Multiple Modals in the Wild

A Study of 24,530 Multiple Modal Sequences in Naturalistic North American Speech

Steven Coats
English, University of Oulu, Finland, steven.coats@oulu.fi



ISLE 6 Workshop: The March of Data
University of Eastern Finland, 2 June 2021

Outline

1. Corpus of North American Spoken English (CoNASE): YouTube captions files, data collection, geolocation, data filtering

2. Double modals in North America

3. Methods: Regex, manual inspection/annotation

4. Results: Inventories, maps, interpretation

5. Caveats, summary, future outlook
   
   
   
   

Slides for the presentation are on my homepage at https://cc.oulu.fi/~scoats

CoNASE: YouTube captions files

City Council Meeting of Monroe, Georgia, January 8th, 2019

WebVTT file

CoNASE data collection and geolocation

• Focus on local government channels
• Channel selection with YT API (Coats 2019, 2020), transcripts with Youtube-DL
• Channel selection by scraping municipal websites (Coats in review)
• Use of Tor for http requests to YT search and for YT-DL
• Pass channel name + state/province/territory name to geolocator

CoNASE size

• 579 channels from US, 49k videos, 28k hours of video, 236m words, collected 2018 (Coats 2019, 2020)
• 1,993 additional channels from US and Canada obtained by scraping municipal websites, ~250k additional videos, ~135k hours, ~1b words (Coats in review)
• All US states and Canadian provinces/territories represented

CoNASE channel locations

• After additional filtering steps (only transcripts with word timing, only transcripts ≥ 100 words, only ASR transcripts, no translations)

1,252,066,371 words, 2,572 channels, 302k videos, 154k hours of video

Excerpt from corpus table

	country	state	channel_title	video_title	video_id	name	type	channel_id	channel_username	video_length	location	address	nr_words	text_pos	latlong
0	USA	Alabama	Baldwin County Alabama	Loxley School	KAFtEkKE_ik	COUNTY OF BALDWIN	general purpose	UCVEowdBqDvlT_TJDjf8YuyA	BCCommission	1651.38	Baldwin County, AL, USA	Baldwin County, AL, USA	3513	the_DT_25.619 Baldwin_NNP_26.619 County_NNP_27...	(30.6010744, -87.77633329999999)
1	USA	Alabama	Baldwin County Alabama	2019 Baldwin County Sewer Utilities Informatio...	6KvmxXRoMGA	COUNTY OF BALDWIN	general purpose	UCVEowdBqDvlT_TJDjf8YuyA	BCCommission	6110.09	Baldwin County, AL, USA	Baldwin County, AL, USA	16090	welcome_UH_1.37 everybody_NN_2.37 glad_JJ_3.89...	(30.6010744, -87.77633329999999)
2	USA	Alabama	Baldwin County Alabama	Heritage Museum of Baldwin County Alabama	bGL_FlSulZ4	COUNTY OF BALDWIN	general purpose	UCVEowdBqDvlT_TJDjf8YuyA	BCCommission	1255.31	Baldwin County, AL, USA	Baldwin County, AL, USA	2318	located_VBN_22.039 in_IN_23.039 the_DT_23.13 t...	(30.6010744, -87.77633329999999)
3	USA	Alabama	Baldwin County Alabama	AL State Veterans Memorial Cemetery at Spanish...	BX18sxhdL5Y	COUNTY OF BALDWIN	general purpose	UCVEowdBqDvlT_TJDjf8YuyA	BCCommission	4376.31	Baldwin County, AL, USA	Baldwin County, AL, USA	7991	good_JJ_15.94 morning_NN_16.94 everyone_NN_17....	(30.6010744, -87.77633329999999)
4	USA	Alabama	Baldwin County Alabama	USO Canteen Dance Re-enactment	VqCvFSku9Es	COUNTY OF BALDWIN	general purpose	UCVEowdBqDvlT_TJDjf8YuyA	BCCommission	3893.26	Baldwin County, AL, USA	Baldwin County, AL, USA	6846	[Music]_XX_5.1 [Music]_XX_14.86 [Music]_XX_104...	(30.6010744, -87.77633329999999)

Multiple modals in North America

Non-standard spoken-language feature in Southern United States (also in Scotland/N. Ireland, Caribbean Englishes/creoles)

• Y'all think you might could help me?
• I may can help you

Studied in the context of dialectology and sociolinguistics (Feagin 1979; Labov 1972, Mishoe & Montgomery 1994, Montgomery 1998), syntax (Batistella 1995; Di Paolo 1989; Hasty 2012a, 2012b, 2014; Nagle 2003), language history (Fennell & Butters 1996, Montgomery & Nagle 1993), English varietal typology (Zullo et al. 2021), and from other perspectives

Previous research

For most studies, data is heterogenous and limited in geographical scope

• Acceptability judgment surveys, sentence completion tasks (e.g. Butters 1973; Di Paolo 1989; Hasty 2011, 2012a, 2012b, 2014; Williamson 2018)
• Regional linguistic atlases (e.g. Montgomery 1989, 1998; Mishoe & Montgomery 1994)
• Observational studies (e.g. Labov 1972, Feagin 1979, Mishoe & Montgomery 1994)

Heterogeneity of collected data: "it is doubtful that atlas data can more than give us an outline of the prevalence and distribution of MMs" (Montgomery 1998: 103)

"Double modals occur with very low frequency in real-life utterances, and they are quite difficult to elicit in sufficient quantities and in a reliable fashion" (Fennell & Butters 1996: 265)

(cf. Grieve et al. 2014)

Preliminary research questions and objectives

Which double or multiple modals are used in naturalistic speech? Where?

• Verification of prior research, accounting for DMs in new locations, accounting for previously unattested DMs, mapping frequent types

Additional considerations

• Syntax, disfluency and "self-repair", status of epistemicity

Double and triple modals in North America: Methods

Corpus search → Creation of a table of occurrences with URLs: Every instance can be checked

13 modals/semi-modals:

can, could, may, might, must, shall, should, will, would, ought to, oughta, 'll, used to

= 156 possible double modals (excluding repetitions)

Double and triple modals in North America: Methods

Regex for timed-token corpus:

"\\s+("+x[0]+"n?_\\w+_\\S+\\s+(?:i?_\\w+_\\S+\\s+|we?_\\w+_\\S+\\s+|you?_\\w+_\\S+\\s+|he?_\\w+_\\S+\\s+|she?_\\w+_\\S+\\s+|it?_\\w+_\\S+\\s+|they?_\\w+_\\S+\\s+|haven?_\\w+_\\S+\\s+|'ve_\\w+_\\S+\\s+|'t_\\w+_\\S+\\s+|not_\\w+_\\S+\\s+){0,3}"+x[1]+"n?_\\w+_\\S+\\s+(?:haven?_\\w+_\\S+\\s+|'ve_\\w+_\\S+ |not_\\w+_\\S+\\s+|'t_\\w+_\\S+\\s+)?)"

This matches two-modal sequences

• might could and negations, perfect constructions, questions/inverted constructions with personal pronouns might not could, might have could, mightn't she could, etc.
• Less comprehensive triple modal regex
• 24,530 hits in corpus

Double modals: False positives

Many two-modal sequences are not actually "true" double modals

• "Food from a can would not taste as good" (homonym)
• "Will can help you with that" (personal name/homonym)
• "If I may, could I remark..." (clause/phrase overlap)
• "We will, if we can." "Would it cost a lot?" (speaker overlap)
• ASR errors ("Mike" = "might", "we'll" = "will", many others)
• "That woul- could be expensive" (self-repair, disfluency)
• "If you will... um... can help me" (self-repair, filled pause)
• Audio errors (video cuts off)
• Video unavailable

Double modals: Ocurrence data frame

• Editable html page with link to video of each DM instance found in the corpus at time of utterance
• Every occurrence is available for inspection

Double modals: Manual verification

• Disfluencies often marked by discrepancies in word timing or prosody (Levelt 1983, Postma 2000, Lickley 2015)
• If there are no disfluencies and the speech is semantically and logically coherent and consistent, it is a "true" double modal t
• Otherwise, annotate as sr self-repair, o overlap (phrase/clause boundary or speaker), a ASR error, u audio unintelligible, i video/transcript error
• Attempt to annotate ~10% of hits for each type

Double modals: Summary table

• Proportions of "true" double modals vs. disfluencies or transcript errors

	group	total	checked	% checked	true	self-repair	overlap	homonym/phone	asr fp	video error	audio problem
40	oughta could	1	1	1.000000	1.00	0.00	0.00	0.00	0.00	0.00	0.00
32	might could	393	346	0.880407	0.72	0.10	0.10	0.00	0.17	0.00	0.01
79	would oughta	5	3	0.600000	0.67	0.00	0.00	0.00	0.33	0.00	0.33
23	might can	426	355	0.833333	0.66	0.06	0.08	0.00	0.29	0.01	0.01
31	might should	88	15	0.170455	0.60	0.20	0.07	0.00	0.13	0.00	0.00
65	'll might	89	9	0.101124	0.56	0.22	0.00	0.00	0.33	0.00	0.00
77	would might	744	66	0.088710	0.52	0.44	0.05	0.03	0.12	0.00	0.03
102	should oughta	3	2	0.666667	0.50	0.00	0.00	0.00	0.50	0.00	0.00
39	oughta should	2	2	1.000000	0.50	0.50	0.50	0.00	0.00	0.00	0.00
38	oughta would	3	2	0.666667	0.50	0.00	0.00	0.00	0.50	0.00	0.50
100	should might	61	4	0.065574	0.50	0.25	0.25	0.00	0.25	0.00	0.00
73	'll could	58	2	0.034483	0.50	0.00	0.00	0.00	0.50	0.00	0.00
103	should must	18	2	0.111111	0.50	0.50	0.00	0.00	0.00	0.00	0.00
8	may should	105	62	0.590476	0.45	0.15	0.18	0.05	0.23	0.00	0.03
29	might would	434	36	0.082949	0.42	0.14	0.28	0.00	0.22	0.00	0.00
12	can might	219	13	0.059361	0.38	0.23	0.23	0.00	0.38	0.00	0.00
75	would may	360	11	0.030556	0.36	0.45	0.09	0.00	0.45	0.00	0.00
34	oughta can	3	3	1.000000	0.33	0.00	0.00	0.33	0.33	0.00	0.00
98	should may	77	3	0.038961	0.33	0.67	0.33	0.00	0.33	0.00	0.00
50	must could	4	3	0.750000	0.33	0.33	0.00	0.00	0.67	0.00	0.00

A few preliminary results

• Double modals are mostly southern, but can be found everywhere in North America
• There are more double modal types than has been previously thought
• Combinations from same root (can could, could can, will would, would will, 'll would and would 'll) are almost all disfluencies/self-repairs
• Shall is very rare (legal texts)
• The vast majority of modal-pronoun-modal sequences are false positives (phrase overlap)

"True" double modals (i.e. manually inspected) by type and state

• So far: 1,035 "true" double modals
• Concentrated in the South, but more widespread than previously thought
• DM inventory of 67 types judged to be "true" double modals (132 types if syntactic variants are considered, i.e. might could ≠ might not could, etc., cf. LAGS: 25 types)
• Triple modals: From 55 hits, 5 possible "true" (questionable)

		dm
group	state
might could	Tennessee	48
	North Carolina	38
	Florida	34
	Texas	24
	Alabama	21
	Georgia	9
	South Carolina	9
	Arkansas	7
	Kentucky	7
	Virginia	6
	Mississippi	5
	Louisiana	4
	Massachusetts	4
	Utah	4
	California	3
	Colorado	3
	Ontario	3
	Illinois	2
	Michigan	2
	Oklahoma	2
	Wisconsin	2
	Arizona	1
	British Columbia	1
	Connecticut	1
	Indiana	1
	Kansas	1
	Maine	1
	Maryland	1
	Minnesota	1
	Oregon	1
	South Dakota	1
	Washington	1
might can	North Carolina	54
	Tennessee	35
	Georgia	27
	Florida	21
	Alabama	15
	Texas	12
	Maryland	8
	Kentucky	7
	South Carolina	7
	Virginia	7
	California	6
	Mississippi	4
	Arkansas	3
	Connecticut	3
	Illinois	3
	Massachusetts	3
	Michigan	3
	Colorado	2
	Louisiana	2
	Alberta	1
	Kansas	1
	Minnesota	1
	New Mexico	1
	New York	1
	North Dakota	1
	Ohio	1
	Oklahoma	1
	Ontario	1
	Oregon	1
	Pennsylvania	1
	Rhode Island	1
	Wisconsin	1
might would	North Carolina	18
	Florida	11
	Tennessee	9
	Alabama	7
	South Carolina	7
	Michigan	5
	Georgia	4
	Kentucky	4
	Massachusetts	4
	Mississippi	3
	Ohio	3
	South Dakota	3
	Texas	3
	Virginia	3
	Washington	3
	Illinois	2
	Louisiana	2
	Minnesota	2
	Alberta	1
	Arizona	1
	Arkansas	1
	British Columbia	1
	California	1
	Connecticut	1
	Kansas	1
	Missouri	1
	New Jersey	1
	New York	1
	Ontario	1
	Oregon	1
	Pennsylvania	1
	Wyoming	1
would might	California	13
	Arizona	8
	Colorado	6
	Tennessee	6
	Alabama	3
	Florida	3
	Massachusetts	3
	Ontario	3
	Arkansas	2
	Connecticut	1
	Kansas	1
	Michigan	1
	New York	1
	Ohio	1
	Pennsylvania	1
	Virginia	1
will can	Texas	5
	California	4
	North Carolina	4
	Arkansas	3
	Minnesota	2
	Ohio	2
	South Carolina	2
	Alabama	1
	Arizona	1
	Florida	1
	Maryland	1
	Mississippi	1
	Missouri	1
	Nevada	1
	New Hampshire	1
	New Jersey	1
	Oklahoma	1
	Ontario	1
	Tennessee	1
	Utah	1
	Virginia	1
could might	Florida	4
	Massachusetts	3
	California	2
	Colorado	2
	Maryland	2
	Michigan	2
	North Carolina	2
	Ohio	2
	Alabama	1
	Alberta	1
	Arizona	1
	British Columbia	1
	Illinois	1
	Kansas	1
	Maine	1
	Missouri	1
	Nevada	1
	New Jersey	1
	Tennessee	1
	Texas	1
	Wisconsin	1
may should	California	4
	Texas	4
	Florida	3
	Massachusetts	2
	Michigan	2
	Tennessee	2
	Arizona	1
	Arkansas	1
	Colorado	1
	Illinois	1
	Kentucky	1
	Minnesota	1
	Missouri	1
	North Carolina	1
	Rhode Island	1
	Virginia	1
	Washington	1
may could	Tennessee	5
	Georgia	4
	North Carolina	4
	California	2
	Louisiana	2
	Michigan	2
	Arizona	1
	Colorado	1
	Illinois	1
	Indiana	1
	Maryland	1
	Mississippi	1
	Ontario	1
	South Carolina	1
	Texas	1
may can	Florida	7
	Alabama	5
	Georgia	4
	California	3
	Tennessee	3
	Kentucky	2
	Louisiana	2
	Colorado	1
	Virginia	1
would should	California	5
	Florida	4
	Arkansas	3
	Wisconsin	3
	British Columbia	2
	Colorado	1
	Illinois	1
	New York	1
	Ohio	1
	Ontario	1
	Pennsylvania	1
	South Dakota	1
	Tennessee	1
	Texas	1
might should	Tennessee	6
	Florida	3
	Georgia	3
	Arizona	2
	Illinois	2
	North Carolina	2
	Alberta	1
	California	1
	Massachusetts	1
	Pennsylvania	1
	Washington	1
would could	Arizona	8
	California	4
	Arkansas	2
	Connecticut	1
	Tennessee	1
should might	Michigan	2
	Alabama	1
	California	1
	Connecticut	1
	Iowa	1
	Maine	1
	New Mexico	1
	Ohio	1
	Oregon	1
	South Carolina	1
	Tennessee	1
	Wisconsin	1
will could	Georgia	3
	Wisconsin	2
	Alabama	1
	Arizona	1
	California	1
	Florida	1
	Louisiana	1
	New Jersey	1
	Tennessee	1
'll will	California	6
	Arizona	2
	Alabama	1
	Colorado	1
	Kentucky	1
	Ontario	1
might will	Illinois	3
	Alabama	1
	Arkansas	1
	Georgia	1
	Kentucky	1
	Louisiana	1
	Ontario	1
	Wisconsin	1
may will	North Carolina	3
	Arizona	2
	Alberta	1
	New Jersey	1
	Ontario	1
	Pennsylvania	1
	Virginia	1
would can	Alabama	1
	Arizona	1
	Ohio	1
	Ontario	1
	Pennsylvania	1
	Tennessee	1
will must	Indiana	1
	Iowa	1
	North Carolina	1
	Pennsylvania	1
	Texas	1
must might	California	2
	Massachusetts	1
	Minnesota	1
	Pennsylvania	1
might must	California	2
	Michigan	1
	Nebraska	1
	Wisconsin	1
may ought to	Alabama	1
	California	1
	North Carolina	1
	Tennessee	1
	Texas	1
may might	Arizona	1
	Colorado	1
	Kansas	1
	Oregon	1
	Virginia	1
can might	Arizona	1
	Maryland	1
	New York	1
	Tennessee	1
	Texas	1
'll might	Ontario	2
	Alabama	1
	Arizona	1
	California	1
would may	Alabama	1
	Arizona	1
	Florida	1
	Ontario	1
must may	Massachusetts	1
	Pennsylvania	1
	Texas	1
	Virginia	1
might 'll	New Hampshire	1
	New Jersey	1
	Oklahoma	1
	Tennessee	1
may would	Arizona	1
	Connecticut	1
	Florida	1
	Tennessee	1
can will	California	2
	Florida	1
	Ontario	1
will may	Arizona	1
	Ontario	1
	Tennessee	1
must can	California	2
	Tennessee	1
can may	Tennessee	2
	Florida	1
would oughta	Florida	1
	Minnesota	1
will might	Iowa	1
	North Carolina	1
might may	Ontario	1
	Tennessee	1
may must	Minnesota	1
	Ontario	1
could will	Ohio	1
	Ontario	1
could used to	Massachusetts	2
could ought to	Alaska	1
	Oregon	1
could may	Kentucky	2
can would	California	1
	Tennessee	1
can must	Florida	1
	Kentucky	1
can could	Connecticut	1
	Florida	1
'll may	Alabama	1
	Virginia	1
'll can	California	1
	Ontario	1
would must	Kansas	1
will should	Connecticut	1
should shall	Kansas	1
should oughta	Tennessee	1
should must	British Columbia	1
should may	Florida	1
should can	Indiana	1
shall would	Wisconsin	1
oughta would	Michigan	1
oughta should	Louisiana	1
oughta could	Minnesota	1
oughta can	Idaho	1
must would	Alberta	1
must will	Pennsylvania	1
must should	Illinois	1
must could	South Dakota	1
might oughta	North Carolina	1
could would	Ontario	1
could should	Arizona	1
can should	Kansas	1
'll could	Tennessee	1

How useful is the raw YT transcript data in estimating the extent of DMs?

• Compare smoothed spatial distribution of all hits in corpus with distribution of "true" hits

Might can, might could, and will can

Normalized frequencies for all occurrences and manually-verified occurrences of might could, might can, and will can used to calculate Getis-Ord G^*_i values based on a 100-nearest-neighbor binary weights matrix

• For frequent types/high proportion of true dms: Messy data gives a good approximation of spatial distribution
• For infrequent types/low proportion of true dms: Not as much

Double modals: Preliminary findings

• Double modals are used all accross N.A., but most widely used in the U.S. South
• Evidence for DMs in places not previously attested (West, North, New England, Canada)
• Changing semantics of modals (Coates 1983, Biber et al. 1999, Facchinetti et al. 2003, Gresset 2003, Leech 2003, Leech et al. 2009, Myhill 1995)

→ "Careful epistemicity" with double forms?

Caveats

• Method gets aggregate transcripts (no diarization)
• Algorithm mostly omits fillers such as uh, um, like
• ASR algorithm is a "black box" and changes over time
• Word error rates can be high in ASR captions files (Coats in review)
• False negatives need to estimated
• "Circumscribing the variable context": Status of potential alternants unclear
• Disfluencies are often marked prosodically, but not always
• How do we distinguish between covert repairs and fluent production of utterances generated by a speaker using a non-standard grammar?

Summary and outlook

Double and triple modals were examined in CoNASE

• Editable html page with links to utterance in corresponding YouTube video
• Manual annotation as true DMs or false positives
• Potentially interesting for dialectology, syntax, "self-repair" in a conversation analysis approach (Schegloff et al. 1977)
• Work continues!

Thanks for your attention!

References

Battistella, E. (1995). The syntax of the double modal construction. LLinguistica Atlantica 17, 19–44.

Biber, D., Johansson, S., Leech, G., Conrad, S. & Finegan, E. (1999). Longman Grammar of Spoken and Written English. Harlow: Pearson Education.

Butters, R. (1973). Acceptability judgments for double modals in Southern dialects. In Bailey, R. & Shuy, R. (eds.), New ways of analyzing variation in linguistics, 276–286. Washington, DC: Georgetown University Press.

Coates, J. (1983). The semantics of the modal auxiliaries. London and Canberra: Croom Helm.

Coats, S. (2019). A corpus of regional American language from YouTube. In Navarretta, C. et al. (Eds.), Proceedings of the 4th Digital Humanities in the Nordic Countries Conference, Copenhagen, Denmark, March 6–8, 2019 (pp. 79–91). Aachen, Germany: CEUR.

Coats, S. (2020). Articulation rate in American English in a corpus of YouTube videos. Language and Speech. https://doi.org/10.1177/0023830919894720

Coats, S. (in review). Dialect corpora from YouTube

Di Paolo, M. (1989). Double modals as single lexical items. American Speech, 64(3): 195–224.

Feagin, C. (1979). Variation and change in Alabama English: A sociolinguistic study of the White community. Washington, DC: Georgetown University Press.

Gresset, S. (2003). Towards a contextual micro-analysis of the non-equivalence of might and could. In Facchinetti et al. (eds.), 81–102.

Grieve, J., Nini, A, Guo, D. & Kasakoff, A. (2015). Using social media to map double modals in modern American English. Presented at NWAV 44, University of Toronto, October 22–25, 2015.

Hasty, J. D. (2011). I might not would say that: A sociolinguistic study of double modal acceptance. University of Pennsylvania Working Papers in Linguistics 17(2), 91–98.

Hasty, J. D. (2012a). This Might Could Help Us Better Understand Syntactic Variation: The Double Modal Construction in Tennessee English. Ph.D. Dissertation, Michigan State University.

Hasty, J. D. (2012b). We might should oughta take a second look at this: A syntactic re-analysis of double modals in Southern United States English. Lingua 122(14), 1716–1738.

Hasty, J. D. (2014). We might should be thinking this way: Theory and practice in the study of syntactic variation. In Zanuttini, R. & Horn, L. R. (eds.), Micro-syntactic variation in North American English, 269–293. Oxford: Oxford University Press.

Labov, W. (1972). Language in the inner city: Studies in the Black English vernacular. Philadelphia, PA: University of Pennsylvania Press.

Leech, G. (2003). Modality on the move: The English modal auxiliaries 1961–1992. In Facchinetti et al. (eds.), 223–240.

Leech, G., Hundt, M., Mair, C. & Smith, N. (2009). Change in contemporary English: A grammatical study. Cambridge: CUP.

Levelt, W. J. M. (1983). Monitoring and self-repair in speech. Cognition 14(4), 41–104.

Lickley, R. J. (2015). Fluency and disfluency. In M. Redford (ed.), The handbook of speech production, 445–469. Wiley-Blackwell.

McDavid, R. I. (1981). The conduct of an Atlas interview in the Gulf States. Ed. by S. E. Leas. LAGS Working Papers, First Series, No. 2 http://www.lap.uga.edu/Projects/LAGS/LAGS-WorkingPapers/1%20Working%20Papers%20first%20series%20paper%20%232.pdf

Mishoe, M. & Montgomery, M. (1994). The pragmatics of multiple modal variation in North and South Carolina. American Speech 69(1), 3–29.

Montgomery, M. (1989). Exploring the roots of Appalachian English. English World-Wide 10(2), 227–278.

Montgomery, M. (1998). Multiple modals in LAGS and LAMSAS. In Montgomery, M. & Nunnally, T. E. (eds.), From the Gulf States and beyond: The legacy of Lee Pederson and LAGS, 90–122. Tuscaloosa: University of Alabama Press.

Montgomery, M. & Nagle, S. J. (1994). Double modals in Scotland and the Southern United States: Trans-Atlantic inheritance or independent development? Folia Linguistica Historica, 14(1-2): 91–108.

Myhill, John. (1995). Change and continuity in the function of the American English modals. Linguistics 33, 157–211.

Nagle, S. J. (2003). Double modals in the Southern United States: Syntactic structure or syntactic structures? In Facchinetti et al. (eds.), 349–371.

Pederson, L. et al. (1986–1992). Linguistic Atlas of the Gulf States (6 vols.). Athens, GA: University of Georgia Press.

Postma, A. (2000). Detection of errors during speech production: a review of speech monitoring models. Cognition 77, 97–131.

Schegloff E. A., Jefferson G. & Sacks H. (1977). The preference for self-correction in the organization of repair in conversation. Language 53, 361–382.