal accent and stød (Andersen 1958, Skautrup 1968:136). The talk discusses the division of labor between stød and tonal accent in these dialects. What is contrastive in the dialects, what is the functional load of these contrasts, and which grammars underlie their distribution? The talk also raises questions related to representation. Grønnum et al. (2012) suggest that tone and stød is represented in Danish on separate tiers in autosegmental representation – with which I principally agree. Matters seem more complicated, however, when the dialectal variation is taken into account. Clearly tonal allophones of stød are observed in the dialects of Eastern Jutland (Molbæk Hansen 1978, Kyst 2008) and so-called ‘parasitic consonants’ are found in various dialects, where stød is realized as a fricative or plosive (Andersen 1955). Further, the phenomena ‘Western Jutlandic stød’ (Ringgaard 1960) and ‘Zealandic short vowel stød’ (Ejskjær 1967), which are phonetically very similar to stød, are generally argued to be different prosodemes due to their distribution, which is very different from that of stød. The main claim of the talk is that it may be possible to identify prosodic dialect continua between the dialects of North Germanic. The talk further discusses some of the empirical and theoretic problems involved in establishing the prosodeme inventories of the traditional dialects of Danish.

References Andersen, Poul. 1955. »Klusilspring« i danske Dialekter. Nordisk Tidsskrift for Tale og Stemme 15(1). 70–77. Andersen, Poul. 1958. Fonemsystemet i Østfynsk. Copenhagen: Schultz. Basbøll, Hans. 2005. The phonology of Danish (The Phonology of the World’s Languages). Oxford ; New York: Oxford University Press. Ejskjær, Inger. 1967. Kortvokalstødet i sjællandsk (Kortvokalstødet i Sjællandsk A22). Copenhagen: Akademisk forlag. Ejskjær, Inger. 1990. Stød and pitch accents in the danish dialects. Acta Linguistica Hafniensia 22(1). 49–75. https://doi.org/10.1080/03740463.1990.10411522. Goldshtein, Yonatan. In press. Stødets naturlige historie. In: Yonatan Goldshtein, Inger Schoonderbeek Hansen & Tina Thode Hougaard. MUDS 17. Aarhus: Aarhus Universitet Grønnum, Nina, Miguel Vazquez-Larruscaín & Hans Basbøll. 2013. Danish Stød: Laryngealization or Tone. Phonetica 70(1–2). 66–92. doi:10.1159/000354640. Kyst, Bodil. 2008. Trykgruppens toner i århusiansk regiolekt. Danske talesprog 9. 1–64. Molbæk Hansen, Peter. 1978. Stød and syllabicity in a Jutlandic dialect. Annual report of the Institute of Phonetics 12. 15–24. Morén, Bruce. 2005. Danish stød and Eastern Norwegian pitch accent: the myth of lexical tones. Talk given and the 13th Phonology Meeting Manchester Ringgaard, K. 1960. Vestjysk stød. Aarhus: Univseristetsforlaget i Aarhus. Skautrup, Peter. 1968. Det danske sprogs historie: Fra J. P. Jacobsen til Johs. V. Jensen. Vol. 4. Copenhagen: Gyldendalske Boghandel.

An alternative, phonetically based phoneme analysis of the Danish consonant system Camilla Søballe Horslund1, Rasmus Puggaard2 & Henrik Jørgensen3 1Aarhus University, 2Leiden University, 3Aarhus University

The standard phoneme analysis of the Danish consonant system (e.g. Rischel 1970, Basbøll 2005, Grønnum 2005) presents a good diachronic description, but we argue that it is a poor synchronic description due to three serious problems. 1) The standard analysis results in a large number of neutralisations that cannot be dissolved. A coda [ɪ̯ ] can represent two different consonant phonemes (/ɡ, j/) while a coda [ʊ̯ ] can represent three different consonant phonemes (/b, ɡ, v/), meaning that [ɪ̯] and [ʊ̯ ] can represent the same phoneme (/ɡ/). Due to lacking alternations for a large number of lexemes, it is simply impossible to determine their phonemic form, which conflicts with the widespread assumption that speakers store phonemic forms in their lexicon (see e.g. Hayes 2009; Gussenhoven & Jacobs 2017). 2) The standard analysis is based on historical variation that no longer occurs in standard Danish. Due to the loss of voicing in /b, d, ɡ/ and the loss of the so-called soft g (the velar continuant) (Brink & Lund 2018), the standard analysis establishes several phonemes whose allophones lack common phonetic content. The different realisations of /ɡ/, for instance, i.e. [k, ɪ̯ , ʊ̯ , Ø], do not have a single phonetic property in common. As lexical representations have to be built bottom-up from the data available in the language input, this is a huge problem for acquisition, since there is nothing in the phonetic input to suggest that these phones have anything in common. 3) The analysis does not apply to the bulk of the Danish lexicon as it is based on morphological alternations of a small subset of the Danish vocabulary, specifically irregular verbs (e.g. ba[ɪ̯ ] – ba[k]te, bake! – baked) and loanwords (e.g. filolo[Ø] – filolo[k]i). This is a problem for acquisition, as we assume these alternations to be acquired at a point at which the phonological system is already in place, given that 75% of Danish children have been found to have acquired all consonantal allophones by the age of five and a half years (Heger 1979). We argue that the above-mentioned characteristics make the proposed system unlearnable from the input, and the standard analysis is hence an implausible description of the phonological system we expect native speakers of Danish to have. We therefore propose an alternative phonological system inspired by Ács & Jørgensen (2016), who proposed a different set of phonemes in onset and coda, thereby presenting a system that exhibits biuniqueness as each phoneme has one single realisation. This analysis moves the complexity from phonology to morphology under the argument that the complexities are better placed within the morphological domain than within the phonological one. While we agree with this sentiment, we suggest that neither the phonological domain nor the morphological domain need to account for this complexity stemming from irregular alternations. Instead, we propose to move the load of these irregular alternations to the lexicon, as we suggest that they are rote learned on an individual basis. We propose to keep the same set of phonemes in onset and coda with defective distribution of a single plosive /ɡ/, which only occurs in onsets. [ɪ̯ ] and [ʊ̯ ] are always considered allophones of the phoneme whose onset realisation they most closely match (i.e. /j/ and /ʋ/). Our analysis has a number of advantages on the standard analysis. First, it is largely biunique and consequently does not result in static neutralisations, thereby making it possible to determine the phonemic form for all lexical entries. Second, all allophones of a phoneme share at least one phonetic property, thereby making the connection between allophones of the same phoneme detectable in the input. Third, our analysis does not require children to base their phonological system on irregular verb patterns or loanwords they are unlikely to learn in early childhood. We believe our analysis presents a phonological system that is learnable from the input, unlike the system presented by the standard analysis. The advantage on Ács & Jørgensen is that we suggest a smaller phonological inventory, since Ács & Jørgensen assume the semivowels to be individual phonemes and we do not.

References Ács, Péter & Henrik Jørgensen. 2016. ”Hvorfor er dansk vanskeligt? Danske konsonantsegmenters to ansigter”. Skandinavisztikaj Füzetek 10: 89-100. Basbøll, Hans. 2005. The phonology of Danish. Oxford: Oxford University Press. Brink, Lars & Jørn Lund. 2018. Yngre nydansk. In Ebba Hjort, Henrik Galberg Jacobsen, Bent Jørgensen, Birgitte Jacobsen, Merete Korvenius Jørgensen & Laurids Kristian Fahl (Eds.) Dansk Sproghistorie 2: Ord for Ord. Aarhus: Aarhus Universitetsforlag. Grønnum, Nina. 2005. Fonetik og fonologi. Almen og dansk. Copenhagen: Akademisk forlag. 3rd edition. Gussenhoven, Carlos & Haike Jacobs. 2017. Understanding Phonology. Oxon and New York: Roudtledge. 4th edition. Hayes, Bruce. 2009. Introductory Phonology. Malden, Massachusetts, Oxford and West Sussex: Wiley-Blackwell Heger, Steffen. 1979. ”Den fonetiske udvikling hos børn”. In Mogens Jansen & Jørn Lund (Eds.), Børnenes sprog - sprogene omkring børn, 13-67. Copenhagen: Hans Reitzel. Rischel, Jørgen. 1970. “Consonant gradation: A problem in Danish phonolgy and morphology”. In Benediktsson, The Nordic Languages and Modern Linguistics, 460-480, Reykjavík: Vísindafélag Íslendinga. The phonological and phonetic representation of phonemes in the SibInv database and sibilant inventory typology Joachim Kokkelmans Free University of Bozen­Bolzano

This contribution addresses two issues pertaining to establishing phonemicity in typological re­ search, and illustrates how these issues are solved in a typology of sibilant inventories formalised in Optimality Theory (Prince and Smolensky 1993/2004) using OTWorkplace (Prince et al. 2007−). While segment inventory typologies crucially depend on what is considered a phoneme, 1) many analyses fail to recognise the actual phonetic nature of phonemes, which has repercussions on the typologies themselves; 2) determining whether a phoneme is native, marginal or absent is a recur­ rent and tricky question. The OT typology is based on an own database (SibInv) of 258 languages. Importantly, a description which is informed by the typology can prevent misclassifications and predict marginal phonemes by detecting differences between underlying and observed inventories. 1. A problem at the source: the lack of phonetic accuracy in phonological descriptions Phonological theory and typology rely crucially on the accurate description of segments (Kiparsky 2018: 55, Hyman 2009: 214) as well as on precise phonetic knowledge, regardless of the impor­ tance attributed by phonologists to phonetic data (Rischel 1991: 238, Hayes and Steriade 2004). Despite the importance of phonetic accuracy, sibilants are frequently misclassified in typology. For instance, Flemming (2018)’s typology and databases like UPSID (Maddieson 1984) and P­ base (Mielke 2007) ignore the difference between alveolar and retracted alveolar sibilants (e.g. English [s] vs. Icelandic [s̠]; Vijūnas 2010). Yet, this difference is crucial for typology, because [s] and [s̠] are predominantly found in inventories with 2 places of articulation (PoA) and 1 PoA respectively, i.e. two different grammars. Representing both [s] and [s̠] as the phoneme /s/ leads to the erroneous assumption that they behave identically, which is counterproven by sound shift patterns (e.g. s­retraction in /rs/, Trask 1996; Kokkelmans 2020), loanword adaptation and phono­ logical behaviour. This phonetic difference has implications for phonology and is therefore crucial. Solutions to such common misapprehensions about sibilants include the awareness of all possible sibilant realisations ([s] vs. [s̠], [ɕ] vs. [ʃ],̻ [ʃ]̺ vs. [ʂ]̠ etc.) and diversification of descriptive sources. The OT typology I propose allows to detect such misclassifications (e.g. if a single language does not follow predicted patterns and turns out to have been misanalysed in the descriptive grammar(s)). 2. Phoneme status: the native vs. marginal problem solved by accidental gaps in OT The question whether languages possess a (marginal or native) sibilant phoneme (e.g. German /ʒ/ and /d͡ʒ/, Żygis et al. 2012: 306) is subject to avoidable debates. I propose that sibilant inventories have an underlying grammar, modelled in OT (see below), which predicts only certain sibilants (e.g. German: [s z ʃ ʒ t͡s d͡z t͡ʃ d͡ʒ]). Not all of these potential ‘slots’ are necessarily filled by sibi­ lant phonemes, for grammar­external reasons: historical developments create accidental gaps (e.g. OHG [tʰ] > German [t͡s], but no OHG *[dʰ] ergo no German [d͡z]). Accidental gaps are filled by loanwords (Maddieson 1986; e.g. German [d͡ʒ]ihad) and do not pose an articulatory difficulty for native speakers. The issue is therefore solved by considering ‘marginal’ phonemes as phonemes which are predicted to be possible by the underlying grammar but constitute an accidental gap on the surface. The OT typology thus allows to determine the phonemic status of a segment. The OT typology has 4 PoAs ([s] [s̠] [ʃ] [ʂ]̠ ) as candidates, together with 6 constraints: m.Sib against sibilants, f.Ident(Sib) as I­O faithfulness, m.Periph against [s] and [ʂ],̠ m.Post(C)D against [ʃ] and [ʂ],̠ m.Voice against voiced sibilants and m.Affr against sibilant affricates. References

Flemming, Edward (2018): “Systemic markedness in sibilant inventories”. Poster presented at the Annual Meeting on Phonology (AMP) 2018, UCSD. Hayes, Bruce and Steriade, Donca (2004): “Introduction: the Phonetic Bases of Phonological Markedness”. In: Hayes, Bruce, Kirchner, Robert and Steriade, Donca (eds.): Phonetically Based Phonology. Cambridge: University Press, 1–33. Hyman, Larry M. (2009): “How (not) to do phonological typology: the case of pitch­accent”. In: Language Sciences 31(2–3), 213–238. Kiparsky, Paul (2018): “Formal and empirical issues in phonological typology”. In: Hyman, Larry M. and Plank, Frans (eds.): Phonological Typology [Phonology and Phonetics 23], 54–106. Kokkelmans, Joachim (2020): “Middle High German and modern Flemish s­retraction in /rs/­ clusters”. In: De Vogelaer, Gunther, Koster, Dietha and Leuschner, Torsten (eds.): German and Dutch in Contrast. Synchronic, Diachronic and Psycholinguistic Perspectives. [Konvergenz und Divergenz 11]. Berlin, Boston: De Gruyter, 213–238. Maddieson, Ian (1984): Pattern of Sounds. Cambridge: University Press. Maddieson, Ian (1986): “Borrowed sounds”. In: Fishman, J. A. et al.: The Fergusonian Impact Vol.1: From Phonology to Society. Berlin: Mouton de Gruyter. Mielke, Jeff (2007): P­base (online). URL: https://pbase.phon.chass.ncsu.edu/. Prince, Alan and Smolensky, Paul (1993/2004): Optimality Theory: Constraint Interaction in Gen­ erative Grammar. Oxford: Blackwell. Prince, Alan and Tesar, Bruce and Merchant, Nazarré (2007­2020): OTWorkplace. [Computer software]. URL: http://sites.google.com/site/otworkplace/. Rischel, Jørgen (1991): “The Relevance of Phonetics for Phonology: A Commentary”. In: Pho­ netica 48(2–4), 233–262. Trask, Robert L. (1996): “Coalescence”. In: A Dictionary of Phonetics and Phonology. London, New York: Routledge. Vijūnas, Aurelijus (2010): “The Proto­Indo­European Sibilant */s”. In: Historische Sprachforschung 123(1), 40–55. Żygis, Marzena, Fuchs, Susanne and Koenig, Laura L. (2012): “Phonetic explanations for the infre­ quency of voiced sibilant affricates across languages”. In: Laboratory Phonology 3(2), 299–336.

Stem-initial accent and its effects on the phoneme inventory of Kam (Nigeria) Jakob Lesage Humboldt-Universität zu Berlin

Kam [Glottocode: kamm1249] is a Niger-Congo language spoken in central east Nigeria by approximately 8,000 to 12,000 people. This language is part of the Macro-Sudan belt, a linguistic area spanning a large part of Northern Sub-Saharan Africa (cf. Güldemann 2008). Like other Macro-Sudan belt languages, Kam distinguishes prosodically prominent and prosodically weak syllables, where prosodic strength is stem-initial and weakness is stem-final (cf. Idiatov & Van de Velde 2015). This distinction is equivalent to the distinction between stressed and unstressed syllables in other languages. In Kam, prosodic prominence exists alongside a tonal system with three contrastive tones.

This prominence asymmetry manifests itself in the segmental phonology of Kam. The phoneme inventory in accented positions is larger. There is free variation between certain sound pairs in prosodically weak positions that are in phonological opposition in strong positions (l vs. ɾ, i vs.ɨ and e vs. ɨ). Finally, stem-initial consonants are not subject to lenition rules that apply in prosodically weak positions. As a result, prominence asymmetry affects the analysis of the language’s consonant inventory. Concerning the vowel inventory, prominence asymmetry helps account for the rarity of three out of eight vowels (/i/, /e/ and /o/), which primarily occur in accented syllables.

This presentation outlines the accentual system of Kam with a focus on its manifestation in consonants and vowels. The relationship of the accentual system to the realization and position of consonants merits comparison to languages with similar prosodic systems. Describing consonant- emphasis prosody systems allows for higher resolution analysis of established linguistic areas such as the Macro-Sudan belt. In addition, it helps explain the emergence of cross-linguistically rare labial-velar consonants that are so characteristic of the area (Idiatov & Van de Velde 2015).

REFERENCES: Güldemann, Tom. 2008. The Macro-Sudan belt: Towards identifying a linguistic area in northern sub- Saharan Africa. In Heine, Bernd & Nurse, Derek (eds.), A Linguistic Geography of Africa, 151– 185. Cambridge, U.K.: Cambridge University Press. Idiatov, Dmitry & Van de Velde, Mark. 2015. Areal features in northern sub-Saharan Africa: Introduction. Kyoto University. (Presented at the Areal phenomena in northern sub-Sahara Africa, 8th World Congress of African Linguistics, Kyoto University.) (http://idiatov.mardi.myds.me/WOCAL8_Areal_Phenomena_NSSA/IDIATOV_VAN%20DE%2 0VELDE_2015_WS_Areal_phenomena_in_NSSA_SLIDES.pdf)

Estonian vs. the phoneme: one nil Markus A. Pöchtrager ([email protected]) University of Vienna

Background. The phonological segment as a locus of independent phonological difference has been central in the last century of phonological theorising (Anderson 1985, Halliday 2000), with occasional dissent: Twaddell (1935) criticised the phoneme (more broadly, any type of segment) as pure fiction. Both Firth (1948) and Z. Harris (1944) argue that there are phonological phenomena larger than the segment, an idea taken up in generative Autosegmental Phonology (Goldsmith 1976). Claim. My talk discusses the Estonian length system, which (i) is seen as rather unusual (Hint 1973, Lehiste 1960, 1965, Posti 1950, Prince 1980, Tauli 1973 etc.), and (ii) illustrates why phonology must go beyond the phoneme and segment. My theoretical framework is a further develop-ment of Government Phonology (GP; Kaye, Lowenstamm & Vergnaud 1985, 1990, Kaye 1995) as presented in Pöchtrager (2006). GP subscribes to the non-arbitrariness principle, requiring a direct connection between a phonological event and its context. Analysis. While English or Italian make do with a distinction between short and long (1–2), Estonian (3) displays three degrees of length (short, long, over-long). Two comments are in order, though. Firstly, bisyllabic words (3a) are claimed to establish a three-way segmental ("phonemic") contrast. Such claims disregard differences in the final nucleus, which is longer in (3ai) and (3aii) than in (3aiii): Differences between individual words are not located in one single point. (They also disregard morphological structure, as forms of type (3aiii) always involve analytic (stem-level) morphology.) Secondly, monosyllabic words (3b) show a trade- off similar to Italian: the more room is taken up by the consonant (C), the less remains for the preceding vowel (V) and vice versa. Such trade-offs again show that phonological differences can have multiple exponence, unlike what the notion of segment suggests. Close inspection of English reveals parallels to Estonian: (4) illustrates "pre-fortis clipping", known from the literature (cf. pioneering work by Peterson & Lehiste 1960), though usually disregarded as phonologically irrelevant because of its predictability (for arguments against such a position cf. Chomsky 1964 or Harris 1999, who I side with). Before voiceless consonants (bit, beat) the vowel is shorter than before neutral ("voiced") ones (bid, bead). Such a seemingly arbitrary interaction fails the non-arbitrariness principle, suggesting that the difference between English voiceless and neutral consonants should not be seen as melodic (qualitative), but rather as structural (quantitative): d, say, is the short version of t (5). As a result, the trade-off in (5) parallels (2) and (3b): The more room is needed by the consonant, the less remains for the preceding vowel (and vice versa). This shift in perspective allows for a non-arbitrary analysis: English does not display an arbitrary interaction of vowel length and consonant quality, but rather a length trade-off like other languages, such as Italian or Estonian: For example, Estonian [li::v] 'sand' and English leave [li::v] are identical, both with an over-long [i::] (cf. (6); space restrictions preclude discussion of the representational format here). The behaviour of length in the two languages is strikingly similar, in distribution and alternations, and the few areas where we do find differences are easily delimited (e.g. English does not have over-long consonants, while Estonian does, cf. (3biii)). My talk contributes to our understanding of Universal Grammar by arguing for the essential identity of two phonological systems usually seen as quite different, thus delimiting the range of linguistic variation. It also shows that phonemes/segments are not suffient to understand the nature of recurrent phonological patterns: The aforementioned trade-offs illustrate that differences can be lodged in more than one place simultaneously.

(1) English: fit ≠ feet, full ≠ fool, bet ≠ bait etc. (2) Italian: fato ['fa:to] 'fate' casa ['ka:za] 'house' V V C fatto ['fat:o] 'done' cassa ['kas:a] 'till' V C C (3a) Estonian: i. [lina(.)] 'linen (nom.sg.)' [sada(.)] 'hundred (nom.sg.)' ii. [lin:a(.)] 'city (gen.sg.)' [sa:da(.)] 'send! (imper.)' iii. [lin::a] 'city (par.sg.)' [sa::da] 'to receive (inf.)' (3b) Estonian: i. [ge::b] 'it boils' [si::d] 'silk (nom.sg.)' V V V C ii. [ge:b:] 'cape (nom.sg.)' [gi:d:] 'thanks (nom.sg.)' V V C C iii. [geb::] 'stick (nom.sg.)' [jud::] 'story (nom.sg.)' V C C C (4) English: bid [bi:d] bead [bi::d] bit [bit] beat [bi:t] (5) English bid [bI:d] V V C bead [bi::d] V V V C (reinterpreted): bit [bId:] V C C beat [bi:d:] V V C C (6)

References. Chomsky, Noam. 1964. Current Issues in Linguistic Theory. The Hague: Mouton. • Firth, John R. 1948. Sounds and Prosodies. Transactions of the Philological Society. 127–152. • Goldsmith, John. 1976. Autosegmental Phonology. PhD Dissertation, MIT. • Harris, Zellig. 1944. Simultaneous Components in Phonology. Language 20, 181–205. • Harris, John. 1999. Release the captive coda. UCL Working Papers in Linguistics 11. 165–194 • Hint, Mati (1973): Eesti keele sõnafonoloogia I. Tallinn: Eesti NSV Teaduste Akadeemia. Keele ja kirjanduse instituut. • Kaye, Jonathan, Lowen- stamm, Jean & Vergnaud, Jean-Roger (1985): The internal structure of phonological elements: a theory of charm and government. Phonology Yearbook, 2, 303–328. • Kaye, Jonathan, Lowenstamm, Jean & Vergnaud, Jean-Roger (1990): Constituent structure and government in phonology. Phonology, 7, 2, 193–231. • Kaye, Jonathan (1995): Derivations and interfaces. In: Jacques Durand & Francis Katamba (eds): Frontiers of Phonology. Longman Linguistics Library. London, New York: Longman. 289–332. • Lehiste, Ilse (1960): Segmental and Syllabic Quantity in Estonian. In: Indiana University Committee on Uralic Studies (ed.): American Studies in Uralic Linguistics, Bloomington: Indiana University Publications. 21–82. • Lehiste, Ilse (1965): The function of quantity in Finnish and Estonian. Language, 41:3, 447–456. • Peterson, Gorden E. & Lehiste, Ilse (1960): Duration of Syllab- ic Nuclei in English. The Journal of the Acoustical Society of America, 32:6, 693–703. • Pöchtrager, Markus A. (2006): The Structure of Length. Ph.D. thesis, University of Vienna. • Posti, Lauri (1950): On Quantity in Estonian. Suomalais-Ugrilaisen Seuran Aikakauskirja/Journal de la Société Finno- Ougrienne. 54:2, 1–14. • Prince, Alan S. (1980): A Metrical Theory for Estonian Quantity. Linguistic Inquiry, 11:3, 511–562. • Tauli, Valter (1973): Standard Estonian Grammar. Part I. Phonology, morphology, wordformation. Uppsala: Almqvist & Wiksell. • Twaddell, W. Freeman. 1935 [1957]. On defining the Phoneme. Language Monograph 16. Repr. in Martin Joos (ed.). 1957. Readings in Linguistics, vol. I. Washington: American Council of Learned Societies. 55–80. A unified treatment of segments and clusters Péter Rebrus and Péter Szigetvári Research Institute for Linguistics / Eötvös Loránd University

We argue that segments and clusters can be viewed as phonological constructions (Bybee 2001, Trón & Rebrus 2001, Hoffmann & Trousdale 2013), subject to similar formal and distri-butional properties. Systematic restrictions on the possible segments and their sequences in a language (inventory and phonotactics) are both modelled by contiguous ranges of available phonological constructions in the given language. These constructions can be arranged in scales in which they are ordered by their complexity/​markedness; cf. Harris 1990 on representations, Steriade 1994 on perception, Shariatmadari 2006 on articulation, Ohala 2009, Cser 2020: 64 on consonant clusters). The constructions available in a language form a contiguous range on the complexity scale: that is, the interval can be defined by the minimally and maximally complex constructions available in the given language. Constructions that fall between these two limits are expected to be available. We exemplify one of these scales by some monomorphemic consonant + plosive clusters (CT) in four different languages, taking t-final clusters as an example, (1). The scale cannot be derived from the sonority scale (cf. Rebrus & Szigetvári 2017). In these cases minimal complexity is the homorganic nasal + plosive nt (Eus, Rus) or the less complex geminate tt (Ita, Hun). Maximal complexity is fricative + plosive st (Eus, Ita) or non-homorganic plosive + plosive kt (Rus, Hun). (There are, of course, other language types with higher or lower maximal complexity, and perhaps higher minimal complexity, including languages where CT clusters are missing altogether, cf., e.g, Prince 1984, Piggott 1999.) A similar scale may be set up for plosives: here the glottal stop represents zero complexity, places other than alveolar, velar and labial (e.g., palatal, uvular) are the most complex in the scale, (2). Vowel inventories are also subject to such generalizations, (3). The occurrence of central vowels (ɨ, ə) in vowel inventories is similar to that of ʔ in the plosive inventory. The less complex non-low vowels (i, e, u, o) have one salient property, front or round, the most complex ones (ü, ö) have both. We can also posit a scale for restricted nonhigh vowel+glide clusters, (4), also known as diphthongs (cf. Maddieson 1984). Similar complexity scales can be set up for all other segment and cluster types. A further parallelism is that the availability of constructions is context sensitive, that is, the more marked the context, the smaller the range of constructions available in the same scale. For example, the sets of both consonant clusters and consonantal segments are typically more limited domain finally, preconsonantally, or after long vowels/diphthongs or reduced vowels. Vocalic constructions, on the other hand, may be limited in unstressed position and at domain edges. These phenomena can be dealt with by setting up minimal and maximal complexity values for available constructions for each potential left and right context. Thus, we claim that

1. phonological constructions available in a language form contiguous patterns in universal complexity scales, typically there are no isolated phonological constructions; 2. the restriction on the availability of phonological construction described above holds uniformly for both inventories (segments) and phonotactic patterns (clusters of segments), that is, from this point of view there is no formal difference between segments and clusters; 3. the scales may be relativized by phonological context (e.g., word edge, vocalic/consonantal context, stress), retaining the same formal restrictions in the relevant contexts; as the markedness of the context increases the range of available constructions decreases.

The advantage of the present analysis is that it provides a consistent formulation of phonological restrictions on (sequences of) segments without having to make reference to complex well-​ formedness constraints on feature combinations. The contiguity condition universally restricts the range of available constructions in a given language. (1) Consonant + t clusters LANGUAGE CONSTRUCTION SCALE MIN MAX zero < nas < liq < fric < plos < … Basque *tt nt lt st *kt nas fric Russian *tt nt lt st kt nas plos Italian tt nt lt st *kt zero fric Hungarian tt nt lt st kt zero plos

(2) Plosives LANGUAGE CONSTRUCTION SCALE MIN MAX zero < alv < vel < lab < pal < … French *ʔ t k p *c alv lab Hungarian *ʔ t k p c alv pal Tagalog ʔ t k p *c zero lab Wolof ʔ t k p c zero pal

(3) Nonlow vowels LANGUAGE CONSTRUCTION SCALE MIN MAX zero < front < round < front.round < … Italian *ɨ, *ə i, e u, o *ü, *ö front round Finnish *ɨ, *ə i, e u, o ü, ö front front.round Romanian ɨ, ə i, e u, o *ü, *ö zero lab Votic ɨ, ə i, e u, o ü, ö zero front.round

(4) Restricted vowel + glide sequences LANGUAGE CONSTRUCTION SCALE MIN MAX homorganic < low+high < heterorganic mid+high German, Latin *ej, *ow aj, aw oj *ew lo+hi o+j Southern Welsh *ej ow aj, aw oj ew o+w e+w English ej ow aj, aw oj *ew zero o+j Fijian, Ancient Gk. ej ow aj, aw oj ew zero e+w

References Bybee, Joan. 2001. Phonology and Language Use. Cambridge: Cambridge University Press. Cser, András. 2020. The Phonology of Classical Latin. Wiley-Blackwell. Harris, John. 1990. Segmental complexity and phonological government. Phonology 7: 255–300. Hoffmann, Thomas and Graeme Trousdale (eds.). 2013. The Oxford Handbook of Construction Grammar. Ohala, John J. 2009. Languages’ sound inventories: The devil in the details. In: François Pellegrino, Egidio Marsico, Ioana Chitoran and Christophe Coupé (eds.) Approaches to Phonological Complexity. Phonology and Phonetics [PP], 16. De Gruyter Mouton. Piggott, Glyne L. 1999. At the right edge of words. The Linguistic Review 16: 143–185. Prince, Alan S. 1984. Phonology with tiers. In Mark Aronoff and Richard T. Oehrle (eds). Language Sound Structure. Cambridge, Mass.: The MIT Press. 234–244. Rebrus, Péter and Péter Szigetvári. 2017. A complexity-based typology of consonant clusters. Paper presented at SinFonIJA 10, Dubrovnik, 23–24 October 2017. Shariatmadari, David. 2006. Sounds difficult? Why phonological theory needs ‘ease of articulation’. SOAS Working Papers in Linguistics 14: 207–226. Steriade, Donca. 1994. Positional neutralization and the expression of contrast. http://lingphil.mit.edu/papers/steriade/contrastive-gesture.pdf Trón, Viktor and Péter Rebrus. 2001. Morphophonology and the hierarchical lexicon. Acta Linguistica Hungarica 48: 101–136. Phonemes and features may coexist in phonological representations in a neural network: the role of feature economy Klaas Seinhorst, Paul Boersma, Silke Hamann University of Amsterdam

Although phonemes and features are traditionally considered to be in a hierarchical relation, with phonemes being bundles of features, we argue that they may coexist in a single representation, as Chládková (2014: 95) found for vowel systems. We hypothesise that phonemes are more likely to be represented in small inventories with low feature economy, where we quantify economy using Hall’s (2007: 176) “Exploitation” measure (hereafter E), defined as the number of categories divided by the number of possible categories. The sibilant system {ʂ ɕ s ʐ}, for instance, shows a binary voicing contrast and a ternary place contrast, yielding six possible categories; however, it exploits only four of those options, hence E = 0.67. In order to represent a fully economical system (where E = 1.0), the most parsimonious option may be to employ features only, certainly for larger inventories; on the other hand, we could imagine the system {ʂ ɕ s ʐ} to be represented more succinctly as something like “[–voi] + /ʐ/”, for instance. We test this prediction by modeling the acquisition of different sibilant inventories in a symmetric neural network (Boersma, Benders & Seinhorst 2018). The network contains three levels of representation: an auditory-phonetic form, a phonological surface form, and a lexical representation (Chládková 2014; Seinhorst, Boersma & Hamann 2019). The input to the network consists of sound-meaning pairs: in a learning step, fixed activities are applied at the auditory and meaning layers, and activation is then allowed to spread through the network before the weights of the connections are updated. We train the model on sibilant inventories defined on two separate auditory continua (VOT and spectral centre of gravity), the maximal inventory being {ʂ ɕ s ʐ ʑ z}; we restrict ourselves to three systems with different economy values, namely {ʂ ɕ s ʐ} (E = 0.67), {ʂ ɕ s ʐ ʑ} (E = 0.83), and {ʂ ɕ s ʐ ʑ z} (E = 1.0). A network before learning is shown in Fig. 1a. This network will be trained on the inventory with four phonemes (the four groups of nodes in the meaning level correspond to {ʂ ɕ s ʐ}, respectively); the two auditory continua ([AudF]) can be seen at the bottom of the figure; the intermediate level is the phonological surface form (/SF/). Fig. 1b shows the same network after 20,000 learning steps. We see that the phonological level has come to encode both features and phonemes: SF nodes 1 and 5 are connected to the “ʐ” phoneme only, while nodes 2, 3 and 6 are connected to all voiceless segments, and node 4 to both retroflexes. We find such mixed representations for the {ʂ ɕ s ʐ} inventory only; for the larger, more economical systems, SF nodes always represent feature values but never phonemes, as we expected. Our findings suggest that phonemes and features need not stand in a hierarchical relation; they can be part of the same representation, depending on properties of the input language. We like to compare this to seeing a car, where perceiving the car as a whole does not preclude one from simultaneously seeing its parts (e.g. the wheels).

“ʂ” “ɕ” “s” “ʐ” “ʂ” “ɕ” “s” “ʐ” “meaning”

/SF/

[AudF] Fig. 1a. The neural network before learning… Fig. 1b. …and after learning is complete. References:

Boersma, Paul, Titia Benders & Klaas Seinhorst (2018). Neural network models for phonology and phonetics. Manuscript, University of Amsterdam. Chládková, Kateřina (2014). Finding phonological features in perception. Doctoral dissertation, University of Amsterdam. Hall, Daniel Currie (2007). The role and representation of contrast in phonological theory. Doctoral dissertation, University of Toronto. Seinhorst, Klaas, Paul Boersma & Silke Hamann (2019). Iterated distributional and lexicon- driven learning in a symmetric neural network explains the emergence of features and dispersion. In Proceedings of the 19th International Congress of Phonetic Sciences.

Lexical Non-Syllabic Vowels in Portuguese: The case for /j/ ad /w/ João Veloso University of Porto (Portugal) – Faculty of Humanities (FLUP) & Centre of Linguistics (CLUP)

1. Mainstream phonologists of Portuguese1 (Mateus 1982; Mateus & Andrade 2000; Mateus et al. 2003) postulate glides [j] and [w] (non-syllabic vocoids, traditionally referred to as “semivowels”) as the phonetic counterparts of lexical /i/ and /u/. That is to say, in the phonological segmental inventory of Portuguese, phonemic /j/ and /w/ are not accepted as such by the most authoritative phonological descriptions of the language. Accordingly, [j] and [w] are always interpreted as the surface result of glidisation rules that reespecify /i/ ([j]) and /u/ ([w]). The main arguments in favour of this interpretation are twofold: (1) [j] and [w] always occur phonetically in the adjacency of other vowel (pai [paj] ‘father’; pau [paw] ‘stick’); they are, in this way, contextual; (2) [j] and [w] often alternate with [i] and [u]: desfiar [] ‘unthreading’ vs. fio [] ‘thread’; luar [] ‘moonlight’ vs. lua [] ‘moon’. 2. The traditional interpretation of 1., albeit preserving the economy of the segment inventory of Portuguese, does not take into account, though, the lexical/grammatical behaviour of [j] and [w], nor the fact that the relations between phonemes and their phonetic realisations are not strictly biunivocal. 3. In this presentation, we shall offer an alternative view regarding the phonological status of Portuguese phonetic glides. Our explanation integrates both linguistic behaviour and non-biunivocality. Inspired by Van Oostendorp’s (1998) rationale for his categorisation of different phonological statuses for schwas, we argue that: (a) it is plausible to postulate that Portuguese glides derive from lexically syllabic vowels whenever they alternate with such vowels (examples in (2)); (b) however, when such alternations are not found in the language, no strong evidence supports the view that glides are the reespecification of different phonological representations. As for the latter ((b)), these cases are more often found in the so-called “falling diphthongs” ([VG]) – see (1)), where glides usually do not alternate with vowels. Moreover, and differently from what is observed in the Portuguese “rising diphthongs” ((2)): their realisation as non-syllabic vocoids is always mandatory; they inhibit vowel reduction in unstressed position; they are found in stable lexical oppositions contrasting with other “semi- vowels” or consonants (e.g.: pai [paj] ‘father’ ~ pau [paw] ‘stick’ ~ par [] ‘pair’). For these realisations, our proposal is that [j] and [w] stand for phonological /j/ and /w/, i.e., phonemes of the Portuguese segment inventory distributionally obliged to follow a syllabic vowel (in stead of viewing this distributional constraint as a contextual effect affecting a vowel for which no strong arguments are found in the contemporary stage of the language, we assume that certain phonemes can be distributionally restricted at the lexical level, as it is the case, for instance, of the palatal nasal, unadmitted in word-initial position in Portuguese).

In sum, the core of our proposal is as follows: A – [j] and [w] Portuguese glides are allophonic variants of /i/ and /u/ when they alternate with [i] and [u] (as it is the case, basically, in the so-called “rising diphthongs”, equated with phonological hiatuses); B – [j] and [w], in Portuguese, are the phonetic realisations of phonemic /j/ and /w/ when no alterations with [i] and [u] are found in the contemporary stage of the language (this happens mostly with the “falling diphthongs” of this language, the main exception being found in word-final position, where these “semivowels” can be accepted as the glidised allophones of noun class markers, as will be discussed, too, in this presentation).

The prosodic status of these “phonemic non-syllabic vocoids” (as V2 of branched nuclei or as C-like slots of syllable codas (Barbosa 1994; Magalhães 2016) will also be discussed in this talk.

1 In this abstract, we restrict “Portuguese” to European Portuguese only.

REFERENCES:

Barbosa, J. M. 1994. Introdução ao Estudo da Fonologia e Morfologia do Português. Coimbra: Almedina. Magalhães, J. 2016. Main Stress and Secondary Stress in Brazilian and European Portuguese. In: W. L. Wetzels et al. (Eds.). The Handbook of Portuguese Linguistics. Oxford: Wiley-Blackwell, 107-124. Mateus, M. H. M. 1982. Aspectos da Fonologia Portuguesa. 2nd ed. Lisboa: INIC. Mateus, M. H.; Andrade, E. 2000. The Phonology of Portuguese. Oxford: Oxford University Press. Mateus, M. H. M. et al. 2003. Gramática da Língua Portuguesa. 5ª ed. Lisboa: Caminho. Van Oostendorp, M. 1998. Schwa in Phonological Theory. Glot International. 3(5): 3-8.