English version of Herring, S.C. 1998. Le style du courrier électronique: variabilité et changement. Terminogramme 84-85 (March): 9-16.

Variability and Change in E-mail Style

Susan C. Herring

University of Texas at Arlington

susan@ling.uta.edu

Introduction

Computer-mediated communication, especially that which takes place via the global network known as the Internet, is fertile soil for language change. From a sociolinguistic perspective, a number of social conditions for change are met:

• Internet culture is ubiquitous (at least in the Western world) and statusful.

• The Internet is especially popular among younger users, including adolescents, who are traditionally one of the most linguistically innovative populations.

• The number of users and frequency of interaction is more than enough to cause usages to become popular and spread.

• Diverse groups of speakers come into contact—indeed, the Internet represents a contact situation of unprecedented scale (Paolillo, 1996; cp. Weinreich, 1953).

These factors might be taken to suggest that a particular style—that favored by younger users with technologically savvy who spend a lot of time online—will spread, and that others who wish to orient towards membership in the statusful group will converge towards that style.

In fact, however, very few scholars have looked at computer-mediated communication (henceforth CMC) from a sociolinguistic perspective. Most CMC research has been carried out by scholars in communication and social psychology, who point to the unique features of the computer medium as the most likely determinants of language use and language change. They typically cite factors such as the following:

• The channel is "lean" (i.e., lacks audio and visual information), obliging users to compensate verbally for social (age, gender, race, etc.) and expressive (prosodic, kinesic, etc.) information (Daft and Lengel, 1984).

• Participants are "anonymous", that is, they can't see one another, may not know one another, and are typically separated by space and time. This reduces social accountability, and is claimed to lead to decreased politeness and attention to addressees' social "face" (Kiesler et al., 1984).

• Small screen sizes and limited message buffers/message storage as well as attentional limits of users may limit message length, sentence length and complexity.

Few of these claims, however, have been investigated empirically by linguists. One might reasonably expect such effects to be present, but are they really, and to what extent? Moreover, there has as yet been no study of CMC that looks at variability and change over time, although people are fond of speculating that language is changing as a result of computer use. Does frequent computer-mediated interaction change the way we use language, and ultimately, perhaps, the structure of language itself?

Predictions about CMC and language change

In this article, I evaluate four kinds of predictions that have been advanced in the literature regarding variation and change in computer-mediated language. The predictions which follow are based primarily on remarks made by Baron (1984) and Hale (1996):

1. CMC use over time leads to reduced syntactic complexity; a tendency towards short, direct sentences and sentence fragments.

2. CMC use over time leads to reduced formality; a tendency towards casual usage and increased typographical errors.

3. CMC use over time leads to reduced politeness; a tendency towards bluntness and "flaming".¹

4. CMC use over time leads to reduced variability; a tendency towards stylistic homogeneity.²

Optimists about the new technology see the first three as liberating trends. Constance Hale, an editor for Wired magazine, quotes a Wired writer as describing messages posted to the Internet as "a whole new fractured language—definitely not as elegant or polished as English used to be, but in a way, much more vital."³

Others are more pessimistic. Naomi Baron, a linguist writing in the early 1980's, posits that "computer conferencing fosters the use of a very particular (and homogeneous) conversational style", characterized by frequent arguments and flaming. She further predicts that since "CMC is ill-suited for "social" uses of language", it "may lead human language [to] lose the majority of its functions" and "undermine the basic social fabric that any society needs to survive."⁴

The Study

In order to test these four predictions, I searched for the oldest corpus I could find of continuous computer-mediated interactions. Thanks to a tip I received on an Internet discussion group devoted to "cyber history",⁵ I was able to locate an archive containing 11 years of e-mail messages posted to an early electronic distribution list on the Arpanet (the predecessor of the Internet).⁶ The MsgGroup discussion list, which operated between 1975 and 1986, is a moderated forum devoted to discussion and development of protocols for computer mesasging systems. Its roughly 100 participants are computer science professionals; most are white, middle-class males working in the United States in universities, the military and private computer companies. As on contemporary Internet discussion lists, interaction on MsgGroup is many-to-many, and is transacted via e-mail messages posted to a central site and then redistributed to all subscribers. All messages posted to MsgGroup are in English.

For the present study, I extracted six 50-message samples from the MsgGroup archives at roughly 2-year intervals. For each message, I coded and counted features of syntactic complexity, formality, and politeness. I also conducted statistical analyses of variance and examined measures of standard deviation for evidence of change in degree of variability.

Results

While the overall results are mixed, a number of features significantly change in frequency between 1975 and 1986. Moreover, the changes are in the direction predicted by Baron (1984) and Hale (1996), with one exception. The results are presented separately for each category of prediction below.

Syntactic complexity

As measures of syntactic complexity, I counted the frequency per sentence of subordinate clauses, complement clauses, relative clauses, passive verb phrases, and "heavy" noun phrases (defined as noun phrases made up of four or more juxtaposed nouns, for example "the domain style naming system"). The results of a 1-way ANOVA for each of these categories are presented in Table 1:

	Subord. Cl.		Compl. Cl.		Relative Cl.		Passive VP		Heavy NP
	Avg.	SD	Avg.	SD	Avg.	SD	Avg.	SD	Avg.	SD
1975	.25	.23	.47	.33	.21	.23	.29	.26	.02	.05
1977	.28	.18	.40	.33	.36	.40	.24	.28	.02	.06
1979	.28	.32	.32	.32	.30	.32	.44	.52	.06	.14
1982	.38	.42	.47	.39	.38	.41	.48	.41	.03	.06
1984	.33	.29	.45	.40	.29	.27	.40	.40	.05	.13
1986	.24	.20	.55	.29	.23	.22	.29	.24	.02	.06
msgs.	N=150^7		N=150		N=150		N=150		N=150
	p = .556		p = .290		p =.307		p = .122		p = .586

Table 1. Syntactic complexity measures (1-way ANOVA)

As Table 1 shows, the frequency of each feature remains remarkably stable over time. There is some variation between samples, but no significant trend emerges for any feature of syntactic complexity.

This result is illustrated graphically in Figure 1, for all features of syntactic complexity combined.⁸ To smooth out the variation between samples, Figure 1 combines the results for the first two samples, the middle two samples, and the last two samples into 'early', 'middle' and 'late' periods, respectively:

Figure 1. Change in syntactic complexity over time

Rather than showing a decrease in syntactic complexity over time, the syntax of MsgGroup messages remains at a relatively high level of complexity throughout the group's existence.

I also counted three measures of syntactic "simplicity": coordinate clauses (conjoined by 'and', 'or', 'but' or 'so'), simple sentences (containing only a subject, a predicate, and optionally an adverbial), and sentence fragments. The ANOVA results are summarized in Table 2:

	Coordinate Clause		Simple Sentence		Fragment
	Avg.	SD	Avg.	SD	Avg.	SD
1975	.26	.19	.25	.29	.12	.17
1977	.13	.15	.24	.33	.15	.25
1979	.22	.29	.15	.23	.12	.24
1982	.34	.40	.12	.15	.08	.13
1984	.16	.21	.22	.26	.02	..05
1986	.23	.19	.18	.14	.14	.17
msgs.	N=150		N=150		N=150
	p = .072		p = .335		p =.131

Table 2. Syntactic simplicity measures (1-way ANOVA)

Again, there is no evidence in these results of significant increase or decrease over time. This is represented graphically in Figure 2 for all features combined:

Figure 2. Change in syntactic simplicity over time

The frequency of simple syntactic constructions remains relatively low throughout the life of MsgGroup. Features of syntactic complexity and simplicity will be illustrated in examples given below.

Formality

To measure formality and informality, I counted a cluster of features for each. Formality features include marked syntactic complexity (i.e., those features of syntactic complexity that stand out as "formal sounding" in the context of the overall level of formality of the group), formal lexicon, nominalizations, and fixed formal phrases. Informality features include colloquial words and phrases, contractions, abbreviations, errors, emoticons (sideways "smiley" faces constructed of ascii characters), and spoken language features in general. I also counted and analyzed errors separately, in order to determine whether they increase over time. Errors include obvious typos, misspellings, lack of obligatory capitalization (e.g., at the beginning of a proper noun), missing words, extra words, etc.

The results of a 1-way ANOVA for each of these categories are presented in Table 3. Averages refer to tokens per sentence:

	Formality		Informality		Errors
	Avg.	SD	Avg.	SD	Avg.	SD
1975	.59	.54	1.01	.78	.25	.35
1977	.49	.44	.92	.80	.24	.57
1979	.43	.38	.87	.69	.22	.42
1982	.71	.64	1.20	.78	.09	.19
1984	.30	.27	.85	.65	.10	.20
1986	.65	.62	1.55	.94	.15	.24
msgs.	N=285^9		N=285		N=285
	p <.001		p <.0001		p =.088

Table 3. Formality and informality features (1-way ANOVA)

As Table 3 shows, formality decreases and informality increases over time; these results are significant. Moreover, the increase in informality is not simply caused by an increase in errors as people become more "casual" in their e-mail style; on the contrary, errors tend to decrease over time, although this result does not achieve statistical significance. These results are shown graphically in Figures 3-5.

Figure 3. Change in formality over time

Figure 4. Change in informality over time

Figure 5. Change in frequency of errors over time

Let us now consider some examples of the phenomena discussed thus far. Example (1), from a message posted to MsgGroup in 1986, is a highly formal passage.

(1) The normal method for distribution of RFCs is for interested parties to copy the documents from the NIC online library using FTP. Requests for special distribution should be addressed to either the author of the RFC in question or to NIC@SRI-NIC.ARPA. Unless specifically noted otherwise on the RFC itself, all RFCs are for unlimited distribution. [1986.10]

The formality in this passage is due to heavy use of features of syntactic complexity such as nominalizations, complement clauses, subordinate clauses, and passives (indicated in boldface).

In contrast, example (2), also posted in 1986, contains many features of informality.

(2) You know, Doug's right. You know, Jon's also right.

What makes sense, I guess, is to have organisations that are clearly US-specific, like the US GOV or MIL, be required to be .GOV.US or .MIL.US (maybe .ARMY.MIL.US, but I don't think so). Then we can have HQ.AMC.MIL.US, and HQ.AMC.MIL.NL, or whatever. Those other groups that wish to identify themselves as US-specific could follow suit, as TIME.COM.US vs TIME.COM.CA(nada, not lifornia). Those that wished to be thought of as multi-national or supra-national could drop the country loyalty entirely, like IBM.COM or ATT.COM. Oh, you say .ATT is a top-level domain now? ;-) [1986.19]

The informal features bolded in this example include the spoken discourse markers 'you know' and 'oh', personal 1st and 2nd person pronouns, contractions, hedges, informal lexis (e.g., 'drop the loyalty'), word play (with 'CA'), and a winking "smiley" face.

Example (3) illustrates a variety of errors (indicated in boldface).

(3) PTTs are not a ruling class. They have poilitical AND economic realities. (...)

I saw a network-creation proposal that indicated that a country would do what germany was doing and I wrote and said "why do that... (...) why not just get connections to EARNET and avoid waiting for germany to devlope their three stage PTT included system?" A year later I heard that the country involved connected to EARNET...I believe the high-level, avoid obstacles and get things going thinking people won out over the high-level "what to do about CCITT and the European PTTs" lost. [1986.5]

This short passage contains typos, lack of "obligatory" capitalization at the beginning of sentences and proper names, and an extra word, 'lost', which seems to be the result of poor syntactic planning. That this example has an unusually high number of errors is indicated by the fact that another MsgGroup participant, in responding to the message, wrote: "[I agree with everything you said]...except for spelling & grammar".

Politeness

The results for politeness also show change over time. I counted observances and violations of politeness, focusing on the effects on the addressee's positive face (in the sense of Brown & Levinson 1987). Positive face refers to people's desire to be ratified and accepted; positive politeness behaviors include thanks, compliments, and expressions of support. Face-threatening acts (FTAs) refer to behaviors that threaten the addressee's positive face, including on-record criticisms, disagreements, and put-downs.

The ANOVA results for the two politeness categories are given in Table 4:

	Positive Politeness		FTAs		FTAs (conflated)
	Avg.	SD	Avg.	SD	Avg.	SD
1975	.126	.248	.045	.164
1977	.046	.108	.013	.073	.027	.125
1979	.063	.213	.030	.095
1982	.020	.063	.055	.116	.042	.106
1984	.071	.214	.070	.192
1986	.031	.080	.089	.192	.079	.191
msgs.	N=285		N=285		N=285
	p <.05		p =.131		p <.05

Table 4. Politeness and face-threatening acts (1-way ANOVA)

Table 4 shows that positive politeness decreases significantly over time, while the incidence of on-record FTAs increases,¹⁰ as predicted by Baron (1984). These overall results are represented graphically in Figures 6 and 7.

Figure 6. Change in positive politeness over time

Figure 7. Change in face-threatening acts over time

Example (4), below, includes several explicit manifestations of positive politeness: appreciation, compliments, encouragement, and thanks.

(4) I really like the "TO: First Addressee" feature for messages which I originate. Very nice!

I think user prompting features in Bananard are still better then NMSG. Giving the user a prompt for what to do next after tying a command should be easy. For example: Forward (message seq):

We're getting there! Thanks for fixing the help files. [1975.16]

Example (5) contains face-threatening acts which violate positive politeness: strong expressions of displeasure directed at the addressee, criticism, and commands. These are expressed linguistically via the use of rhetorical questions, repeated punctuation, and a negative imperative ('do not put').

(5) TOM ELLIS:

Why do we put up systems with out documentation??????

NMSG - ? # yeilds "Documentation for # Not Available"

Why is NMSG available without this????

I really hate to get mad about this but will we never learn?

Please do somethingto get some documentation for NMSG online or pull it off the system.

In the future do not put any system "up" without documentation. [1975.11]

Both of the examples above were posted in 1975, and both, in fact, were posted by the same person. This shows that synchronic variation was present, even as politeness decreased over time.

Variability

Does e-mail style become more homogeneous over time? An examination of the standard deviation (SD) values for Tables 1-4 reveals no clear pattern of increase or decrease in degree of variation over time for any feature or cluster of features in the MsgGroup data. There is considerable variability during all periods, especially in regard to formality and informality. This is true even in messages posted by the same individual.

Example (6) was posted by the same man who posted example (1). Whereas example (1) is highly formal and syntactically complex, example (6) is stylistically informal (relative to the corpus as a whole) and syntactically simple:

(6) Hello:

My name is Jon Postel and i am mostly a user and infrequently an implementer of message systems. My background with the ARPANET and with computer message system goes back a ways. The first time i played with a computer message system was in 1967. (...) Since that time i have been involved in protocol development for the ARPANET and more recently for other networks too. I have also been involved in development of operating systems and interactive applications programs. I am quite interested in the design of user command languages. [1977.31; cf. example (1)]

The writer of this message, Jon Postel, has an official role in the group, which is as archivist of technical reports, but he is also a user of the technology and thus a "regular" participant as well. Via style shifting, he clearly demarcates and separates these two social roles.

Further evidence that variation in formality serves social functions comes from the group's moderator, Einar Stefferud, who signs his messages 'Stef'. Example (7) is an excerpt from a message posted by Stef in which he alternates between an informal and a formal style (informal features are bolded; formal features are italicized):

(7) at least, lets stop beating on Jon and SRI-NIC. Instead, lets look at what we have done to ourselves, and discuss what we might separately and jointly do about it.

I am not advocating reregistration of every domain and host, but I do advocate careful rethinking of the choices that have been made so far, and careful thought about new choices to be made.

Methinks that mostly, some education is needed for some administrators, including ourselves. We don't need blame. We need understanding. [1986.22]

In this message, Stef shifts into formal style (formal lexis; nominalizations; passives) to give his professional opinion on the issue under discussion, but switches to an informal style in the first and third paragraphs to chastise other group members for their behavior. As moderator, he has the authority to regulate others' behavior directly, but instead he opts to use an informal, solidarity-enhancing style to soften the impact of his face-threatening acts. This use of stylistic variation is also clearly socially motivated.

Discussion

The results for each prediction are summarized in Table 5.

Predictions	Supported?	Comments
decreased syntactic complexity	NO	No trends. Syntax is complex overall.
increased syntactic simplicity	NO	No trends.
decreased formality	YES
increased informality	YES
increased errors	NO	Trend is in opposite direction.
decreased politeness	YES	Tone is civil overall.
increased face threats	YES
decreased variability	NO	No trend. Data are highly variable, esp. formality/informality features.

Table 5. Summary of findings

We can see from Table 5 that the predictions for syntactic complexity, errors and variability are not supported. However, the MsgGroup data support the predictions for formality and politeness.

How can these results be explained? Let us first consider the negative results. It should be welcome news to parents and educators that sending e-mail regularly does not necessarily lead to a decline in correctness or syntactic sophistication. Why, then, do Baron and Hale claim that computer-mediated language is fragmented, non-standard and simple? Baron wrote in 1984, before computer-mediated communication as such was widely known or used, so perhaps she can be forgiven for making a prediction that turns out to be incorrect. However, Hale wrote in 1996, presumably on the basis of extensive experience with e-mail.

I propose two possible reasons for why my results differ. First, 11 years may be an insufficiently long period of time for syntactic change to take place, even in a high-contact environment such as the Internet. Nonetheless, most people have been using the Internet for fewer than 11 years; any stylistic changes claimed to have been observed would have had to take place in a shorter period of time. A more likely explanation is that MsgGroup has different users and different purposes for use from the groups observed by Hale and commented on in the popular media. The MsgGroup participants are highly educated, and they are engaged in professional communication. If Hale was basing her observations on casual, social interactions among younger users, as seems likely, differences in user demographics and communicative purpose could explain differences in the rate of "errors" and syntactic complexity.

It is more challenging to explain why there is no decrease in variability, since many aspects of the Internet situation predict that stylistic convergence should take place. What is clear is that if convergence is taking place, it co-exists with a high degree of stylistic variability. Indeed, some evidence suggests that variance may actually increase over time in experienced users. Stef, the moderator of MsgGroup, is the most consistent contributor to the forum,¹¹ and his use of all the stylistic features in Table 5 increases over time, even those (such as formality and politeness) for which the group as a whole shows a decrease. As was illustrated in example (7), Stef style-shifts strategically to negotiate social alignments and manage addressees' face. It may be that stylistically marked features of any sort are resources for socially expressive code alternation, and that regular long-time users learn to exploit such features as a means to compensate for the relative lack of social cues in the computer medium. That is, there may be an impetus towards increased stylistic variability that competes with or partially cancels out an impetus towards increased stylistic convergence.

The remaining MsgGroup findings support the predictions made by Baron and Hale. However, while Baron and Hale draw on essentially technologically deterministic explanations to support their claims, the MsgGroup data suggest that social factors also play an important role. Formality is claimed to decrease because the nature of the technology makes it easy to exchange messages quickly with a minimum of fuss. Yet it is also the case that during the life span of MsgGroup, social changes were taking place in the way the Internet was viewed and used. From a government-sponsored defense project in the 1970s, the Internet of the 80s had spread to use by non-computer scientists, including university students, who were using it for social as well as professional purposes (Hafner & Lyon, 1996). In other words, the definition of the Internet as a communication medium was itself changing to incorporate more 'informal' uses, and informality on MsgGroup may have been a reflection of this larger cultural trend. This is in addition to a familiarity effect that would have taken place within MsgGroup itself, such that formality decreases as participants get to know one another and the group over time.

Politeness is claimed to decrease, according to Baron and social psychologists such as Kiesler et al. (1984) and Spears et al. (1990), because the computer medium has a depersonalizing or "deindividuating" effect that causes users to become less inhibited and less concerned with the human beings at the receiving end of their messages. However, this effect is selective—it affects males more than females (Herring, 1994), and it doesn't account for why MsgGroup is relatively civil compared to other groups on the Internet (cf. Kim & Raja, 1991) where "flaming" is the norm and users metaphorically don "asbestos suits" (as one discussion group participant put it) before posting messages. Clearly, computer-mediated communication in and of itself does not cause users to flame; social accountability, user demographics and local group norms appear to exercise a stronger influence on politeness than the "disinhibiting" effects of the medium.

Conclusion

The results of this study are inconsistent with any strong view of technological determinism. Rather, they suggest the importance of social and ecological influences—such as user demographics, communicative purpose, the degree of social accountability in the relationships among participants, and the culture of the Internet at large—on computer-mediated language.

The results also suggest that stylistic variability is socially adaptive in text-only computer-mediated communication, which is a "lean" channel with reduced social context cues, and that this effect may partially or wholly offset any trends towards convergence. To show this, more fine-grained measures of variation and convergence may be necessary, e.g., measures which take into account users' level of experience. It may prove to be the case that experienced users increase their linguistic variability to enhance their stylistic repertoires, as in the case of Stef in MsgGroup, while new users converge initially towards dominant Internet norms in order to fit in and gain acceptance in online groups (Weber, forthcoming). Such a finding would explain the lack of overall change in variance for the MsgGroup corpus.

In the meantime, it is clear that at least in one long-lived computer-mediated discussion group, e-mail style exhibits socially motivated variation and undergoes change over time. The computer medium may enhance or accelerate this change, consistent with the claims of technological determinism, but a sociolinguistic approach is also needed in order to explain why change is variable across users, groups, purposes and situations.

Notes

¹ "Flaming" refers to the practice of sending electronic messages that are hostile or insulting; see, e.g., Lea et al. (1992).

² This claim is only made by Baron (1984, p.130).

³ Jon Katz, quoted in Hale (1996, p.9).

⁴ Baron (1984, pp.130, 136-7).

⁵ Cyhist@sjuvm.stjohns.edu.

⁶ For a historical account of the development of the Arpanet and the Internet, see Hafner and Lyon (1996).

⁷ Syntactic complexity and syntactic simplicity measures were carried out on the first 25 messages in each sample, for a total of 150 messages.

⁸ Quantitative results in tables were calculated as averages of averages for purposes of the ANOVAs, and are sensitive to variation in message length. In order to show global trends in the data independent of message length, the quantitative results in the figures were calculated as averages of the aggregate total for each sample. Both methods of calculation indicate average number of tokens per sentence.

⁹ Fifteen messages which appeared to have been composed and edited off-line (e.g., newspaper articles, technical reports) were eliminated from the larger sample, resulting in a total corpus of 285 messages.

¹⁰ The number of tokens is too small for this trend to emerge as significant when the six samples are compared individually. When adjacent samples are combined, however (as shown in the last column of Table 4), the trend achieves statistical significance.

¹¹ Stef is the only participant who posted messages during all six sample periods.

References

Baron, Naomi. 1984. "Computer mediated communication as a force in language change." Visible Language XVIII 2, 118-141.

Brown, Penelope and Stephen Levinson. 1987. Politeness: Some Universals in Language Usage. Cambridge: Cambridge University Press.

Daft, R. L. and R. H. Lengel. 1984. "Information richness: A new approach to managerial behavior and organization design." In B. M. Staw and L. L. Cummings (eds.), Research in Organizational Behavior, volume 6. Greenwich, CT: JAI Press, 191-233.

Hafner, Katie and M. Lyon. 1996. Where Wizards Stay Up Late: The Origins of the Internet. NY: Simon & Schuster.

Hale, Constance. 1996. Wired Style: Principles of English Usage in the Digital Age. San Francisco: HardWired.

Herring, Susan C. 1994. "Politeness in computer culture: Why women thank and men flame." In M. Bucholtz, A.C. Liang, L. Sutton, and C. Hines (eds.), Cultural Performances: Proceedings of the Third Berkeley Women and Language Conference. Berkeley: Berkeley Women and Language Group, 278-294.

Kiesler, Sara, Jane Siegel and Timothy W. McGuire. 1984. "Social psychological aspects of computer-mediated communication." American Psychologist 39.1123-1134.

Kim, Min-Sun and Narayan S. Raja. 1991. "Verbal aggression and self-disclosure on computer bulletin boards." Paper presented at the International Communication Association, Chicago, IL. Washington, D.C.: ERIC Clearinghouse on Languages and Linguistics. ED334620.

Lea, Martin, Tim O’Shea, Pat Fung and Russell Spears. 1992. "‘Flaming' in computer-mediated communication: Observations, explanations, implications." In M. Lea (ed.), Contexts of Computer-Mediated Communication. New York: Harvester Wheatsheaf, 89-112.

Paolillo, John C. 1996. "Language choice on soc.culture.punjab." Electronic Journal of Communication/La revue électronique de communication 6(3). Special issue on Computer-Mediated Discourse Analysis, ed. by S. Herring. Available from comserve@cios.com

Spears, Russell, Martin Lea & Stephen Lee. 1990. "De-individuation and group polarization in computer-mediated communication." British Journal of Social Psychology 29, 121-134.

Weber, H.L. Forthcoming. "Missed cues: How disputes can socialize virtual newcomers." In S. Herring (ed.), Computer-Mediated Conversation.

Weinreich, Uriel. 1953. Languages in Contact: Findings and Problems. Publications of the Linguistic Circle of New York, no.1. The Hague: Mouton and Co.