Phonology and Phonetics of L2 Telugu English

The paper provides a partial phonological and phonetic description of the segmental structure of L2 Telugu English (TE). Previous research on the subject has been carried out in the context of a more general notion of Indian English (IE), so the properties of TE as distinct from other varieties of IE (e.g., Gujarati English) have largely remained unexplored. We have primarily focused on areas that previous research identified as prominent issues in the study of IE: vowel inventory and production, representation and realization of liquids, word-final obstruent phenomena, and allophones of /w/. To account for these aspects of TE, we have combined a generative approach to the study of an individual’s linguistic competence with linguistic fieldwork as a means of collecting first-hand data. On the basis of collected data, we have conducted a spectrographic analysis of TE vowels and a distributional analysis of TE consonants. The paper provides the first description of the acoustic spaces of TE vowels. We found that all vowels except [ɔ] and [i] are more central in TE than in General American English. /r/ was realized as either [r] or [ɻ] without a specific pattern, and occasionally as [ɽ] in the intervocalic position. /l/ was realized as [ɭ] in word-final position and as [l] elsewhere. TE displayed word-final obstruent devoicing for all obstruents except for /b/, which was consistently unreleased. /w/ was realized as [ʋ] before front vowels and as [w] elsewhere. While previous research that concentrated on the broad notion of Indian English recognized the issue of /w/-allophony, it has not provided a principle that governs the exact distribution of /w/’s allophones. By combining the generative framework with linguistic fieldwork, we have accounted for this long-standing puzzle with a single rule: /w/ → [ʋ] / __ [–CONS, –BACK].


Introduction
The regional variety of English that is acquired as either a first (L1) or a second language (L2) in India is often referred to as Indian English (IE) (Sailaja, 2009). Most speakers of IE are also native speakers of other languages of India such as Hindi and Telugu (Sirsa & Redford, 2013, p. 393). IE is the most widely spoken non-native language in India, adopted by some 129 million people (according to the Ethnologue). A complicating factor in the study of IE is its great variability (Maxwell & Fletcher 2009).
When acquiring a second language, there are numerous factors that interact and affect the acquisition process, including the grammatical rules of a speaker's native language(s), the structure of the target language, and the quantity and quality of observable linguistic data (Wiltshire, 2006;Maskara, 2013;Slabakova, 2016). The nature of linguistic data that enter into L2 acquisition and shape the final outcome is to a large extent determined diachronically, as a result of the historical influence from different native languages. The data available to L2 IE learners are highly variable because the influencing native languages of India are numerous and span several different language families. The major families are the Dravidian family to which, for example, Telugu and Tamil belong; the Indo-Aryan family to which Hindi belongs; and the Tibeto-Burman family to which Manipuri and Himalayish belong (Kaur & Saini, 2014, p. 54). The notion of IE is a generalization over more than a hundred million speakers. In other words, 'Indian English' is a label that covers millions of individual linguistic competences. Such a high-level generalization, as we show below, obscures certain linguistic patterns of IE, patterns that only become apparent once a finer-grained approach to the study of IE is adopted.
Our main goal is to examine the properties of a single speaker's L2 competence of what is loosely called Indian English (Note 1). While various studies have focused on English of unspecified groups of speakers whose native languages are Hindi, Gujarati and Kannada (see below), there is a scarcity of information about the L2 English of native Telugu speakers. This study therefore focuses on aspects of Telugu English (TE), bearing in mind that this is merely a loose (but convenient) label and that the actual object of this study is a particular I-language.
To constrain the study, our focus will be limited to several phonological and phonetic phenomena that have been recognized in the literature as prominent characteristics of IE (Sailaja, 2009, p. §2) (Note 2).
Prabhakar Babu's (1976) doctoral thesis-a phonetic and phonological study of Telugu's influence on English-is the most comprehensive work on the topic. However, due to its reliance on the massively variable notion of IE and without revealing the sources of its Telugu English data, the study fails to detect several significant linguistic patterns that become apparent only once an I-language perspective is adopted. Apart from Babu's thesis, only cursory remarks on the phonology and phonetics of L2 TE exists in the literature (discussed below), and a thorough study of the topic within a generative www.scholink.org/ojs/index.php/sll Studies in Linguistics and Literature Vol. 5, No. 1, 2021 48 Published by SCHOLINK INC.
framework is lacking.

Figure 1. Monophthong Phonemes of GIE according to Pandey (2015, p. 305)
According to Pandey (2015, p. 305 a much greater extent than the consonant allophones from other varieties of English in terms of their phonetic realization. Almost each vowel is different in quality from RP". In neither case are these differences quantified nor is any further detail provided. We thus find not only a discrepancy in vowel inventory accounts but also an absence of a quantitative phonetic description of TE vowels. In their acoustic phonetic analysis of monophthongs produced by L2 speakers of IE whose L1 is either Hindi or Punjabi, Maxwell & Fletcher (2009, p. 68) conclude that the vowel inventory of this variety of English consists of the monophthongs /ɪ, iː, oː, eː, ɛ, ae, ʊ, uː, ɐ, ɐː/. In a follow-up study, Maxwell & Fletcher (2010) (Chand, 2009;Fuchs, 2019Fuchs, , p. 1382. Wiltshire (2005, p. 282)  These conclusions were based on the study of the production of English by native speakers of several Tibeto-Burman languages and of Gujarati, i.e., languages which belong to two different language families, and it is unclear if any of the conclusions apply to Telugu English, which belongs to yet another language family.
Moreover, Dinkar (2013, p. 40) presents the following insight about Kannada English (note that Kannada, like Telugu, is a Dravidian language): "If the labiodental approximant occurs, it is likely to occur in the /w/ initial onset position, and it is unlikely to occur in bilabial /w/ onset clusters and fricative /v/ onset environments. Results are mixed when looking at whether glide /ʋ/ occurs when bilabial /w/ in the initial onset position precedes rounded vowels".
In regard to the "/w/ initial onset position", e.g., for words such as with and way, Dinkar (2013, p. 32) suggests that the approximant [ʋ] will be used for the "bilabial /w/" by some speakers of Kannada, while [w] will be used by others. As for /w/ in onset clusters only, in words such as twelve and twenty, there were no instances in which any of the speakers realized /w/ as [ʋ]. When the fricative /v/ occurred in onset environments, Dinkar (2013, p. 33) reports that it was only realized as [ʋ] 0.013% of the time.  Sailaja (2009, pp. 19-20) claims that the trill /r/ is the only rhotic phoneme in IE, appearing in words such as rock and round. Scholars such as Babu (1976, p. 54) and Fuchs (2016, p. 25 Wiltshire (2006, p. 3) reports that speakers of Tibeto-Burman English tend to devoice obstruents in word-final positions. As an explanation for this phenomenon, she offers the observation that Tibeto-Burman languages generally do not allow codas. However, it is unclear why in light of that fact these speakers devoice word-final coda obstruents rather than deleting them, which would be more in line with the patterns of their native language. Unlike Tibeto-Burman languages, Gujarati and Hindi do allow voiced obstruents in codas (Wiltshire, 2006, p. 3). It is not surprising, then, that speakers of Tibeto-Burman languages such as Angami and Mizo devoice English word-final voiced obstruents much more often than speakers of Hindi and Gujarati (Wiltshire, 2006, p. 3). Overall, this topic is still understudied, particularly in regard to Telugu.
Retroflexion of the lateral approximant /l/ in TE has also been inconsistently described in previous literature. Babu (1976, p. 54) states that the lateral alveolar approximant /l/ can occur either at the beginning, middle or end of a word, in words such as look, fellow and signal. He suggests that an underlying /l/ can sometimes be realized as a retroflex lateral approximant [ɭ] when it occurs in word-final position, e.g., in words such as girl and pull. On the other hand, Pandey (2015, p. 304) claims that "the alveolar lateral tends to be retroflexed [ɭ] intervocalically". Sailaja (2009, p

Participants
Since the goal of this study was to provide a partial description of the linguistic competence of an individual (consistent with the I-language perspective; cf. Chomsky, 1986;Vaux & Cooper, 1999), the study included a single consultant, whose linguistic performance served as a source of evidence. Engineering at Concordia University (Montreal, Canada), with both programs administered exclusively in English.

Materials
The study consisted of a 30-minute and a 90-minute elicitation session; the two sessions were spaced two months apart to avoid priming and overloading the consultant. On the first occasion, the consultant was presented with a PowerPoint presentation that consisted of 20 isolated English words, 10 English sentences, 20 isolated Telugu words and 20 Telugu sentences. On the second occasion, he was presented with 52 English words and 30 English sentences. The presentation was set up to include one word per slide, accompanied by a picture, in order to facilitate his understanding. In regard to sentences, there were 2 per slide, also presented to the consultant on a computer screen. Words and sentences were chosen so as to provide a representative variety of phonological forms, thus allowing for an exploration of various segmental phenomena. A MacBook Pro was used to present the relevant material to the consultant as well as to record the elicitation session via an external microphone. Praat was used for the spectrographic analysis of the vowels, and R Studio for the creation of Figure 2.

Procedure
The consultant was given a concise explanation of how the elicitation sessions would proceed. He was told that he would be provided with English words and sentences in written form, which he would have to pronounce. The session began by eliciting monosyllabic words such as boat, disyllabic words such as island and finally some trisyllabic words like sunglasses or spectacles.
After all 20 isolated words were elicited, we moved on to sentences. Here, the consultant was provided with 10 English sentences, which included the 20 words that he was previously asked to produce. He was simply asked to read the English sentences aloud. After going through all 10 sentences, we moved away from eliciting data for a few minutes to speak about the consultant's background, in order to avoid fatigue or boredom potentially brought by repetitive tasks (Vaux & Cooper 1999, p. 19 Such a procedure comprised the first 30-minute elicitation session. Two months later, a similar procedure was implemented two more times, with a larger set of English words and sentences. These two sessions were split by a 30-minute break. Ultimately, the entire experimental paradigm yielded three 30-minute sessions of the form just described. R Studio was used to plot all vowels in terms of their F1-by-F2 values, as well as to create ellipses that capture the 95% confidence intervals around the occurrences of all vowel tokens.  (117) 1970 (164) 2254 (139) 1780 (165) 1612 (131) 1527 (114) 1106 (279) 1213 (143) 986 (147) 974 (149) 1204 (  American English (GAE; as described by Ladefoged & Johnson, 2010, p. 193)

Vowels (in black)
The consultant consistently produced the following diphthongs (and only those): [aɪ] in words such as price and kind, [aʊ] in mouth and pound, [ɔɪ] in words like boy and poison.

[w] and [ʋ]
In both isolated words and in connected speech, the consultant realized the underlying TE /w/ with two  There is seaweed on my towel. [siʋiʈ] [ʈaewʊɭ] My grandmother shares her wisdom while we eat waffles. [ʋɪzɖəm] [wɑiɭ] [ʋi] [ʋaefʊɭs] Does he take a shower after doing his homework? [ʃaewə] [homwək] The stars twinkle in the sky. [ʈʋɪkəɭ] The water washed the castle away. [wɔʈər] [wɑʃɪʈ] [ɛʋe̝ ] I could hear the bird tweeting.
[ʈʋiʈiŋ] be seen in both isolated words and in connected speech, the speaker's use of [ɻ] is only apparent in connected speech, the only exception being the word raspberry. These results are displayed in Table 4.

Word-final obstruents
Different voiced obstruents behaved differently at the end of words. While the voiced obstruents /ɖ, z, g, d ͡ ʒ, ʒ/ underwent devoicing in word-final position, the voiced bilabial plosive /b/ was unreleased (i.e., realized as [b ̚ ]) word-finally. An acoustic inspection verified that word-final bilabial stops maintained their voicing throughout their closure phase, despite lacking an audible sound burst. Obstruents did not undergo devoicing word-initially or word-medially. Results of both of these alternations, which occurred both in isolated words and in connected speech, can be seen in Table 6 and Table 7.

Laterals
In addition to the variation in the production of rhotics (see §3.3), the consultant's pronunciation of TE also featured a variation in the production of the lateral approximant. As shown in Table 8  island

account of Hindi English and Punjabi
English in that we did not find the diphthongs /ɪɘ, eə, ʊə/ in words like NEAR, SQUARE, and CURE respectively, but rather monophthongs /i, e, u/. The results are well aligned with Pandey's (2015) description of General Indian English (GIE) in terms of the monophthongs, but less aligned in terms of the diphthongs. Pandey's (2015) monophthong inventory, reproduced in Figure 1, also contains 11 vowels; the only difference is that in our results the vowel in words like FATHER is almost equal in backness to [ʊ] (see Figure 3) Babu's (1976) and Pandey's (2015) claims that [ə] occurs in both stressed and unstressed syllables without /ʌ/ ever appearing in any of those environments. Figure 2 provides the first description of the acoustic spaces of TE vowels (Note 5), following the methodology that Peterson and Barney (1952) and Hillenbrand et al. (1995) used in their seminal studies of GAE vowels. A notable pattern visible in Figure 2 and also reflected in the SD values in Table 1, is that front vowels exhibit less variability than back vowels.
In Figure 3, we have plotted TE vowels (red) and GAE vowels ( , she does not specify in which contexts each of these allophones is to be found. Also, Pandey (2015, p. 303) says that in GIE "[t]he labio-dental approximant /υ/ is substituted for two native English phonemes, the labial approximant /w/ [...] and the labio-dental fricative /v/, both of which are distinguished in restricted environments allophonically"; again, the exact distribution of /w/'s allophones is left unstated. We think that the reason why its distribution has remained a mystery is the following: conflating the production of many millions of speakers spread across many dialects belonging to several different language families, as is standard practice in the study of this issue (Maxwell & Fletcher, 2009, p. 53), obscures the patterns that would be detectable had a more fine-grained perspective been adopted. In other words, 'Indian English', as conceived in the literature, is such a massively variable notion, an amalgam of so many different things, that it precludes the statement of many important regularities. Specifically, it is not possible to find a rule that governs the distribution of /w/'s allophones unless IE is decomposed into less variable components. In order to achieve that, we have adopted a standard generative perspective which views language as a property of an individual's mind (following Chomsky 1986 and subsequent work), and have combined it with linguistic fieldwork in order to obtain first-hand data (following Vaux & Cooper 1999).
The results in Table 2 show that the roundedness of the following vowel has no impact on the realization of /w/: it is realized as [w] when followed by rounded vowels in words such as wood, towel and wall; also, it is realized as [ʋ] before unrounded vowels in words such as seaweed, wave and window. This shows that, contrary to (1), the distribution of allophones [w] and [ʋ] is not determined by the feature [ROUND] of the following vowel, but rather by the specification of its [BACK] feature. We have also not found any evidence supporting claims (2) and (3): the position of /w/ within a syllable did not play any role in determining the distribution of its allophones. We therefore conclude that although Dinkar's (2013) Table 6)-a well-known phenomenon in IE (Sirsa & Redford, 2013, p. 404)-so we are assuming that he internalized /ʈ/ and /ɖ/, and not /t/ and /d/, as part of his underlying inventory. Furthermore, word-final /b/ (but no other word-final stop) was consistently unreleased. We can postulate two phonological rules to account for these patterns: Similarly to the Tibeto-Burman languages which generally do not allow codas (Wiltshire, 2006, p. 3), Telugu does not allow consonants in word-final position except for /m/, /j/ and /w/ (Sailaja, 1999, p. 744). Babu (1976, p. 30) specifically states that stops never occur in word-final position in Telugu. It is possible that these restrictions on word-final obstruents in the consultant's native Telugu are reflected in his production of L2 TE. For the Tibeto-Burman English (e.g., non-native English of native speakers of Angami and Ao), Broselow (2018, p. 5) reports that both of these word-final effects can frequently be observed in all obstruents, and attributes those effects to the influence of the native language.
However, if that is the case, then several aspects of the influence of native Telugu on the production of non-native English remain unclear. Why is /b/ consistently unreleased word-finally while other obstruents are devoiced, i.e., why are these voiced obstruents treated differently? If Telugu doesn't allow obstruents word-finally, then why are the obstruents merely different in TE and not completely absent? Broselow (2018) invokes the notion of markedness to explain the general tendency of voiced obstruents to change into something less marked (in Optimality Theory, this would amount to the high ranking of the VOP 'voiced obstruent prohibition' constraint), but the typical markedness-based reasoning (e.g., that voiced obstruents are harder to articulate than their voiceless counterparts) is clearly insufficient to account for the two questions raised here.
While sources such as Masoko andTrinidad (2017, p. 11) andFuchs (2016, p. 24) Babu (1976, p. 37) suggests that /ɭ / never occurs at the beginning of words and contrasts medially and word-finally with /l/ in Telugu. In regard to Telugu English, Babu (1976, p. 54) says that /l/ is realized as a retroflex in word-final positions in some instances; however, he explicitly states that that is an irregular alternation which does not always occur. He also mentions /l/ occurs initially, medially and

Conclusion
In this paper, we have provided a partial phonological and phonetic description of the segmental structure of Telugu English (TE), a variety of English spoken mostly as a second language in the Indian states Andhra Pradesh and Telangana. The few existing phonological and phonetic descriptions of TE have been carried out in the context of a more general notion of Indian English (IE). Therefore, the properties of TE as distinct from other varieties of IE (e.g., Gujarati English) have largely remained unexplored. IE displays massive phonological variability due to the influence of many different native languages spoken in India. This in turn can obscure many relevant linguistic patterns. To mitigate the situation, we have combined a generative approach to the study of an individual's linguistic competence with linguistic fieldwork as a means of collecting first-hand data. Thus, a single native speaker of Telugu, whose second language is TE, participated in a series of elicitation sessions designed to obtain representative data in both isolated words and in connected speech. We have found that the TE vowel inventory consists of monophthongs /i, ɪ, e, ɛ, ae, ə, u, ʊ, o, ɔ, ɑ/ and diphthongs /aɪ, aʊ, ɔɪ/. This study provides the first spectrographic analysis of TE vowels (see Table 1) and a description of their acoustic spaces (Figure 2) in the vein of Hillenbrand et al. (1995)  in GAE (Figure 3).
Unlike the previous literature on the subject (e.g., Wiltshire, 2005)  Word-final obstruent devoicing was consistently featured in all voiced obstruents except for /b/, which was realized without an audible release but with preserved voicing. We have assumed that these word-final obstruent phenomena stem from the fact that Telugu is a language of the Dravidian family which generally does not allow consonants in word-final position except for /m/, /j/ and /w/ (Sailaja, 1999, p. 744), but the explanation for the exact nature of the pattern (i.e., its relative unnaturalness) is yet to be found.
Finally, the distribution of the lateral approximant /l/ was predictable and consistent: alveolar /l/ occurred both word-initially and word-medially, but never word-finally; in the word-final position, /l/ was always realized as the retroflex approximant [ɭ].
While the I-language perspective has allowed us to provide new insight into several long-standing problems in the study of IE, the main drawback of such an approach is that, due to its individualistic nature, its implications for the study of inter-speaker variation are very limited. It would therefore be beneficial in future research to employ a comparable methodology (i.e., combining generative linguistics and linguistic fieldwork) in order to eventually arrive at a more complete account of TE both in terms of individual metal grammars and their inter-speaker variation.